3_Text_Generation_RNN.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "hcD2nPQvPOFM"
   },
   "source": [
    "# Text generation using a RNN with eager execution\n",
    "\n",
    "(Adapted from the tensorflow example https://www.tensorflow.org/tutorials/sequences/text_generation)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "BwpJ5IffzRG6"
   },
   "source": [
    "This tutorial demonstrates how to generate text using a character-based RNN. We will work with a dataset of Shakespeare's writing from Andrej Karpathy's [The Unreasonable Effectiveness of Recurrent Neural Networks](http://karpathy.github.io/2015/05/21/rnn-effectiveness/). Given a sequence of characters from this data (\"Shakespear\"), train a model to predict the next character in the sequence (\"e\"). Longer sequences of text can be generated by calling the model repeatedly.\n",
    "\n",
    "This tutorial includes runnable code implemented using [tf.keras](https://www.tensorflow.org/programmers_guide/keras) and [eager execution](https://www.tensorflow.org/programmers_guide/eager). The following is sample output when the model in this tutorial trained for 30 epochs, and started with the string \"Q\":\n",
    "\n",
    "<pre>\n",
    "QUEENE:\n",
    "I had thought thou hadst a Roman; for the oracle,\n",
    "Thus by All bids the man against the word,\n",
    "Which are so weak of care, by old care done;\n",
    "Your children were in your holy love,\n",
    "And the precipitation through the bleeding throne.\n",
    "\n",
    "BISHOP OF ELY:\n",
    "Marry, and will, my lord, to weep in such a one were prettiest;\n",
    "Yet now I was adopted heir\n",
    "Of the world's lamentable day,\n",
    "To watch the next way with his father with his face?\n",
    "\n",
    "ESCALUS:\n",
    "The cause why then we are all resolved more sons.\n",
    "\n",
    "VOLUMNIA:\n",
    "O, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, it is no sin it should be dead,\n",
    "And love and pale as any will to that word.\n",
    "\n",
    "QUEEN ELIZABETH:\n",
    "But how long have I heard the soul for this world,\n",
    "And show his hands of life be proved to stand.\n",
    "\n",
    "PETRUCHIO:\n",
    "I say he look'd on, if I must be content\n",
    "To stay him from the fatal of our country's bliss.\n",
    "His lordship pluck'd from this sentence then for prey,\n",
    "And then let us twain, being the moon,\n",
    "were she such a case as fills m\n",
    "</pre>\n",
    "\n",
    "While some of the sentences are grammatical, most do not make sense. The model has not learned the meaning of words, but consider:\n",
    "\n",
    "* The model is character-based. When training started, the model did not know how to spell an English word, or that words were even a unit of text.\n",
    "\n",
    "* The structure of the output resembles a play—blocks of text generally begin with a speaker name, in all capital letters similar to the dataset.\n",
    "\n",
    "* As demonstrated below, the model is trained on small batches of text (100 characters each), and is still able to generate a longer sequence of text with coherent structure."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Goals\n",
    "\n",
    "- Understand the evolution of samples during training, and the relationship between the number of training ephocs and the quality of the generated output. \n",
    "- Produce (or download) your own textual corpus to experiment with parameters, training, and results. In [his article](http://karpathy.github.io/2015/05/21/rnn-effectiveness/) cited above, Karpathy plays with sources including Shakespeare, Wikipedia articles, Algebraic Geometry (Latex), and Linux source code, and baby names.\n",
    "E"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "srXC6pLGLwS6"
   },
   "source": [
    "## Setup"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "WGyKZj3bzf9p"
   },
   "source": [
    "### Import TensorFlow and other libraries"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "yG_n40gFzf9s"
   },
   "outputs": [],
   "source": [
    "from __future__ import absolute_import, division, print_function\n",
    "\n",
    "import tensorflow as tf\n",
    "tf.enable_eager_execution()\n",
    "\n",
    "import numpy as np\n",
    "import os\n",
    "import time"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "EHDoRoc5PKWz"
   },
   "source": [
    "### Download the Shakespeare dataset\n",
    "\n",
    "Change the following line to run this code on your own data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "pD_55cOxLkAb"
   },
   "outputs": [],
   "source": [
    "path_to_file = tf.keras.utils.get_file('shakespeare.txt', 'https://storage.googleapis.com/download.tensorflow.org/data/shakespeare.txt')\n",
    "\n",
    "# path_to_file = \"try_with_your_own_file.txt\"\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "UHjdCjDuSvX_"
   },
   "source": [
    "### Read the data\n",
    "\n",
    "First, look in the text."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "aavnuByVymwK"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Length of text: 747567 characters\n"
     ]
    }
   ],
   "source": [
    "# Read, then decode for py2 compat.\n",
    "text = open(path_to_file, 'rb').read().decode(encoding='utf-8')\n",
    "# length of text is the number of characters in it\n",
    "print ('Length of text: {} characters'.format(len(text)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "Duhg9NrUymwO"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "Reading Reference\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "Narratology: Introduction to the theory of Narrative, 3rd ed. Mieke Bal. 2009.\n",
      "\n",
      "A narrative test is a text in which an agent or subject conveys to an addressee ('tells' the reader) a story in a particular medium, such as langu\n"
     ]
    }
   ],
   "source": [
    "# Take a look at the first 250 characters in text\n",
    "print(text[:250])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "IlCgQBRVymwR"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "110 unique characters\n"
     ]
    }
   ],
   "source": [
    "# The unique characters in the file\n",
    "vocab = sorted(set(text))\n",
    "print ('{} unique characters'.format(len(vocab)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "rNnrKn_lL-IJ"
   },
   "source": [
    "## Process the text"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "LFjSVAlWzf-N"
   },
   "source": [
    "### Vectorize the text\n",
    "\n",
    "Before training, we need to map strings to a numerical representation. Create two lookup tables: one mapping characters to numbers, and another for numbers to characters."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "IalZLbvOzf-F"
   },
   "outputs": [],
   "source": [
    "# Creating a mapping from unique characters to indices\n",
    "char2idx = {u:i for i, u in enumerate(vocab)}\n",
    "idx2char = np.array(vocab)\n",
    "\n",
    "text_as_int = np.array([char2idx[c] for c in text])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "tZfqhkYCymwX"
   },
   "source": [
    "Now we have an integer representation for each character. Notice that we mapped the character as indexes from 0 to `len(unique)`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "FYyNlCNXymwY"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{\n",
      "  '\\t':   0,\n",
      "  '\\n':   1,\n",
      "  '\\x0c':   2,\n",
      "  '\\r':   3,\n",
      "  ' ' :   4,\n",
      "  '!' :   5,\n",
      "  '\"' :   6,\n",
      "  '#' :   7,\n",
      "  '%' :   8,\n",
      "  '&' :   9,\n",
      "  \"'\" :  10,\n",
      "  '(' :  11,\n",
      "  ')' :  12,\n",
      "  '*' :  13,\n",
      "  '+' :  14,\n",
      "  ',' :  15,\n",
      "  '-' :  16,\n",
      "  '.' :  17,\n",
      "  '/' :  18,\n",
      "  '0' :  19,\n",
      "  ...\n",
      "}\n"
     ]
    }
   ],
   "source": [
    "print('{')\n",
    "for char,_ in zip(char2idx, range(20)):\n",
    "    print('  {:4s}: {:3d},'.format(repr(char), char2idx[char]))\n",
    "print('  ...\\n}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "l1VKcQHcymwb"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "'\\n\\nReading Ref' ---- characters mapped to int ---- > [ 1  1 52 69 65 68 73 78 71  4 52 69 70]\n"
     ]
    }
   ],
   "source": [
    "# Show how the first 13 characters from the text are mapped to integers\n",
    "print ('{} ---- characters mapped to int ---- > {}'.format(repr(text[:13]), text_as_int[:13]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "bbmsf23Bymwe"
   },
   "source": [
    "### The prediction task"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "wssHQ1oGymwe"
   },
   "source": [
    "Given a character, or a sequence of characters, what is the most probable next character? This is the task we're training the model to perform. The input to the model will be a sequence of characters, and we train the model to predict the output—the following character at each time step.\n",
    "\n",
    "Since RNNs maintain an internal state that depends on the previously seen elements, given all the characters computed until this moment, what is the next character?\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "hgsVvVxnymwf"
   },
   "source": [
    "### Create training examples and targets\n",
    "\n",
    "Next divide the text into example sequences. Each input sequence will contain `seq_length` characters from the text. \n",
    "\n",
    "For each input sequence, the corresponding targets contain the same length of text, except shifted one character to the right. \n",
    "\n",
    "So break the text into chunks of `seq_length+1`. For example, say `seq_length` is 4 and our text is \"Hello\". The input sequence would be \"Hell\", and the target sequence \"ello\".\n",
    "\n",
    "To do this first use the `tf.data.Dataset.from_tensor_slices` function to convert the text vector into a stream of character indices."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "0UHJDA39zf-O"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "\n",
      "\n",
      "R\n",
      "e\n",
      "a\n"
     ]
    }
   ],
   "source": [
    "# The maximum length sentence we want for a single input in characters\n",
    "seq_length = 100\n",
    "examples_per_epoch = len(text)//seq_length\n",
    "\n",
    "# Create training examples / targets\n",
    "char_dataset = tf.data.Dataset.from_tensor_slices(text_as_int)\n",
    "\n",
    "for i in char_dataset.take(5):\n",
    "  print(idx2char[i.numpy()])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "-ZSYAcQV8OGP"
   },
   "source": [
    "The `batch` method lets us easily convert these individual characters to sequences of the desired size. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "l4hkDU3i7ozi"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "'\\n\\nReading Reference\\n\\n\\n\\n\\nNarratology: Introduction to the theory of Narrative, 3rd ed. Mieke Bal. 2009'\n",
      "\".\\n\\nA narrative test is a text in which an agent or subject conveys to an addressee ('tells' the reade\"\n",
      "'r) a story in a particular medium, such as language, imager, sound, buildings, r a combination thereo'\n",
      "\"f. A story is the fontent of that text, and produces a particular manifestation, inflection, and 'col\"\n",
      "\"oring' of a fabula; the fabula is presented in a certain manner. A fabula is a series of logically an\"\n"
     ]
    }
   ],
   "source": [
    "sequences = char_dataset.batch(seq_length+1, drop_remainder=True)\n",
    "\n",
    "for item in sequences.take(5):\n",
    "  print(repr(''.join(idx2char[item.numpy()])))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "UbLcIPBj_mWZ"
   },
   "source": [
    "For each sequence, duplicate and shift it to form the input and target text by using the `map` method to apply a simple function to each batch:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "9NGu-FkO_kYU"
   },
   "outputs": [],
   "source": [
    "def split_input_target(chunk):\n",
    "    input_text = chunk[:-1]\n",
    "    target_text = chunk[1:]\n",
    "    return input_text, target_text\n",
    "\n",
    "dataset = sequences.map(split_input_target)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "hiCopyGZymwi"
   },
   "source": [
    "Print the first examples input and target values:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "GNbw-iR0ymwj"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Input data:  '\\n\\nReading Reference\\n\\n\\n\\n\\nNarratology: Introduction to the theory of Narrative, 3rd ed. Mieke Bal. 200'\n",
      "Target data: '\\nReading Reference\\n\\n\\n\\n\\nNarratology: Introduction to the theory of Narrative, 3rd ed. Mieke Bal. 2009'\n"
     ]
    }
   ],
   "source": [
    "for input_example, target_example in  dataset.take(1):\n",
    "  print ('Input data: ', repr(''.join(idx2char[input_example.numpy()])))\n",
    "  print ('Target data:', repr(''.join(idx2char[target_example.numpy()])))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "_33OHL3b84i0"
   },
   "source": [
    "Each index of these vectors are processed as one time step. For the input at time step 0, the model receives the index for \"F\" and trys to predict the index for \"i\" as the next character. At the next timestep, it does the same thing but the `RNN` considers the previous step context in addition to the current input character."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "0eBu9WZG84i0"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Step    0\n",
      "  input: 1 ('\\n')\n",
      "  expected output: 1 ('\\n')\n",
      "Step    1\n",
      "  input: 1 ('\\n')\n",
      "  expected output: 52 ('R')\n",
      "Step    2\n",
      "  input: 52 ('R')\n",
      "  expected output: 69 ('e')\n",
      "Step    3\n",
      "  input: 69 ('e')\n",
      "  expected output: 65 ('a')\n",
      "Step    4\n",
      "  input: 65 ('a')\n",
      "  expected output: 68 ('d')\n"
     ]
    }
   ],
   "source": [
    "for i, (input_idx, target_idx) in enumerate(zip(input_example[:5], target_example[:5])):\n",
    "    print(\"Step {:4d}\".format(i))\n",
    "    print(\"  input: {} ({:s})\".format(input_idx, repr(idx2char[input_idx])))\n",
    "    print(\"  expected output: {} ({:s})\".format(target_idx, repr(idx2char[target_idx])))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "MJdfPmdqzf-R"
   },
   "source": [
    "### Create training batches\n",
    "\n",
    "We used `tf.data` to split the text into manageable sequences. But before feeding this data into the model, we need to shuffle the data and pack it into batches."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "p2pGotuNzf-S"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<BatchDataset shapes: ((64, 100), (64, 100)), types: (tf.int64, tf.int64)>"
      ]
     },
     "execution_count": 58,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Batch size \n",
    "BATCH_SIZE = 64\n",
    "steps_per_epoch = examples_per_epoch//BATCH_SIZE\n",
    "\n",
    "# Buffer size to shuffle the dataset\n",
    "# (TF data is designed to work with possibly infinite sequences, \n",
    "# so it doesn't attempt to shuffle the entire sequence in memory. Instead, \n",
    "# it maintains a buffer in which it shuffles elements).\n",
    "BUFFER_SIZE = 10000\n",
    "\n",
    "dataset = dataset.shuffle(BUFFER_SIZE).batch(BATCH_SIZE, drop_remainder=True)\n",
    "\n",
    "dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "r6oUuElIMgVx"
   },
   "source": [
    "## Build The Model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "m8gPwEjRzf-Z"
   },
   "source": [
    "Use `tf.keras.Sequential` to define the model. For this simple example three layers are used to define our model:\n",
    "\n",
    "* `tf.keras.layers.Embedding`: The input layer. A trainable lookup table that will map the numbers of each character to a vector with `embedding_dim` dimensions;\n",
    "* `tf.keras.layers.GRU`: A type of RNN with size `units=rnn_units` (You can also use a LSTM layer here.)\n",
    "* `tf.keras.layers.Dense`: The output layer, with `vocab_size` outputs."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "zHT8cLh7EAsg"
   },
   "outputs": [],
   "source": [
    "# Length of the vocabulary in chars\n",
    "vocab_size = len(vocab)\n",
    "\n",
    "# The embedding dimension \n",
    "embedding_dim = 256\n",
    "\n",
    "# Number of RNN units\n",
    "rnn_units = 1024"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "NNVB-jmMEOzP"
   },
   "source": [
    "Next define a function to build the model.\n",
    "\n",
    "Use `CuDNNGRU` if running on GPU.  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "wjKZrC39ELy0"
   },
   "outputs": [],
   "source": [
    "if tf.test.is_gpu_available():\n",
    "  rnn = tf.keras.layers.CuDNNGRU\n",
    "else:\n",
    "  import functools\n",
    "  rnn = functools.partial(\n",
    "    tf.keras.layers.GRU, recurrent_activation='sigmoid')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "MtCrdfzEI2N0"
   },
   "outputs": [],
   "source": [
    "def build_model(vocab_size, embedding_dim, rnn_units, batch_size):\n",
    "  model = tf.keras.Sequential([\n",
    "    tf.keras.layers.Embedding(vocab_size, embedding_dim, \n",
    "                              batch_input_shape=[batch_size, None]),\n",
    "    rnn(rnn_units,\n",
    "        return_sequences=True, \n",
    "        recurrent_initializer='glorot_uniform',\n",
    "        stateful=True),\n",
    "    tf.keras.layers.Dense(vocab_size)\n",
    "  ])\n",
    "  return model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "wwsrpOik5zhv"
   },
   "outputs": [],
   "source": [
    "model = build_model(\n",
    "  vocab_size = len(vocab), \n",
    "  embedding_dim=embedding_dim, \n",
    "  rnn_units=rnn_units, \n",
    "  batch_size=BATCH_SIZE)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "RkA5upJIJ7W7"
   },
   "source": [
    "For each character the model looks up the embedding, runs the GRU one timestep with the embedding as input, and applies the dense layer to generate logits predicting the log-liklihood of the next character:\n",
    "\n",
    "![A drawing of the data passing through the model](https://tensorflow.org/tutorials/sequences/images/text_generation_training.png)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "-ubPo0_9Prjb"
   },
   "source": [
    "## Try the model\n",
    "\n",
    "Now run the model to see that it behaves as expected.\n",
    "\n",
    "First check the shape of the output:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "C-_70kKAPrPU"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(64, 100, 110) # (batch_size, sequence_length, vocab_size)\n"
     ]
    }
   ],
   "source": [
    "for input_example_batch, target_example_batch in dataset.take(1): \n",
    "  example_batch_predictions = model(input_example_batch)\n",
    "  print(example_batch_predictions.shape, \"# (batch_size, sequence_length, vocab_size)\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "Q6NzLBi4VM4o"
   },
   "source": [
    "In the above example the sequence length of the input is `100` but the model can be run on inputs of any length: "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "vPGmAAXmVLGC"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_2 (Embedding)      (64, None, 256)           28160     \n",
      "_________________________________________________________________\n",
      "cu_dnngru_2 (CuDNNGRU)       (64, None, 1024)          3938304   \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (64, None, 110)           112750    \n",
      "=================================================================\n",
      "Total params: 4,079,214\n",
      "Trainable params: 4,079,214\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "uwv0gEkURfx1"
   },
   "source": [
    "To get actual predictions from the model we need to sample from the output distribution, to get actual character indices. This distribution is defined by the logits over the character vocabulary. \n",
    "\n",
    "Note: It is important to _sample_ from this distribution as taking the _argmax_ of the distribution can easily get the model stuck in a loop.\n",
    "\n",
    "Try it for the first example in the batch:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "4V4MfFg0RQJg"
   },
   "outputs": [],
   "source": [
    "sampled_indices = tf.random.multinomial(example_batch_predictions[0], num_samples=1) \n",
    "#sample_indices = tf.random.categorical(example_batch_predictions[0], num_samples=1)\n",
    "sampled_indices = tf.squeeze(sampled_indices,axis=-1).numpy()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "QM1Vbxs_URw5"
   },
   "source": [
    "This gives us, at each timestep, a prediction of the next character index:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "YqFMUQc_UFgM"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([ 41,  77,  62,  51,  23,   3,  20,   8,  52,   8,  74,   8,  14,\n",
       "       107,  12,   9,   9,  64,  21,   7, 108,  33,  94,  25,  94,  50,\n",
       "        89,  21,  86, 105,  84,  80,  82,  81,  41,  73,  86,  68,  58,\n",
       "        17,  85,  64,  95,  54,  61,  43,  75,  28,  22,  80,   1,  12,\n",
       "        75,  79,  68,  46,  57, 104,  89,  77,  96,  81,   1,  69,  28,\n",
       "        55,  74, 103,  74,  18,  12,  38,  90,  94,   1,  78,  94,  92,\n",
       "        38,  60,  58,  60,  62,  65,  98,  60,  45,  52,  61,   3,   9,\n",
       "        22,  14,  44,  97,  28,  99,  55,   9,  73])"
      ]
     },
     "execution_count": 66,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sampled_indices"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "LfLtsP3mUhCG"
   },
   "source": [
    "Decode these to see the text predicted by this untrained model:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "xWcFwPwLSo05"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Input: \n",
      " 'pologist Claude Levi-Strauss characterizes two modes of thought, the mythical and the scientific. My'\n",
      "\n",
      "Next Char Predictions: \n",
      " 'Gm]Q4\\r1%R%j%+Ó)&&`2#Õ>\\x8d6\\x8dPy2vÑtprqGivdX.u`\\x8eT[Ik93p\\n)kodLWÐym\\x93q\\ne9UjÉj/)Dz\\x8d\\nn\\x8d}DZXZ]a\\x9dZKR[\\r&3+J\\x999\\x9fU&i'\n"
     ]
    }
   ],
   "source": [
    "print(\"Input: \\n\", repr(\"\".join(idx2char[input_example_batch[0]])))\n",
    "print()\n",
    "print(\"Next Char Predictions: \\n\", repr(\"\".join(idx2char[sampled_indices ])))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "LJL0Q0YPY6Ee"
   },
   "source": [
    "## Train the model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "YCbHQHiaa4Ic"
   },
   "source": [
    "At this point the problem can be treated as a standard classification problem. Given the previous RNN state, and the input this time step, predict the class of the next character."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "trpqTWyvk0nr"
   },
   "source": [
    "### Attach an optimizer, and a loss function"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "UAjbjY03eiQ4"
   },
   "source": [
    "The standard `tf.keras.losses.sparse_softmax_crossentropy` loss function works in this case because it is applied across the last dimension of the predictions. \n",
    "\n",
    "Because our model returns logits, we need to set the `from_logits` flag. \n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "4HrXTACTdzY-"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Prediction shape:  (64, 100, 110)  # (batch_size, sequence_length, vocab_size)\n",
      "scalar_loss:       4.701376\n"
     ]
    }
   ],
   "source": [
    "def loss(labels, logits):\n",
    "#   return tf.keras.losses.sparse_categorical_crossentropy(labels, logits, from_logits=True)\n",
    "  return tf.nn.sparse_softmax_cross_entropy_with_logits(labels=labels, logits=logits)\n",
    "\n",
    "example_batch_loss  = loss(target_example_batch, example_batch_predictions)\n",
    "print(\"Prediction shape: \", example_batch_predictions.shape, \" # (batch_size, sequence_length, vocab_size)\") \n",
    "print(\"scalar_loss:      \", example_batch_loss.numpy().mean())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "jeOXriLcymww"
   },
   "source": [
    "Configure the training procedure using the `tf.keras.Model.compile` method. We'll use `tf.train.AdamOptimizer` with default arguments and the loss function. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "DDl1_Een6rL0"
   },
   "outputs": [],
   "source": [
    "model.compile(\n",
    "    optimizer = tf.train.AdamOptimizer(),\n",
    "    loss = loss)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "ieSJdchZggUj"
   },
   "source": [
    "### Configure checkpoints"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "C6XBUUavgF56"
   },
   "source": [
    "Use a `tf.keras.callbacks.ModelCheckpoint` to ensure that checkpoints are saved during training:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "W6fWTriUZP-n"
   },
   "outputs": [],
   "source": [
    "# Directory where the checkpoints will be saved\n",
    "checkpoint_dir = './text100'\n",
    "# Name of the checkpoint files\n",
    "checkpoint_prefix = os.path.join(checkpoint_dir, \"ckpt_{epoch}\")\n",
    "\n",
    "checkpoint_callback=tf.keras.callbacks.ModelCheckpoint(\n",
    "    filepath=checkpoint_prefix,\n",
    "    save_weights_only=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "3Ky3F_BhgkTW"
   },
   "source": [
    "### Execute the training"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "IxdOA-rgyGvs"
   },
   "source": [
    "To keep training time reasonable, use 3 epochs to train the model. In Colab, set the runtime to GPU for faster training."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "7yGBE2zxMMHs"
   },
   "outputs": [],
   "source": [
    "EPOCHS=3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "UK-hmKjYVoll"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "116/116 [==============================] - 7s 63ms/step - loss: 2.8899\n",
      "Epoch 2/10\n",
      "116/116 [==============================] - 7s 57ms/step - loss: 2.1782\n",
      "Epoch 3/10\n",
      "116/116 [==============================] - 6s 55ms/step - loss: 1.8924\n",
      "Epoch 4/10\n",
      "116/116 [==============================] - 7s 56ms/step - loss: 1.6824\n",
      "Epoch 5/10\n",
      "116/116 [==============================] - 6s 55ms/step - loss: 1.5467\n",
      "Epoch 6/10\n",
      "116/116 [==============================] - 6s 55ms/step - loss: 1.4590\n",
      "Epoch 7/10\n",
      "116/116 [==============================] - 6s 55ms/step - loss: 1.3957\n",
      "Epoch 8/10\n",
      "116/116 [==============================] - 7s 57ms/step - loss: 1.3442\n",
      "Epoch 9/10\n",
      "116/116 [==============================] - 7s 56ms/step - loss: 1.3018\n",
      "Epoch 10/10\n",
      "116/116 [==============================] - 6s 54ms/step - loss: 1.2615\n"
     ]
    }
   ],
   "source": [
    "history = model.fit(dataset.repeat(), epochs=EPOCHS, steps_per_epoch=steps_per_epoch, callbacks=[checkpoint_callback])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "kKkD5M6eoSiN"
   },
   "source": [
    "## Generate text"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "JIPcXllKjkdr"
   },
   "source": [
    "### Restore the latest checkpoint"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "LyeYRiuVjodY"
   },
   "source": [
    "To keep this prediction step simple, use a batch size of 1.\n",
    "\n",
    "Because of the way the RNN state is passed from timestep to timestep, the model only accepts a fixed batch size once built. \n",
    "\n",
    "To run the model with a different `batch_size`, we need to rebuild the model and restore the weights from the checkpoint.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "zk2WJ2-XjkGz"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'./text100/ckpt_10'"
      ]
     },
     "execution_count": 73,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tf.train.latest_checkpoint(checkpoint_dir)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "LycQ-ot_jjyu"
   },
   "outputs": [],
   "source": [
    "model = build_model(vocab_size, embedding_dim, rnn_units, batch_size=1)\n",
    "\n",
    "model.load_weights(tf.train.latest_checkpoint(checkpoint_dir))\n",
    "\n",
    "model.build(tf.TensorShape([1, None]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "71xa6jnYVrAN"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_3 (Embedding)      (1, None, 256)            28160     \n",
      "_________________________________________________________________\n",
      "cu_dnngru_3 (CuDNNGRU)       (1, None, 1024)           3938304   \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (1, None, 110)            112750    \n",
      "=================================================================\n",
      "Total params: 4,079,214\n",
      "Trainable params: 4,079,214\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "DjGz1tDkzf-u"
   },
   "source": [
    "### The prediction loop\n",
    "\n",
    "The following code block generates the text:\n",
    "\n",
    "* It Starts by choosing a start string, initializing the RNN state and setting the number of characters to generate.\n",
    "\n",
    "* Get the prediction distribution of the next character using the start string and the RNN state.\n",
    "\n",
    "* Then, use a multinomial distribution to calculate the index of the predicted character. Use this predicted character as our next input to the model.\n",
    "\n",
    "* The RNN state returned by the model is fed back into the model so that it now has more context, instead than only one word. After predicting the next word, the modified RNN states are again fed back into the model, which is how it learns as it gets more context from the previously predicted words.\n",
    "\n",
    "\n",
    "![To generate text the model's output is fed back to the input](https://tensorflow.org/tutorials/sequences/images/text_generation_sampling.png)\n",
    "\n",
    "Looking at the generated text, you'll see the model knows when to capitalize, make paragraphs and imitates a Shakespeare-like writing vocabulary. With the small number of training epochs, it has not yet learned to form coherent sentences."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "WvuwZBX5Ogfd"
   },
   "outputs": [],
   "source": [
    "def generate_text(model, start_string):\n",
    "  # Evaluation step (generating text using the learned model)\n",
    "\n",
    "  # Number of characters to generate\n",
    "  num_generate = 1000\n",
    "\n",
    "  # Converting our start string to numbers (vectorizing) \n",
    "  input_eval = [char2idx[s] for s in start_string]\n",
    "  input_eval = tf.expand_dims(input_eval, 0)\n",
    "\n",
    "  # Empty string to store our results\n",
    "  text_generated = []\n",
    "\n",
    "  # Low temperatures results in more predictable text.\n",
    "  # Higher temperatures results in more surprising text.\n",
    "  # Experiment to find the best setting.\n",
    "  temperature = 1.0\n",
    "\n",
    "  # Here batch size == 1\n",
    "  model.reset_states()\n",
    "  for i in range(num_generate):\n",
    "      predictions = model(input_eval)\n",
    "      # remove the batch dimension\n",
    "      predictions = tf.squeeze(predictions, 0)\n",
    "\n",
    "      # using a multinomial distribution to predict the word returned by the model\n",
    "      predictions = predictions / temperature\n",
    "      predicted_id = tf.multinomial(predictions, num_samples=1)[-1,0].numpy()\n",
    "      \n",
    "      # We pass the predicted word as the next input to the model\n",
    "      # along with the previous hidden state\n",
    "      input_eval = tf.expand_dims([predicted_id], 0)\n",
    "      \n",
    "      text_generated.append(idx2char[predicted_id])\n",
    "\n",
    "  return (start_string + ''.join(text_generated))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "ktovv0RFhrkn"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "writing age of poputian out of his task in the discurby identify for meanmanist. \r\n",
      "\r\n",
      "Most torammers in a dialogue identifying the writers. p 64.\r\n",
      "\r\n",
      "Kear; I am working within a gunstitute act an extend works and confusianity; the term \"white against the magically unist as it is playing a view of the lose is in the analogical landscape recipe it we've beel-systematically. p 135.\r\n",
      "\r\n",
      "Ws model is space. p 138.\r\n",
      "\r\n",
      "mayzey until the knowledge, decisions are - then sometimes surface between cembariail beautiful basic voilctives initmed. Untried At He was does long and \"that is technology as jung speaks of Style class down soon a \"piece?'4.\n",
      "\n",
      "Acting. \n",
      "\n",
      "Thuman Popul and Gant, Derri\f",
      ", ose] means it to produce that rendering the to it the unrejected identician--shmists, fall people whole, observes.\"  The Enthore in Rode BrD5 Pronsever, those strictly advertising the piring of return to the body has my top macy and ruseual modern phase stores, or those concression, oppositioned place universe into others a \n"
     ]
    }
   ],
   "source": [
    "# print(generate_text(model, start_string=u\"ROMEO\"))\n",
    "print(generate_text(model, start_string=u\"writing\"))\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "AM2Uma_-yVIq"
   },
   "source": [
    "The easiest thing you can do to improve the results it to train it for longer (try `EPOCHS=30`).\n",
    "\n",
    "You can also experiment with a different start string, or try adding another RNN layer to improve the model's accuracy, or adjusting the temperature parameter to generate more or less random predictions."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "Y4QwTjAM6A2O"
   },
   "source": [
    "## Advanced: Customized Training\n",
    "\n",
    "The above training procedure is simple, but does not give you much control.\n",
    "\n",
    "So now that you've seen how to run the model manually let's unpack the training loop, and implement it ourselves. This gives a starting point if, for example, to implement _curriculum learning_ to help stabilize the model's open-loop output. \n",
    "\n",
    "We will use `tf.GradientTape` to track the gradiends. You can learn more about this approach by reading the [eager execution guide](https://www.tensorflow.org/guide/eager).\n",
    "\n",
    "The procedure works as follows:\n",
    "\n",
    "* First, initialize the RNN state. We do this by calling the `tf.keras.Model.reset_states` method.\n",
    "\n",
    "* Next, iterate over the dataset (batch by batch) and calculate the *predictions* associated with each.\n",
    "\n",
    "* Open a `tf.GradientTape`, and calculate the predictions and loss in that context.\n",
    "\n",
    "* Calculate the gradients of the loss with respect to the model variables using the `tf.GradientTape.grads` method.\n",
    "\n",
    "* Finally, take a step downwards by using the optimizer's `tf.train.Optimizer.apply_gradients` method.\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "_XAm7eCoKULT"
   },
   "outputs": [],
   "source": [
    "model = build_model(\n",
    "  vocab_size = len(vocab), \n",
    "  embedding_dim=embedding_dim, \n",
    "  rnn_units=rnn_units, \n",
    "  batch_size=BATCH_SIZE)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "qUKhnZtMVpoJ"
   },
   "outputs": [],
   "source": [
    "optimizer = tf.train.AdamOptimizer()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "d4tSNwymzf-q"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efed34d30>, <tf.Tensor: id=883691, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-1.6645500e-06, -2.2062575e-06,  1.7071508e-06, ...,\n",
      "         9.0128751e-06, -5.9112317e-06,  2.6318590e-05],\n",
      "       [-1.6155893e-06, -1.9279382e-06,  1.1504717e-06, ...,\n",
      "        -1.3546650e-05, -1.9089233e-05,  1.0909354e-06],\n",
      "       [ 3.5631720e-06, -2.2063753e-06,  6.2005811e-06, ...,\n",
      "        -3.1541535e-05, -8.6429291e-06, -2.2900442e-06],\n",
      "       ...,\n",
      "       [ 2.4355722e-06,  4.9641960e-07,  5.9189250e-07, ...,\n",
      "        -3.0895822e-06, -8.0293930e-06,  1.2242006e-05],\n",
      "       [-1.4867965e-06, -6.2188144e-07,  1.0489189e-06, ...,\n",
      "         1.3251835e-05,  3.2291862e-07,  4.7271287e-06],\n",
      "       [ 4.4425838e-07, -3.6718487e-07,  5.5877337e-07, ...,\n",
      "         2.2209824e-05,  3.8150213e-07,  3.6547201e-05]], dtype=float32)>, <tf.Tensor: id=883692, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-7.7868316e-07, -3.8747717e-06,  8.1323509e-07, ...,\n",
      "        -2.2856702e-04,  4.7539350e-05,  1.7175196e-05],\n",
      "       [-8.7646117e-07, -7.0983896e-07,  1.8373427e-06, ...,\n",
      "         3.0340336e-04, -3.7867554e-05, -3.5753525e-05],\n",
      "       [ 2.4600573e-07, -1.5626619e-06,  5.9918643e-06, ...,\n",
      "         2.5795709e-04, -2.6710604e-05, -6.2687039e-05],\n",
      "       ...,\n",
      "       [-3.4780951e-06, -1.3066121e-05, -5.8463661e-06, ...,\n",
      "        -6.5396132e-05,  9.0942558e-06, -1.5445512e-05],\n",
      "       [ 5.0909043e-06,  8.0257387e-06,  6.6272191e-06, ...,\n",
      "         3.8596927e-04, -5.5514211e-05,  1.7808172e-05],\n",
      "       [-4.1886947e-06, -1.1628587e-05, -1.1881572e-06, ...,\n",
      "        -2.0754265e-05, -4.5177594e-06,  5.0262774e-05]], dtype=float32)>, <tf.Tensor: id=883693, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 5.0201525e-06,  3.1509349e-05, -2.1489934e-06, ...,\n",
      "       -6.5547775e-04,  8.4409025e-04, -1.3341176e-04], dtype=float32)>, <tf.Tensor: id=883596, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-3.7502684e-06, -4.2731466e-05, -4.0010596e-06, ...,\n",
      "        -4.0855221e-06, -4.9188138e-06, -6.3613552e-06],\n",
      "       [-9.0684807e-06, -1.0587991e-04, -1.0107200e-05, ...,\n",
      "        -9.0997955e-06, -1.2533966e-05, -1.4903861e-05],\n",
      "       [ 1.4207642e-06, -4.9556817e-05,  4.5556629e-07, ...,\n",
      "         1.4407465e-07,  1.2956447e-06, -7.4176683e-07],\n",
      "       ...,\n",
      "       [ 2.8106020e-05,  1.4475400e-03,  2.5638634e-05, ...,\n",
      "         3.1472919e-05,  3.5407436e-05,  4.4589364e-05],\n",
      "       [-4.2901238e-06, -3.9466886e-05, -4.4651101e-06, ...,\n",
      "        -4.2129041e-06, -5.5101464e-06, -6.8789504e-06],\n",
      "       [-2.5477648e-06, -4.3839941e-04, -1.7238112e-06, ...,\n",
      "        -3.7799298e-06, -2.0943794e-06, -4.0275986e-06]], dtype=float32)>, <tf.Tensor: id=883588, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.17280659e-04,  1.70412939e-03,  1.12523645e-04, -7.78948306e-04,\n",
      "       -1.33545939e-02,  1.53385685e-04,  1.21548818e-03,  1.28433661e-04,\n",
      "        1.20305136e-04,  1.26101935e-04,  4.13953880e-04, -7.45263387e-05,\n",
      "       -1.10349315e-03,  1.39825876e-04,  1.34333313e-04,  1.65851286e-03,\n",
      "        1.34924645e-04,  2.66461633e-04,  2.09472928e-04, -4.79574606e-04,\n",
      "       -1.18641547e-05,  9.33588308e-06, -3.88096669e-04,  5.08051831e-04,\n",
      "        9.77049931e-05, -1.53027751e-04,  2.79478380e-04,  2.37330576e-04,\n",
      "       -4.97982292e-05, -8.63072491e-05,  2.97657767e-04,  1.36021728e-04,\n",
      "        1.34424423e-04,  1.35352922e-04,  4.85084805e-04, -3.78154182e-05,\n",
      "        5.08630765e-04,  3.71619069e-04,  3.70058959e-04,  6.18343707e-04,\n",
      "        3.56066594e-04,  1.83724333e-04,  6.30510214e-04,  9.05072375e-04,\n",
      "       -1.42282297e-04,  1.09650260e-04, -1.04719365e-04,  5.57463674e-04,\n",
      "        5.55186067e-04,  3.39363469e-04,  2.08059471e-04,  2.10354483e-05,\n",
      "       -8.11028993e-04,  4.18456533e-04,  9.58557066e-04,  3.52294446e-04,\n",
      "        1.18347241e-04, -3.75114585e-04, -4.27143270e-04,  4.98913520e-04,\n",
      "        1.36738599e-04,  3.65277883e-05, -2.74809245e-05,  1.61566029e-04,\n",
      "        1.45481652e-04, -2.10363185e-03, -1.00391963e-03, -1.93116278e-03,\n",
      "        2.15468870e-04, -2.82511418e-03,  3.94980935e-03,  5.92948287e-04,\n",
      "        3.79248778e-03, -5.30807301e-03,  3.63928964e-04, -9.60087345e-05,\n",
      "        1.07537163e-03,  3.88429500e-04,  7.43312703e-05, -5.34732314e-03,\n",
      "        1.98789639e-03,  6.47644571e-04, -2.25885818e-03,  4.74796491e-03,\n",
      "       -2.13698531e-03,  1.55137805e-03, -3.35519173e-04, -5.95076592e-04,\n",
      "        1.35078502e-03,  9.48576548e-04,  7.49138359e-04,  1.22806858e-04,\n",
      "        1.16848620e-04,  1.31800683e-04,  1.32672139e-04,  1.19297860e-04,\n",
      "        1.19194323e-04,  1.26462677e-04,  1.22854122e-04,  1.27212319e-04,\n",
      "        1.20189012e-04,  1.26103405e-04,  1.28541913e-04,  1.22332945e-04,\n",
      "        1.25179417e-04,  1.29833890e-04,  1.20965378e-04,  1.22580139e-04,\n",
      "        1.32479894e-04,  1.46728635e-04], dtype=float32)>]\n",
      "Epoch 1 Batch 0 Loss 2.3907\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdae77f0>, <tf.Tensor: id=884053, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-1.91062691e-06, -1.22211156e-06,  2.11000975e-06, ...,\n",
      "        -2.43145496e-05,  3.61284510e-05,  3.45731278e-05],\n",
      "       [ 3.92355673e-07,  2.75087416e-07,  7.28761904e-08, ...,\n",
      "        -5.85846828e-05,  4.59762987e-06,  3.75785930e-06],\n",
      "       [-1.96742377e-07, -1.15850035e-07, -2.26242776e-07, ...,\n",
      "         4.32442866e-05,  4.25897127e-07,  3.02283534e-05],\n",
      "       ...,\n",
      "       [ 2.45168508e-07,  2.75996939e-07, -6.66799664e-08, ...,\n",
      "         2.60852744e-06, -2.93524477e-06,  4.44227953e-06],\n",
      "       [ 7.34546575e-08,  5.71443657e-08, -1.15025074e-08, ...,\n",
      "        -1.74504876e-05,  1.25296438e-05,  4.97736664e-06],\n",
      "       [ 5.08149128e-07,  4.61059642e-07, -1.57868243e-07, ...,\n",
      "         4.78365255e-05, -3.98816264e-05,  1.35406981e-05]], dtype=float32)>, <tf.Tensor: id=884054, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-4.78963648e-06, -8.68231564e-06, -1.37516972e-05, ...,\n",
      "        -6.09232229e-04,  9.06790447e-05,  8.53701749e-06],\n",
      "       [-5.29965234e-07,  4.58050977e-07, -4.21267487e-07, ...,\n",
      "         2.93442572e-04, -2.85350270e-05, -3.79593548e-05],\n",
      "       [ 2.06640607e-07, -1.32144407e-07,  1.46499679e-06, ...,\n",
      "         5.86031128e-05, -6.06586036e-06,  2.75465754e-05],\n",
      "       ...,\n",
      "       [ 6.57235546e-07,  1.11096949e-06,  7.69923076e-07, ...,\n",
      "        -2.21541959e-05,  1.43219575e-06,  1.83798056e-06],\n",
      "       [ 4.37941299e-06,  3.73320609e-06,  5.52567508e-06, ...,\n",
      "         1.70334984e-04, -2.46319614e-05, -4.34034591e-05],\n",
      "       [ 4.53950406e-06,  1.10144865e-05, -2.11627303e-05, ...,\n",
      "        -1.19714648e-04,  1.80351108e-05,  4.37088493e-05]], dtype=float32)>, <tf.Tensor: id=884055, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 1.3458535e-05,  2.2148224e-06, -3.0887495e-06, ...,\n",
      "       -2.3832892e-05,  1.7271472e-04, -1.2526632e-04], dtype=float32)>, <tf.Tensor: id=883958, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.4128302e-06, -4.6887697e-05, -1.4631971e-06, ...,\n",
      "        -1.7124075e-06, -2.1840783e-06, -2.9503967e-06],\n",
      "       [-2.8261657e-06, -1.7646240e-04, -3.0758970e-06, ...,\n",
      "        -3.2635348e-06, -4.6538416e-06, -6.1823757e-06],\n",
      "       [-8.9043499e-07, -9.8729455e-05, -1.1030342e-06, ...,\n",
      "        -1.2538994e-06, -1.1862405e-06, -2.1714318e-06],\n",
      "       ...,\n",
      "       [ 1.3361987e-05,  6.9077802e-04,  1.2229333e-05, ...,\n",
      "         1.5978017e-05,  1.8587170e-05,  2.5085475e-05],\n",
      "       [-1.3618694e-06, -3.3363205e-05, -1.4009279e-06, ...,\n",
      "        -1.6312663e-06, -2.0802884e-06, -2.8057675e-06],\n",
      "       [-1.3061667e-06, -2.5977267e-04, -1.0271104e-06, ...,\n",
      "        -1.5922270e-06, -1.3240872e-06, -2.2196368e-06]], dtype=float32)>, <tf.Tensor: id=883950, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.40241027e-05,  7.29285879e-04,  1.27315843e-05, -9.56273288e-06,\n",
      "        1.43653562e-03, -1.21952202e-04, -5.59910142e-04,  1.77136244e-05,\n",
      "        1.72482341e-05,  1.67902817e-05,  1.72984216e-03, -1.06023613e-03,\n",
      "       -7.01724086e-04,  2.34385989e-05,  1.83057309e-05,  1.71744800e-03,\n",
      "        9.79830162e-04, -1.73471635e-04, -9.67489104e-05, -5.19085617e-04,\n",
      "        3.25026514e-04,  4.05033934e-05,  3.11883632e-04, -3.75283089e-05,\n",
      "       -4.41072538e-04, -1.12400128e-04, -1.31462948e-04,  1.75207042e-05,\n",
      "        8.19611829e-04,  4.42945631e-04,  2.30484831e-04, -2.91611184e-04,\n",
      "        2.09558984e-05, -2.90153868e-04, -3.74270283e-04,  6.47279550e-04,\n",
      "        7.74976972e-04,  5.64542774e-04,  2.69494078e-04,  6.79499586e-04,\n",
      "       -5.40418667e-04, -2.16168788e-04,  1.69748906e-04,  5.45922667e-04,\n",
      "       -3.06963047e-05,  2.23492389e-04,  4.29330568e-04, -6.36704382e-04,\n",
      "        4.66638419e-04,  4.00922203e-04,  4.12117632e-04,  4.30677428e-05,\n",
      "        2.34911902e-04,  2.72214995e-04,  1.29655667e-03,  1.57695133e-04,\n",
      "        1.09482709e-04,  7.15709699e-04,  5.83004075e-05,  1.15041235e-04,\n",
      "        2.17932120e-05,  1.82259741e-04,  1.31558496e-04,  3.54676995e-05,\n",
      "        2.52714726e-05, -9.73356795e-03, -3.04652750e-03, -8.20935331e-03,\n",
      "       -1.37751177e-03,  9.82791185e-03,  3.55633325e-03,  2.20918620e-04,\n",
      "       -3.09898937e-03, -5.50045725e-03, -7.24886428e-04,  1.12186919e-03,\n",
      "        1.62596861e-03, -4.14906448e-04,  1.87649147e-03, -1.38125243e-03,\n",
      "       -1.81861408e-03, -1.59275980e-04,  1.94529537e-03, -2.49732239e-03,\n",
      "        2.45497562e-03,  1.45663833e-03, -3.68532434e-04,  2.55269511e-03,\n",
      "        2.22849660e-04, -1.62938057e-04, -2.55012012e-04,  1.68114093e-05,\n",
      "        1.39738122e-05,  1.73635290e-05,  2.14704978e-05,  1.46898083e-05,\n",
      "        1.57377472e-05,  1.74584020e-05,  1.66653135e-05,  1.66939481e-05,\n",
      "        1.63764162e-05,  1.77766124e-05,  1.65984020e-05,  1.52461744e-05,\n",
      "        1.58199673e-05,  1.62424185e-05,  1.57888298e-05,  1.67498947e-05,\n",
      "        1.93323594e-05,  2.61500172e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efe315cf8>, <tf.Tensor: id=884414, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-9.92165042e-07, -7.90013758e-07,  1.07315103e-07, ...,\n",
      "         8.09527046e-06,  1.77396501e-07,  3.71212227e-05],\n",
      "       [ 2.33796030e-07,  1.70020243e-07,  9.12660454e-08, ...,\n",
      "        -3.62851279e-05, -1.17323898e-05, -1.50454025e-05],\n",
      "       [-5.59802515e-07,  1.21261365e-07,  1.20367986e-07, ...,\n",
      "         5.04349264e-05,  9.38158064e-06,  3.99650598e-05],\n",
      "       ...,\n",
      "       [-8.71124612e-08, -1.68711651e-07,  3.71117835e-07, ...,\n",
      "        -2.30435603e-06,  1.44866965e-06,  3.53428345e-06],\n",
      "       [ 6.40544116e-08,  1.73362515e-08,  1.96123784e-09, ...,\n",
      "        -6.83516846e-05,  3.01203163e-05, -1.59173414e-05],\n",
      "       [ 8.16338684e-07,  2.29243327e-07, -6.45927898e-08, ...,\n",
      "         3.64767293e-05, -3.75367745e-05,  3.97859403e-05]], dtype=float32)>, <tf.Tensor: id=884415, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-9.3034618e-07, -1.1113289e-06,  1.3634956e-06, ...,\n",
      "        -3.6226269e-05,  7.4098302e-06,  6.4003348e-06],\n",
      "       [ 1.0346317e-06,  2.2273405e-06,  8.6486614e-07, ...,\n",
      "        -4.7171801e-05,  2.0216039e-05, -2.3378558e-05],\n",
      "       [ 4.6397361e-07,  1.2929881e-06,  3.1473050e-07, ...,\n",
      "         3.0861887e-05,  1.1222490e-05, -5.1114519e-05],\n",
      "       ...,\n",
      "       [ 7.1611282e-07,  1.4687282e-06, -1.9140217e-08, ...,\n",
      "        -3.5170018e-05,  3.3218180e-06,  6.6438372e-07],\n",
      "       [ 1.6554368e-06,  3.7960655e-07,  5.4607158e-06, ...,\n",
      "        -1.2474276e-05,  7.9380734e-06, -6.5660920e-06],\n",
      "       [ 6.0278571e-06,  1.4370229e-05, -4.5692177e-06, ...,\n",
      "         7.5107135e-05, -1.0789986e-05, -5.0371196e-05]], dtype=float32)>, <tf.Tensor: id=884416, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 5.9069416e-06, -3.0038614e-06, -3.8197973e-06, ...,\n",
      "       -3.6935529e-05, -7.5300013e-06,  7.4521580e-05], dtype=float32)>, <tf.Tensor: id=884319, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.1884292e-06,  1.2734687e-04, -1.2859648e-06, ...,\n",
      "        -1.4608827e-06, -1.8805089e-06, -2.5753275e-06],\n",
      "       [-2.4558371e-06,  2.1274984e-04, -2.7473170e-06, ...,\n",
      "        -2.8902523e-06, -4.0754367e-06, -5.4992702e-06],\n",
      "       [-7.4041179e-07, -4.9391776e-05, -9.6666690e-07, ...,\n",
      "        -1.0320754e-06, -1.0406877e-06, -1.8200169e-06],\n",
      "       ...,\n",
      "       [ 1.1010035e-05, -1.2294231e-03,  1.0464240e-05, ...,\n",
      "         1.3306466e-05,  1.5654086e-05,  2.1359074e-05],\n",
      "       [-1.1586851e-06,  1.0481143e-04, -1.2403029e-06, ...,\n",
      "        -1.4050971e-06, -1.8027865e-06, -2.4820124e-06],\n",
      "       [-9.9460794e-07,  2.2198362e-04, -7.9051426e-07, ...,\n",
      "        -1.2159771e-06, -9.9568786e-07, -1.7270879e-06]], dtype=float32)>, <tf.Tensor: id=884311, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.15424600e-05, -1.22034526e-03,  1.08642380e-05, -5.01875416e-04,\n",
      "        5.18799387e-03,  2.94652618e-05, -2.95901496e-04,  1.50269798e-05,\n",
      "        1.44876813e-05,  1.42684421e-05,  8.31767393e-04, -1.44273072e-04,\n",
      "       -8.82647437e-05,  2.02687625e-05,  1.55763591e-05, -3.94627394e-04,\n",
      "        1.42067031e-04,  1.14967639e-03,  5.21936381e-05, -4.40070035e-05,\n",
      "        5.20626840e-04,  3.08819057e-04, -4.31157496e-05,  3.86156316e-04,\n",
      "        1.36534058e-04,  2.95773556e-04,  2.82175024e-04, -1.76825532e-04,\n",
      "        3.05004185e-04, -6.80476951e-04, -4.10858600e-04,  1.77305792e-05,\n",
      "        1.75596560e-05,  1.88031518e-05,  2.28940713e-04,  2.72282225e-04,\n",
      "       -1.35473369e-04, -1.21994963e-04,  5.44164832e-05,  1.50269509e-04,\n",
      "        3.46484478e-04,  2.22079310e-04,  2.59956490e-04,  4.54985187e-04,\n",
      "       -5.22605696e-05,  2.02798503e-04, -2.43136776e-04,  4.36996808e-04,\n",
      "       -6.06293834e-05,  5.52721758e-05,  5.43781025e-05,  3.81992286e-05,\n",
      "       -8.60411092e-05,  4.96913097e-04,  3.21013067e-04, -1.62551169e-05,\n",
      "        9.87213280e-05,  2.74407997e-04,  5.21085385e-05,  2.28601784e-04,\n",
      "        1.85835561e-05,  1.08910963e-05, -3.40803817e-05,  3.03259967e-05,\n",
      "        2.15837554e-05, -5.14877308e-03, -2.70703831e-03, -3.30796838e-03,\n",
      "       -2.96186423e-04,  2.22379598e-03,  2.39985017e-03, -1.19815872e-03,\n",
      "       -1.03621511e-04, -2.93985568e-03,  6.56767748e-04,  1.80677351e-04,\n",
      "       -2.72006495e-04,  5.47565869e-04, -2.85782851e-03,  5.08660078e-03,\n",
      "        5.10727114e-04, -3.67396977e-04,  5.43133065e-05, -6.15695305e-03,\n",
      "        2.23490852e-03,  1.43431558e-03,  1.64571940e-03,  6.30948285e-04,\n",
      "       -8.71878001e-04, -7.05925806e-04, -3.08889488e-04,  1.42249446e-05,\n",
      "        1.19806309e-05,  1.43196039e-05,  1.76861922e-05,  1.23284572e-05,\n",
      "        1.31669731e-05,  1.39331069e-05,  1.36584595e-05,  1.38015303e-05,\n",
      "        1.35616638e-05,  1.46259790e-05,  1.38276237e-05,  1.27246094e-05,\n",
      "        1.30328244e-05,  1.36489216e-05,  1.32364012e-05,  1.39203639e-05,\n",
      "        1.62471970e-05,  2.22126073e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efe315710>, <tf.Tensor: id=884775, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 1.28326803e-06, -1.76927722e-06, -4.40268906e-08, ...,\n",
      "        -2.49608802e-05,  6.88207820e-06,  1.58102484e-05],\n",
      "       [ 3.18708366e-08,  3.28677139e-08,  1.43807640e-07, ...,\n",
      "        -1.82334516e-05,  1.98898651e-05, -7.88917168e-06],\n",
      "       [ 1.03624885e-07, -9.78221522e-08,  1.56200990e-08, ...,\n",
      "        -2.90622429e-05, -6.34066919e-06, -1.83521897e-05],\n",
      "       ...,\n",
      "       [ 3.34078862e-08, -7.30141849e-08, -4.57393590e-09, ...,\n",
      "         5.41042209e-06,  8.13849420e-06,  1.01514997e-05],\n",
      "       [ 8.46538413e-08,  5.58799318e-09,  2.51234606e-08, ...,\n",
      "        -1.46347293e-05, -2.49279547e-05,  1.55018188e-05],\n",
      "       [ 4.59177500e-07,  3.28594325e-07, -2.09785227e-07, ...,\n",
      "         4.48340870e-05, -2.94344864e-06,  1.87164587e-05]], dtype=float32)>, <tf.Tensor: id=884776, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-7.4535347e-07, -2.5194940e-06,  2.1092014e-06, ...,\n",
      "        -3.7242638e-04,  4.7485166e-05, -2.5197316e-05],\n",
      "       [-6.4553603e-07, -1.7555458e-06, -1.5552326e-06, ...,\n",
      "        -1.2959092e-04,  4.3151544e-05, -1.0490505e-05],\n",
      "       [-6.8188740e-07, -2.2779298e-06,  3.4569819e-06, ...,\n",
      "         4.9767719e-04, -6.9850983e-05,  5.9137652e-05],\n",
      "       ...,\n",
      "       [-5.2554947e-07, -9.1836273e-07,  3.3646512e-07, ...,\n",
      "         1.9314451e-05, -2.2836648e-06, -2.1991514e-07],\n",
      "       [ 5.3921491e-07, -2.8738184e-06,  2.4851388e-06, ...,\n",
      "        -1.8438759e-04,  2.6886679e-05, -6.1775663e-06],\n",
      "       [-5.5148530e-06, -1.2579427e-05, -1.0694483e-05, ...,\n",
      "         1.6581129e-04, -1.0536856e-05,  4.1646166e-05]], dtype=float32)>, <tf.Tensor: id=884777, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 2.1036465e-07,  4.6614168e-06,  4.9826463e-06, ...,\n",
      "        1.9360234e-05, -1.8922256e-04,  1.6571570e-04], dtype=float32)>, <tf.Tensor: id=884680, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.26264456e-06, -1.00628553e-04, -1.32797641e-06, ...,\n",
      "        -1.56449278e-06, -1.95680968e-06, -2.64339724e-06],\n",
      "       [-2.31551076e-06, -2.46709620e-04, -2.63697325e-06, ...,\n",
      "        -2.71832141e-06, -3.93809660e-06, -5.21003494e-06],\n",
      "       [-1.06398591e-06,  5.20193680e-06, -1.18618971e-06, ...,\n",
      "        -1.49217590e-06, -1.39561666e-06, -2.32821230e-06],\n",
      "       ...,\n",
      "       [ 1.34622233e-05,  1.56150223e-03,  1.20781660e-05, ...,\n",
      "         1.64366975e-05,  1.80487878e-05,  2.44161820e-05],\n",
      "       [-1.20085144e-06, -1.11979694e-04, -1.23755956e-06, ...,\n",
      "        -1.46753086e-06, -1.79015456e-06, -2.43992235e-06],\n",
      "       [-1.49829032e-06, -2.98988074e-04, -1.14477473e-06, ...,\n",
      "        -1.88510171e-06, -1.50630160e-06, -2.37725999e-06]], dtype=float32)>, <tf.Tensor: id=884672, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.41568962e-05,  1.63761852e-03,  1.26074810e-05, -1.31156354e-04,\n",
      "       -1.16302269e-02,  3.32349882e-05,  2.14822614e-03,  1.73824246e-05,\n",
      "        1.75142886e-05,  1.62764300e-05,  1.93194734e-04,  3.29420087e-04,\n",
      "       -5.84315130e-05,  2.34356048e-05,  1.84157070e-05, -1.12860790e-03,\n",
      "       -7.10258202e-04, -8.78034043e-04,  5.57863459e-05,  4.34727059e-04,\n",
      "       -4.79365001e-04, -1.07233680e-03, -4.40453354e-04, -2.52847076e-05,\n",
      "       -7.42553093e-04, -2.20499365e-04,  3.73681483e-04,  2.01387535e-04,\n",
      "        4.56178124e-04,  9.06533533e-05,  5.34529099e-05,  2.13505991e-05,\n",
      "        2.03986456e-05,  2.21201972e-05,  3.78232624e-04, -4.15195507e-04,\n",
      "        4.32162691e-04, -1.37374474e-04,  1.88755337e-04,  4.26886516e-04,\n",
      "        1.70420855e-04,  3.72986804e-04,  9.15106502e-05, -2.29117670e-03,\n",
      "        1.02162805e-04,  2.00448761e-04, -4.15493967e-04, -8.22279777e-04,\n",
      "       -5.42401453e-04, -9.41288017e-05,  5.12904720e-04,  4.19217904e-05,\n",
      "        2.38150140e-04,  2.97825231e-04,  2.81538902e-04, -3.25852889e-04,\n",
      "       -5.13541963e-05, -6.98154443e-04, -5.67226263e-04,  2.28674151e-04,\n",
      "        2.04190583e-05,  1.75627531e-04,  1.36865317e-04,  3.54162803e-05,\n",
      "        2.43786344e-05,  4.80215240e-05, -5.90395648e-04, -1.19537953e-03,\n",
      "       -4.78096452e-04,  5.79411746e-04,  3.41899809e-03, -1.62366708e-03,\n",
      "        2.85200682e-03,  1.84526644e-03, -4.48494859e-04, -1.01073470e-03,\n",
      "        8.12445534e-04,  2.23591644e-03, -5.25091309e-07,  1.73730589e-03,\n",
      "       -4.51924745e-03, -3.88997199e-04,  1.02808978e-03,  6.47366233e-03,\n",
      "        3.04416521e-03,  1.18156429e-03, -2.70586694e-03,  1.84978649e-03,\n",
      "        5.27009252e-06, -7.25056452e-07, -8.35535000e-04, -1.39555836e-04,\n",
      "       -1.42798352e-04,  1.74437373e-05,  2.20578822e-05,  1.45409940e-05,\n",
      "        1.58389757e-05,  1.84724577e-05,  1.66656628e-05,  1.66069622e-05,\n",
      "        1.62108608e-05,  1.81644209e-05,  1.64127905e-05,  1.52632219e-05,\n",
      "        1.61900643e-05,  1.63582263e-05,  1.55891212e-05,  1.72466134e-05,\n",
      "        1.88324284e-05,  2.55499544e-05], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafce80>, <tf.Tensor: id=885136, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-2.08622782e-06, -1.93767005e-06,  1.45472075e-06, ...,\n",
      "         2.16406279e-05, -2.43745740e-06, -1.09734192e-05],\n",
      "       [ 8.34993443e-07,  2.04917114e-07, -7.41905879e-08, ...,\n",
      "         6.37814310e-06,  1.38452160e-05,  1.49512971e-05],\n",
      "       [-2.69676065e-07,  1.44108867e-07,  1.37633066e-07, ...,\n",
      "        -4.49148647e-05, -3.05456433e-06, -2.84641392e-05],\n",
      "       ...,\n",
      "       [-1.95043057e-07,  1.04220426e-07,  1.13053865e-07, ...,\n",
      "        -6.27214877e-06,  2.67478094e-06, -1.41209910e-06],\n",
      "       [ 1.60297731e-07,  3.58165906e-08, -1.68633196e-08, ...,\n",
      "        -3.73492076e-05, -7.98025940e-06,  1.39143340e-05],\n",
      "       [ 7.82574489e-07,  3.95758235e-07, -1.97070321e-07, ...,\n",
      "        -2.19728167e-06, -1.05059135e-05,  4.58074010e-06]], dtype=float32)>, <tf.Tensor: id=885137, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-3.6008355e-06, -7.4439931e-06, -6.1898099e-06, ...,\n",
      "        -4.3672798e-04,  5.7223493e-05, -1.6736300e-05],\n",
      "       [ 3.6224023e-07,  2.6983005e-07, -6.0246975e-07, ...,\n",
      "         1.0137559e-04, -1.8396157e-05, -4.8562793e-05],\n",
      "       [-5.1599272e-07, -1.4891228e-06, -2.5582312e-07, ...,\n",
      "         2.8468066e-04, -2.9030261e-05,  8.8111474e-06],\n",
      "       ...,\n",
      "       [ 7.7838524e-07,  1.3802880e-06,  1.2346596e-06, ...,\n",
      "        -2.9445396e-06, -1.6320210e-07,  4.4740696e-07],\n",
      "       [ 4.9336754e-06,  7.4932868e-06,  7.8582298e-06, ...,\n",
      "        -9.9628924e-05,  9.5908154e-06, -8.5861284e-06],\n",
      "       [-5.9910062e-06, -1.4046961e-05, -5.0256021e-06, ...,\n",
      "        -7.0672009e-05, -4.2994166e-06,  8.4354258e-05]], dtype=float32)>, <tf.Tensor: id=885138, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 1.7085256e-05, -4.5645761e-06,  1.3716598e-06, ...,\n",
      "       -2.5489626e-06, -9.8414865e-05, -7.2399140e-05], dtype=float32)>, <tf.Tensor: id=885041, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.20709728e-06,  3.32877826e-05, -1.25418842e-06, ...,\n",
      "        -1.48332867e-06, -1.85755277e-06, -2.53124790e-06],\n",
      "       [-2.35877883e-06, -9.89550972e-06, -2.59188801e-06, ...,\n",
      "        -2.75432512e-06, -3.87414048e-06, -5.21560651e-06],\n",
      "       [-7.72970225e-07,  1.72186155e-05, -9.37532491e-07, ...,\n",
      "        -1.09409939e-06, -1.06250900e-06, -1.81667019e-06],\n",
      "       ...,\n",
      "       [ 1.16171168e-05, -6.24008942e-04,  1.05062445e-05, ...,\n",
      "         1.40176826e-05,  1.58807561e-05,  2.17739180e-05],\n",
      "       [-1.17969660e-06,  5.35662693e-05, -1.20295090e-06, ...,\n",
      "        -1.43176112e-06, -1.76117305e-06, -2.41450061e-06],\n",
      "       [-1.17051150e-06,  1.63368502e-04, -8.64999151e-07, ...,\n",
      "        -1.40916472e-06, -1.15122123e-06, -1.90288711e-06]], dtype=float32)>, <tf.Tensor: id=885033, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.20634431e-05, -6.73936273e-04,  1.08337945e-05,  6.57448836e-04,\n",
      "       -1.23496205e-02,  2.92357945e-05,  2.32985200e-04,  1.53137917e-05,\n",
      "        1.49182724e-05,  1.43392645e-05, -4.50810767e-04, -1.25891311e-04,\n",
      "       -2.10452781e-04,  2.09511345e-05,  1.62972283e-05,  1.28325669e-03,\n",
      "       -2.42760449e-04, -1.49974448e-03,  5.11508879e-05,  1.41308174e-05,\n",
      "       -2.70563702e-04, -9.97554744e-05, -7.56374560e-04,  2.73751735e-04,\n",
      "       -6.03171124e-04, -9.06506102e-05,  1.79437411e-04,  1.75381472e-04,\n",
      "        6.21233659e-04,  5.62637753e-04, -2.54285464e-04,  1.85539811e-05,\n",
      "        1.79745166e-05,  1.96252950e-05,  2.05078788e-04,  4.87765879e-04,\n",
      "       -3.99023207e-04, -6.13227894e-04, -1.44454607e-04,  5.71632001e-04,\n",
      "       -1.54895621e-04, -1.09008739e-04, -7.01669254e-04,  1.07018370e-03,\n",
      "        2.57579784e-04, -1.19050797e-04, -2.64247414e-04, -1.62790748e-04,\n",
      "        2.23770548e-04,  2.24747695e-04, -1.17826312e-04, -5.85514179e-04,\n",
      "        4.22558922e-04,  2.70575576e-04,  8.45642178e-04, -3.28717841e-04,\n",
      "        1.01183534e-04,  4.09729633e-04,  5.52394558e-05,  5.34570972e-05,\n",
      "       -1.37679162e-04, -1.38235628e-04, -1.82422460e-04,  3.15196412e-05,\n",
      "        2.14640750e-05, -1.28788664e-03,  9.66582957e-05,  1.84196804e-03,\n",
      "       -1.24147424e-04,  1.25260348e-03,  2.19814107e-03,  3.91000183e-04,\n",
      "       -2.25181621e-03,  1.90726749e-03,  3.32352414e-04,  1.40977558e-03,\n",
      "        1.76417665e-03,  2.86393100e-03, -3.41481459e-03,  2.15032767e-03,\n",
      "       -2.56477762e-03,  6.83287071e-05, -1.86995184e-03,  6.98475773e-03,\n",
      "        2.29201885e-03,  2.07939811e-04,  2.69811891e-04, -2.78329290e-03,\n",
      "        2.16677145e-04, -6.79650577e-04,  7.42702978e-04,  1.43093766e-05,\n",
      "        1.19328342e-05,  1.47144165e-05,  1.84943983e-05,  1.25478628e-05,\n",
      "        1.33729518e-05,  1.53023248e-05,  1.40277079e-05,  1.40376524e-05,\n",
      "        1.37037923e-05,  1.52910379e-05,  1.40653619e-05,  1.29657965e-05,\n",
      "        1.34897773e-05,  1.39877220e-05,  1.33515487e-05,  1.45287977e-05,\n",
      "        1.63803525e-05,  2.25260010e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdae76a0>, <tf.Tensor: id=885497, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 2.64402274e-06,  4.51350335e-09, -8.21651668e-07, ...,\n",
      "         7.33833440e-06, -5.01226566e-07,  2.38351840e-05],\n",
      "       [-6.00428862e-08, -1.37801251e-07,  2.80155490e-07, ...,\n",
      "        -2.15004093e-05, -2.73047190e-05, -8.63772766e-06],\n",
      "       [ 1.89715806e-07,  7.06686762e-08, -7.02237202e-08, ...,\n",
      "        -4.23410092e-05,  1.66702903e-05,  1.79525014e-05],\n",
      "       ...,\n",
      "       [-9.57606176e-07, -2.47971371e-07,  2.97520018e-07, ...,\n",
      "        -1.20695177e-05,  1.08543770e-06, -2.86567956e-06],\n",
      "       [ 8.16831829e-08,  3.62669823e-08, -6.29329620e-08, ...,\n",
      "        -2.12788800e-05, -8.15185288e-08, -4.60181909e-05],\n",
      "       [-9.84147874e-08,  1.96264054e-07,  6.12896187e-08, ...,\n",
      "         1.26189025e-05,  1.48477784e-05,  1.01197647e-05]], dtype=float32)>, <tf.Tensor: id=885498, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-8.5349564e-07, -1.7105139e-06,  1.7375302e-07, ...,\n",
      "        -5.6102966e-05,  1.3770629e-05, -2.4865703e-05],\n",
      "       [ 6.2636923e-07,  1.3397516e-06,  5.9896092e-07, ...,\n",
      "        -1.2402829e-04,  2.6493388e-05,  2.2722522e-06],\n",
      "       [-2.0160985e-07, -2.4559441e-07, -6.9540744e-07, ...,\n",
      "        -9.9361307e-05,  1.3312155e-05, -8.3473951e-06],\n",
      "       ...,\n",
      "       [ 4.2084284e-07,  3.7325978e-07,  3.9652729e-08, ...,\n",
      "        -2.1709457e-05,  2.2079894e-06,  2.5231525e-06],\n",
      "       [ 4.8587535e-06,  4.5039619e-06,  1.1904525e-05, ...,\n",
      "        -1.0557714e-04,  1.0748438e-05, -1.0088311e-05],\n",
      "       [-4.1380313e-06, -8.0577702e-06, -3.8408702e-06, ...,\n",
      "        -4.2902015e-05, -2.3043669e-05,  1.7010676e-05]], dtype=float32)>, <tf.Tensor: id=885499, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 5.8531336e-06, -3.7286854e-06,  8.6249963e-07, ...,\n",
      "       -2.2253051e-05, -1.0593206e-04, -4.1928506e-05], dtype=float32)>, <tf.Tensor: id=885402, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.1644618e-06,  4.0151896e-05, -1.2392566e-06, ...,\n",
      "        -1.4429850e-06, -1.8342881e-06,  1.9223791e-05],\n",
      "       [-2.2913439e-06,  1.4378602e-05, -2.5780816e-06, ...,\n",
      "        -2.7254109e-06, -3.8570925e-06,  5.2058585e-05],\n",
      "       [-8.9687035e-07, -5.9769416e-05, -1.0504750e-06, ...,\n",
      "        -1.2396636e-06, -1.2545780e-06, -4.0002742e-05],\n",
      "       ...,\n",
      "       [ 1.1905118e-05,  3.9312197e-04,  1.0987607e-05, ...,\n",
      "         1.4455788e-05,  1.6488199e-05, -1.3314857e-04],\n",
      "       [-1.1812853e-06, -4.7256208e-06, -1.2304241e-06, ...,\n",
      "        -1.4532899e-06, -1.7860900e-06,  2.5680447e-05],\n",
      "       [-1.1535365e-06, -3.5614014e-04, -8.8797105e-07, ...,\n",
      "        -1.4061923e-06, -1.1775894e-06, -3.4033430e-05]], dtype=float32)>, <tf.Tensor: id=885394, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.23765094e-05,  4.34851681e-04,  1.13451879e-05, -6.87901629e-04,\n",
      "       -1.01384241e-03,  3.07730043e-05,  2.96688086e-06, -1.40781151e-04,\n",
      "        1.53530036e-05, -1.41665121e-04, -2.02993024e-03,  2.07819394e-04,\n",
      "        1.87653368e-05,  2.17492179e-05, -1.39579555e-04,  1.98165607e-03,\n",
      "       -1.82812265e-03,  1.22593215e-03, -1.03914717e-04, -5.49751385e-05,\n",
      "       -8.44829832e-04, -2.30657577e-04,  6.22809748e-05, -3.30523879e-04,\n",
      "        2.09678299e-04, -4.67944337e-05, -1.13761074e-04,  3.82917206e-05,\n",
      "       -7.07712097e-05, -1.70699059e-04, -2.19853959e-04,  1.88482700e-05,\n",
      "        1.83535012e-05,  2.04583666e-05,  2.15855645e-04, -8.62004468e-04,\n",
      "        1.98041060e-04,  4.84925840e-04,  4.74295928e-04,  5.48180600e-04,\n",
      "       -1.56589696e-04, -1.10728710e-04,  6.94176488e-05,  8.68419651e-04,\n",
      "        1.00668760e-04, -9.01256979e-04,  5.58143220e-05,  2.41440612e-05,\n",
      "       -9.46769069e-05,  3.72313370e-04, -4.34635993e-04, -1.15413772e-04,\n",
      "        2.84484442e-04, -6.40968283e-05, -3.93912836e-04, -1.11540166e-05,\n",
      "       -5.47956442e-05, -8.81790242e-04,  5.76667553e-05, -1.12919210e-04,\n",
      "        1.89234888e-05,  1.78819755e-04,  1.43454527e-04,  3.24780558e-05,\n",
      "       -2.90208758e-04, -2.63685622e-04,  5.71722048e-05,  4.25672205e-03,\n",
      "       -2.49923533e-03, -7.62938196e-03, -1.27862196e-03, -8.30954639e-04,\n",
      "       -2.19334010e-03,  5.32270456e-03,  1.30672670e-05, -4.65020392e-04,\n",
      "        4.80956427e-04,  2.36063788e-04, -7.99869304e-04,  5.19307097e-03,\n",
      "        2.60376115e-03, -8.72581149e-05,  1.52808067e-03,  3.00277793e-03,\n",
      "       -2.09753262e-03,  2.94697075e-03,  7.26058090e-04, -3.02166492e-03,\n",
      "       -8.43425747e-04, -9.06607951e-04,  9.25664208e-04,  1.52479006e-05,\n",
      "        1.21268749e-05, -1.41124328e-04,  1.84675609e-05,  1.27743024e-05,\n",
      "       -1.42333287e-04,  1.57929608e-05,  1.42982517e-05,  1.41816417e-05,\n",
      "        1.39616368e-05,  1.56650876e-05,  1.45841650e-05,  1.30500321e-05,\n",
      "        1.37425886e-05,  1.43259231e-05,  1.36985682e-05,  1.50105698e-05,\n",
      "        1.70270960e-05, -1.33589798e-04], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdae62e8>, <tf.Tensor: id=885858, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 8.27213512e-07, -6.91254968e-07, -9.88635293e-07, ...,\n",
      "         1.58841522e-05, -4.65007324e-05,  1.96408564e-05],\n",
      "       [ 1.75337973e-07, -2.11044924e-08,  5.75864760e-08, ...,\n",
      "        -2.05098349e-05, -4.72991815e-06,  2.86967020e-06],\n",
      "       [-2.28299569e-07,  1.90219154e-07, -5.96954024e-08, ...,\n",
      "         4.80844756e-05, -1.21685025e-05, -6.31247667e-06],\n",
      "       ...,\n",
      "       [-4.34975334e-07, -6.18830569e-08,  2.37592431e-07, ...,\n",
      "        -3.27916200e-06, -5.36283596e-06, -6.04314528e-06],\n",
      "       [ 9.62794005e-08,  3.82574363e-08,  2.66725824e-08, ...,\n",
      "        -6.35094038e-05,  4.64472078e-06, -1.18533571e-05],\n",
      "       [-2.14205812e-07, -4.21773638e-09,  1.42177981e-07, ...,\n",
      "         1.27240983e-05, -2.90879470e-05, -6.39193831e-06]], dtype=float32)>, <tf.Tensor: id=885859, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-2.5006614e-08,  1.0145715e-07,  3.4897439e-06, ...,\n",
      "         2.3321502e-04, -5.3286971e-05, -2.7462735e-05],\n",
      "       [-6.9260369e-07, -1.7176493e-06, -8.2742065e-07, ...,\n",
      "         6.6122047e-05, -2.1944417e-05,  2.4844594e-05],\n",
      "       [-5.5737581e-07, -8.5782108e-07, -6.8985173e-07, ...,\n",
      "        -5.5588411e-05,  2.5940069e-05,  1.3904995e-05],\n",
      "       ...,\n",
      "       [-1.1814193e-06, -2.1505798e-06, -5.7242335e-07, ...,\n",
      "         3.2415261e-05, -3.3189567e-06,  1.5955162e-06],\n",
      "       [ 7.2016137e-06,  9.7707307e-06,  1.3039825e-05, ...,\n",
      "         2.5771168e-04, -5.3830121e-05, -3.1336236e-05],\n",
      "       [ 1.7888959e-06,  4.8936454e-06, -9.2519285e-06, ...,\n",
      "        -6.9198097e-05, -1.6292808e-05,  2.9594043e-05]], dtype=float32)>, <tf.Tensor: id=885860, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 1.0507891e-05,  7.3328351e-06,  4.2376864e-06, ...,\n",
      "        3.4850844e-05,  2.5341337e-04, -7.0372000e-05], dtype=float32)>, <tf.Tensor: id=885763, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[ 1.1649276e-05,  7.7049284e-05, -1.3352684e-06, ...,\n",
      "        -1.6027775e-06, -1.9708891e-06, -2.6868925e-06],\n",
      "       [ 1.7237786e-05, -2.8375191e-05, -2.7570652e-06, ...,\n",
      "        -2.9254247e-06, -4.1127150e-06, -5.4988209e-06],\n",
      "       [ 2.2932529e-05,  1.3208644e-04, -1.3125771e-06, ...,\n",
      "        -1.6267334e-06, -1.6158275e-06, -2.5507925e-06],\n",
      "       ...,\n",
      "       [-1.4083636e-04, -1.4257699e-03,  1.2232598e-05, ...,\n",
      "         1.6541155e-05,  1.8401164e-05,  2.4955341e-05],\n",
      "       [ 3.3690503e-06,  1.5574141e-04, -1.2964217e-06, ...,\n",
      "        -1.5430214e-06, -1.8856131e-06, -2.6065393e-06],\n",
      "       [ 3.5551024e-05,  1.4826778e-04, -1.0528455e-06, ...,\n",
      "        -1.7192467e-06, -1.4189325e-06, -2.2279460e-06]], dtype=float32)>, <tf.Tensor: id=885755, shape=(110,), dtype=float32, numpy=\n",
      "array([-1.42215053e-04, -1.58423593e-03,  1.26870873e-05,  7.48294871e-04,\n",
      "        4.72251233e-03, -1.21867539e-04, -5.25439391e-04,  1.73739827e-05,\n",
      "        1.74476518e-05,  1.62927299e-05, -7.59294344e-05,  7.28012412e-04,\n",
      "        5.60514629e-04,  2.44950043e-05,  1.92602438e-05, -5.72605641e-04,\n",
      "       -4.31478868e-04,  4.81225958e-04,  5.70574448e-05,  4.35098802e-04,\n",
      "       -4.42578021e-04,  2.92676123e-04,  9.53050039e-05, -1.61645090e-04,\n",
      "       -8.14948507e-05, -6.58764679e-04, -8.87213973e-05,  2.18696383e-04,\n",
      "       -6.43579988e-04,  7.98359164e-04,  2.89892836e-04,  2.14298943e-05,\n",
      "        2.06844470e-05,  2.31329650e-05, -8.95543853e-05, -1.44136691e-04,\n",
      "        6.21212239e-05,  3.62902618e-04,  1.26586910e-05, -2.20132250e-04,\n",
      "        1.88288002e-04,  2.18100016e-04, -4.29284264e-05,  8.38564301e-04,\n",
      "        1.15232418e-04,  2.14069951e-04, -3.90439032e-04, -6.21318395e-05,\n",
      "       -2.27583878e-04, -4.75973939e-05,  4.04997350e-04,  4.65620287e-05,\n",
      "       -4.48555686e-04,  1.24855374e-04,  1.04876875e-03, -3.06378468e-04,\n",
      "       -4.66254714e-05,  3.18183098e-04,  6.45241817e-05, -1.04974424e-04,\n",
      "        2.11226634e-05,  1.95620858e-04,  1.57580173e-04,  3.62677383e-05,\n",
      "        2.45601568e-05,  3.69310728e-03, -8.51812889e-04,  7.90510210e-04,\n",
      "       -1.82800123e-03, -7.06937769e-03, -1.38196640e-03, -1.54630642e-03,\n",
      "       -1.23778952e-03,  4.33805538e-03,  3.75251439e-05, -1.00618543e-03,\n",
      "        1.55132858e-03,  4.44827980e-04, -7.75300490e-04, -3.03826714e-03,\n",
      "       -3.27020942e-04, -7.30716929e-05, -1.25750713e-03,  4.50826390e-03,\n",
      "        9.28612542e-04,  1.25657977e-03,  1.52543839e-03, -1.90480135e-03,\n",
      "       -6.33216114e-04, -2.68907240e-03, -1.74545945e-04,  1.68199294e-05,\n",
      "        1.37795523e-05,  1.72189721e-05,  2.15972268e-05,  1.45694394e-05,\n",
      "        1.58499806e-05,  1.81618088e-05,  1.63446930e-05,  1.64199955e-05,\n",
      "        1.61759253e-05,  1.79195376e-05,  1.62461019e-05,  1.51799659e-05,\n",
      "        1.58716139e-05,  1.63411296e-05,  1.54149711e-05,  1.72277523e-05,\n",
      "        1.90773226e-05,  2.59438093e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafcfd0>, <tf.Tensor: id=886219, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 1.04960338e-06,  3.10538923e-07, -4.56783198e-07, ...,\n",
      "         2.56967382e-06,  3.79433041e-05,  3.57221070e-05],\n",
      "       [ 5.44306488e-07,  3.29090255e-07, -1.04665844e-07, ...,\n",
      "        -4.17628835e-05,  5.76712500e-06, -2.83524832e-06],\n",
      "       [-1.86063943e-07,  7.30147818e-08, -1.01033329e-07, ...,\n",
      "        -1.42865383e-05,  2.59438657e-05,  4.99869639e-05],\n",
      "       ...,\n",
      "       [ 1.96055694e-07, -5.43036798e-08, -7.53504139e-08, ...,\n",
      "        -1.56880753e-06, -1.17329901e-05,  3.87846825e-07],\n",
      "       [ 7.48047047e-09,  4.38645493e-08,  3.19863354e-08, ...,\n",
      "         8.50463039e-06,  4.17638184e-05, -4.54124311e-05],\n",
      "       [-3.89060830e-07, -2.20746273e-07,  2.57334278e-07, ...,\n",
      "        -5.78292713e-07, -6.54834948e-05,  1.81195301e-05]], dtype=float32)>, <tf.Tensor: id=886220, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 2.6165742e-07, -1.1963623e-06,  8.3516807e-06, ...,\n",
      "         5.8812683e-04, -9.2510643e-05,  6.0590028e-05],\n",
      "       [-3.8814267e-07, -5.5033485e-07,  8.4586156e-07, ...,\n",
      "        -1.4280368e-04,  2.0988657e-05,  2.0659458e-05],\n",
      "       [ 4.8774405e-07,  5.4702383e-07,  9.1299188e-07, ...,\n",
      "        -8.9923866e-05,  1.4827889e-05, -1.2570180e-06],\n",
      "       ...,\n",
      "       [ 7.9397466e-08, -7.0441345e-08, -4.3260161e-07, ...,\n",
      "        -1.4074936e-05,  2.3314576e-06, -1.3298912e-06],\n",
      "       [-5.3256490e-06, -1.8270954e-05, -1.3248239e-05, ...,\n",
      "         1.8907373e-04, -2.2454344e-05, -3.4116045e-05],\n",
      "       [ 6.8449676e-06,  1.5551484e-05, -1.7187316e-05, ...,\n",
      "         1.4498051e-04, -4.3468401e-05, -5.0619645e-05]], dtype=float32)>, <tf.Tensor: id=886221, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 6.1204514e-06,  1.3436313e-06, -6.1363958e-06, ...,\n",
      "       -1.4373522e-05,  1.9009844e-04,  1.4601403e-04], dtype=float32)>, <tf.Tensor: id=886124, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.19785250e-06, -2.35914958e-05, -1.28643592e-06, ...,\n",
      "        -1.49386869e-06, -1.88943670e-06, -2.56550015e-06],\n",
      "       [-2.45491810e-06, -7.73195716e-05, -2.78689822e-06, ...,\n",
      "        -2.94451524e-06, -4.14217629e-06, -5.53320751e-06],\n",
      "       [-9.00293799e-07, -9.35466524e-05, -1.07985250e-06, ...,\n",
      "        -1.26473810e-06, -1.27269345e-06, -2.11574411e-06],\n",
      "       ...,\n",
      "       [ 1.11524596e-05,  8.81284359e-05,  1.07753713e-05, ...,\n",
      "         1.36948493e-05,  1.61246517e-05,  2.16294247e-05],\n",
      "       [-1.18579419e-06, -1.48319705e-05, -1.27337159e-06, ...,\n",
      "        -1.46287141e-06, -1.85644285e-06, -2.52009499e-06],\n",
      "       [-7.93843697e-07, -2.18569912e-04, -6.31183241e-07, ...,\n",
      "        -1.00482748e-06, -7.67706410e-07, -1.29630132e-06]], dtype=float32)>, <tf.Tensor: id=886116, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.14927016e-05,  7.15832430e-05,  1.10560277e-05,  8.61734443e-04,\n",
      "        7.56963482e-03,  3.08354720e-05, -1.34384783e-03,  1.48845093e-05,\n",
      "        1.43889993e-05,  1.44677188e-05,  4.10300592e-04,  1.23568054e-04,\n",
      "       -5.37080225e-04,  2.17715442e-05,  1.61766020e-05,  8.08203011e-04,\n",
      "        1.26646497e-04,  1.98672013e-03,  5.23433118e-05, -3.46531684e-04,\n",
      "       -1.88617734e-04,  7.47226368e-05,  1.83069526e-04,  2.52684520e-04,\n",
      "        1.89921804e-04, -1.17651034e-04, -1.58504845e-05, -1.69505802e-06,\n",
      "       -8.99218503e-06,  1.58260271e-04,  1.53560744e-04,  1.77997772e-05,\n",
      "        1.75912937e-05,  1.97736135e-05, -9.79947654e-05, -2.41937145e-04,\n",
      "        5.18552144e-04, -2.71357334e-04, -3.05715395e-04,  6.77190430e-04,\n",
      "        5.00722323e-04, -5.56564599e-04, -5.36028703e-04,  4.57186106e-04,\n",
      "       -4.78607253e-05, -1.02061225e-04,  2.35031752e-04,  2.77523563e-04,\n",
      "       -7.14248337e-04,  2.09098187e-04, -4.85662385e-05,  4.35606235e-05,\n",
      "        2.79836240e-05, -3.52827774e-04,  9.26658860e-04, -1.46144390e-04,\n",
      "        1.03445098e-04, -6.74847397e-04, -2.53869919e-04,  3.27164598e-05,\n",
      "        1.91778545e-05,  1.90688937e-04,  1.52055960e-04,  3.09022398e-05,\n",
      "        2.18552923e-05,  2.76404992e-03, -2.16490589e-03,  1.79875700e-03,\n",
      "       -2.54042167e-03, -6.44364767e-03, -2.52420222e-03, -9.40119091e-04,\n",
      "       -7.05557177e-04,  1.54416694e-03, -5.69440192e-04,  6.81390404e-04,\n",
      "        3.63054802e-03, -3.80955287e-03,  3.08033405e-03,  7.20380805e-04,\n",
      "        4.05757362e-03, -5.28629811e-04, -1.17592223e-03, -5.07332105e-03,\n",
      "        1.85626652e-03, -3.37298331e-03,  7.62412674e-04,  8.56360944e-04,\n",
      "        5.09216567e-04, -3.25784739e-03, -1.51469911e-04,  1.47069022e-05,\n",
      "        1.18691996e-05,  1.42945919e-05,  1.69309515e-05,  1.18672660e-05,\n",
      "        1.31830766e-05,  1.40482680e-05,  1.33631138e-05,  1.32141513e-05,\n",
      "        1.33954345e-05,  1.44614505e-05,  1.37725401e-05,  1.22869469e-05,\n",
      "        1.28136671e-05,  1.33288986e-05,  1.29015698e-05,  1.40982847e-05,\n",
      "        1.65477268e-05,  2.22254148e-05], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafcb00>, <tf.Tensor: id=886580, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 3.1036856e-07, -1.0393370e-06, -2.8686870e-07, ...,\n",
      "         1.1283505e-05,  1.4970736e-05,  7.1279144e-05],\n",
      "       [ 7.1035407e-07,  3.1994836e-07, -1.9404276e-07, ...,\n",
      "        -1.4029039e-06, -7.5122666e-06,  1.1055386e-05],\n",
      "       [ 2.0456642e-07, -1.1740746e-07, -6.9135076e-08, ...,\n",
      "         9.5458654e-06,  7.1952581e-06,  5.5944576e-05],\n",
      "       ...,\n",
      "       [-2.5341063e-07, -1.0001706e-08,  9.6723866e-08, ...,\n",
      "         4.9310693e-06, -8.1125590e-06,  2.2366833e-06],\n",
      "       [ 8.2215202e-08,  6.5079938e-08, -3.6493366e-09, ...,\n",
      "        -3.0755828e-06,  3.7854047e-06, -5.0751805e-05],\n",
      "       [-5.9992453e-07, -1.3791150e-07,  1.1436682e-07, ...,\n",
      "         2.6498366e-05, -2.1198946e-05, -1.5837681e-05]], dtype=float32)>, <tf.Tensor: id=886581, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 7.6546394e-07, -2.9352389e-06,  6.2912668e-06, ...,\n",
      "         3.8398799e-04, -6.7214540e-05,  1.5490214e-05],\n",
      "       [-3.2845841e-07, -1.1166557e-06, -5.3816461e-07, ...,\n",
      "         1.8442181e-04, -2.9212319e-05, -2.8842653e-05],\n",
      "       [ 1.6960571e-07,  2.3692122e-07,  2.5090972e-07, ...,\n",
      "        -2.0263510e-04,  2.5368354e-05,  4.4012127e-06],\n",
      "       ...,\n",
      "       [ 5.3958665e-08,  2.8740044e-07,  2.9742992e-07, ...,\n",
      "         1.1508653e-05, -1.9434990e-06,  3.4571681e-07],\n",
      "       [-8.5754482e-07, -6.0863804e-06, -1.1245111e-06, ...,\n",
      "         3.7301582e-04, -5.6666391e-05, -2.2051619e-05],\n",
      "       [-1.7371377e-07, -2.2031097e-06, -5.2948303e-07, ...,\n",
      "         9.3306538e-05, -5.4970347e-05, -1.9592339e-06]], dtype=float32)>, <tf.Tensor: id=886582, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 7.3862952e-06,  1.9579231e-06, -3.0701617e-06, ...,\n",
      "        1.1787099e-05,  3.7617143e-04,  9.1056638e-05], dtype=float32)>, <tf.Tensor: id=886485, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.27007513e-06,  5.49207616e-05, -1.34005995e-06, ...,\n",
      "        -1.57609520e-06, -1.97174222e-06, -2.69251905e-06],\n",
      "       [-2.50214816e-06, -1.23128528e-04, -2.80698714e-06, ...,\n",
      "        -2.98757413e-06, -4.18445416e-06, -5.56833402e-06],\n",
      "       [-1.05979279e-06,  2.54580809e-04, -1.16840556e-06, ...,\n",
      "        -1.43294756e-06, -1.42798024e-06, -2.32172738e-06],\n",
      "       ...,\n",
      "       [ 1.20547775e-05, -6.66684122e-04,  1.13243914e-05, ...,\n",
      "         1.47265673e-05,  1.70150161e-05,  2.29368598e-05],\n",
      "       [-1.22323047e-06,  6.64827385e-05, -1.30065564e-06, ...,\n",
      "        -1.50980122e-06, -1.88071022e-06, -2.58170144e-06],\n",
      "       [-8.73556019e-07,  8.93474789e-05, -6.49036451e-07, ...,\n",
      "        -1.07843948e-06, -8.11487951e-07, -1.37413679e-06]], dtype=float32)>, <tf.Tensor: id=886477, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.24821654e-05, -7.00699340e-04,  1.16653982e-05,  4.21240955e-04,\n",
      "        6.75590709e-03,  3.23828353e-05,  1.28839296e-04,  1.59208594e-05,\n",
      "        1.55113194e-05,  1.55716316e-05,  1.01571949e-03,  1.29345077e-04,\n",
      "       -3.73628209e-05,  2.29933466e-05,  1.75519781e-05, -2.45765987e-04,\n",
      "       -2.66127230e-04,  5.33273793e-04,  5.37234118e-05, -4.95441142e-04,\n",
      "        7.62027848e-05, -3.26870766e-04, -3.62374238e-04,  1.42297111e-04,\n",
      "       -2.34125429e-04,  2.89855496e-04, -1.24164944e-04, -1.28834508e-04,\n",
      "        3.98210104e-04,  4.46378137e-04,  1.69267398e-04,  1.90788105e-05,\n",
      "        1.87251826e-05,  2.16178851e-05,  5.55907463e-05, -9.33615782e-04,\n",
      "       -1.35251277e-04,  3.69465735e-04,  1.38652504e-05, -1.17214222e-04,\n",
      "       -7.48317223e-04, -7.59143441e-05, -3.37544276e-04,  9.13631171e-04,\n",
      "       -3.55930883e-04,  6.65348489e-05, -6.63998289e-05,  6.29943796e-04,\n",
      "        2.50086072e-04, -3.93951952e-04, -3.47581867e-04,  4.68092439e-05,\n",
      "        5.31098398e-04,  7.64912227e-04,  9.61346785e-04,  1.83299329e-04,\n",
      "        1.08900436e-04,  8.22786707e-04,  6.27634654e-05,  1.92262407e-04,\n",
      "        2.00717677e-05,  1.94896013e-04,  1.55839734e-04,  3.30822804e-05,\n",
      "        2.29149573e-05, -6.76951837e-04,  2.36103265e-03,  3.82513809e-03,\n",
      "       -3.00512346e-03, -2.77912710e-03, -2.25413591e-03, -6.05047855e-04,\n",
      "       -1.22740446e-03, -3.70140246e-04, -8.15811945e-05, -6.30576687e-04,\n",
      "        8.67786934e-04, -1.33220383e-04,  3.99708748e-03, -5.85000031e-03,\n",
      "        1.10168185e-04, -4.83991389e-05,  4.18109260e-03, -4.75275749e-03,\n",
      "        4.93821339e-04, -3.89576773e-03,  2.40340712e-04, -1.01863872e-03,\n",
      "       -3.93115988e-06, -3.84582178e-04,  6.21203159e-04,  1.52803332e-05,\n",
      "        1.25397910e-05,  1.56406768e-05,  1.82930453e-05,  1.27965868e-05,\n",
      "        1.41336577e-05,  1.56351853e-05,  1.43241714e-05,  1.42395993e-05,\n",
      "        1.42277740e-05,  1.56612841e-05,  1.47303290e-05,  1.32607083e-05,\n",
      "        1.39710710e-05,  1.44915675e-05,  1.37071665e-05,  1.52322509e-05,\n",
      "        1.75256191e-05,  2.36468259e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc8d0>, <tf.Tensor: id=886941, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 1.1337797e-06, -3.3241881e-07, -1.8861653e-06, ...,\n",
      "        -4.0769977e-05,  9.2008268e-06,  2.3891855e-05],\n",
      "       [ 2.3398864e-07, -2.6531225e-08,  1.3061411e-08, ...,\n",
      "         1.0512656e-05, -2.0571939e-05,  6.6960406e-06],\n",
      "       [-2.5271416e-07, -8.4512600e-08,  3.7275452e-08, ...,\n",
      "         4.5755423e-06,  5.3480426e-06,  1.8475335e-05],\n",
      "       ...,\n",
      "       [ 2.8172894e-07,  1.8881441e-07, -9.2018844e-08, ...,\n",
      "        -1.3508718e-05,  7.3411613e-08,  1.6285188e-06],\n",
      "       [ 7.7097106e-08,  8.5452534e-08, -1.5683060e-08, ...,\n",
      "         2.7795186e-05, -7.7310688e-06, -5.4476129e-07],\n",
      "       [-3.2178471e-07,  8.1671629e-08,  2.8655422e-09, ...,\n",
      "         5.7245030e-05, -2.6051248e-05,  1.4495250e-05]], dtype=float32)>, <tf.Tensor: id=886942, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 6.68428447e-06,  1.00485013e-05,  1.28428337e-05, ...,\n",
      "         5.51825506e-04, -7.29054227e-05, -3.75549243e-05],\n",
      "       [ 1.16950886e-07,  6.81672248e-07,  2.21957544e-06, ...,\n",
      "         7.36481670e-05, -1.77223937e-05, -2.43342747e-05],\n",
      "       [ 1.08617968e-07,  3.50335455e-07, -1.01488297e-06, ...,\n",
      "        -2.73042067e-04,  4.90436469e-05,  2.39685960e-05],\n",
      "       ...,\n",
      "       [-3.26181748e-07, -7.61439367e-07,  6.70926283e-07, ...,\n",
      "         3.70619564e-05, -5.87867635e-06,  1.26365080e-06],\n",
      "       [ 7.89032401e-06,  1.54100744e-05,  2.01310158e-05, ...,\n",
      "         2.13614374e-04, -3.84028717e-05,  1.47652472e-05],\n",
      "       [-2.24725068e-06, -7.57383214e-06, -1.11993932e-05, ...,\n",
      "        -4.04560618e-04,  4.52019995e-05,  1.19353215e-04]], dtype=float32)>, <tf.Tensor: id=886943, shape=(6144,), dtype=float32, numpy=\n",
      "array([-2.5055671e-05,  8.1286188e-08,  1.2094914e-06, ...,\n",
      "        3.7100490e-05,  2.1972695e-04, -4.0943592e-04], dtype=float32)>, <tf.Tensor: id=886846, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.3197100e-06,  5.2510775e-05, -1.3892322e-06, ...,\n",
      "        -1.6279847e-06, -2.0348127e-06, -2.8272043e-06],\n",
      "       [-2.4395008e-06, -1.8345183e-04, -2.7843923e-06, ...,\n",
      "        -2.8963923e-06, -4.1070571e-06, -5.6365702e-06],\n",
      "       [-9.1405786e-07,  2.5862959e-04, -1.0848298e-06, ...,\n",
      "        -1.2489871e-06, -1.2722578e-06, -2.0982570e-06],\n",
      "       ...,\n",
      "       [ 1.2339702e-05, -8.2201121e-04,  1.1356902e-05, ...,\n",
      "         1.4895603e-05,  1.7014299e-05,  2.3451656e-05],\n",
      "       [-1.2749783e-06,  1.0540429e-04, -1.3255747e-06, ...,\n",
      "        -1.5662936e-06, -1.9130916e-06, -2.6969237e-06],\n",
      "       [-8.7413895e-07,  1.3853752e-04, -6.0027770e-07, ...,\n",
      "        -1.0305926e-06, -7.1939138e-07, -1.2558119e-06]], dtype=float32)>, <tf.Tensor: id=886838, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.27829990e-05, -8.41179164e-04,  1.16926612e-05,  9.16471006e-04,\n",
      "        8.46518297e-03, -1.24095648e-04,  3.57210811e-04,  1.63645218e-05,\n",
      "        1.56151546e-05,  1.59992305e-05, -1.66855892e-03,  2.82834517e-04,\n",
      "        6.11507974e-04,  2.32229813e-05,  1.77023248e-05, -4.18817290e-05,\n",
      "        1.35291251e-03, -2.93831248e-03,  5.36564257e-05,  6.53263822e-04,\n",
      "       -2.99580395e-04, -1.11175039e-04, -1.49054424e-04, -1.45046826e-04,\n",
      "       -3.62367660e-04,  3.47846100e-04, -9.52191986e-05, -1.14356910e-04,\n",
      "        5.77540195e-04, -3.66705353e-04,  1.75331821e-04,  1.93731539e-05,\n",
      "        1.88219637e-05,  2.17860161e-05,  2.08187557e-04,  1.92792591e-04,\n",
      "       -1.22718004e-04,  7.24473648e-05, -1.24335915e-04,  1.90426741e-04,\n",
      "       -5.61888446e-04,  2.51746329e-04,  3.51573079e-04,  6.00036641e-04,\n",
      "        2.81994638e-04,  8.05561358e-05,  4.16430848e-04, -5.79932937e-04,\n",
      "        1.14761482e-04, -5.16322383e-04,  4.62063472e-04, -1.06530126e-04,\n",
      "       -6.45973705e-05,  3.12118442e-04,  7.70852785e-06,  2.00200622e-04,\n",
      "       -3.56715726e-04,  2.93293531e-04, -9.18405713e-05, -1.16103038e-04,\n",
      "        2.06716904e-05, -4.25169448e-04, -4.68874350e-04,  3.34049946e-05,\n",
      "        2.30734531e-05,  4.66178637e-03,  7.66600249e-04,  2.23808514e-04,\n",
      "       -9.75924195e-05,  5.01125073e-03, -2.22921115e-03, -5.17730834e-04,\n",
      "        5.33506740e-03, -5.23050502e-03, -1.14382186e-03, -2.08173384e-04,\n",
      "       -2.95030233e-03,  4.89985803e-03,  4.92782914e-04, -8.58053192e-03,\n",
      "       -2.18214700e-05, -4.47809434e-05,  4.61126119e-03, -6.46992866e-03,\n",
      "       -7.24156154e-03, -1.15078734e-03,  1.74635591e-03,  6.51446171e-04,\n",
      "       -3.97271142e-05, -1.94078617e-04,  1.40158634e-04,  1.54362733e-05,\n",
      "        1.26324703e-05,  1.60463605e-05,  1.88525391e-05,  1.28268321e-05,\n",
      "        1.41348182e-05,  1.56508740e-05,  1.43297257e-05,  1.42248764e-05,\n",
      "        1.41895371e-05,  1.58227849e-05,  1.48549561e-05,  1.33761432e-05,\n",
      "        1.42908439e-05,  1.45623580e-05,  1.38014848e-05,  1.54042154e-05,\n",
      "        1.75213208e-05,  2.41774906e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc780>, <tf.Tensor: id=887302, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 4.9901024e-07, -4.9200958e-07, -1.6734252e-07, ...,\n",
      "        -6.2989311e-05,  1.0484748e-05, -1.9947777e-06],\n",
      "       [ 2.1283775e-07,  8.1788876e-08,  3.9449052e-08, ...,\n",
      "        -1.6206574e-06, -5.1456759e-06,  1.5489410e-05],\n",
      "       [-3.6941373e-07,  6.9184338e-08,  6.7321089e-08, ...,\n",
      "         2.6639325e-05, -9.2692062e-06,  1.4338120e-06],\n",
      "       ...,\n",
      "       [ 1.6121567e-07,  3.1166743e-07,  1.7069269e-07, ...,\n",
      "         6.5806544e-06, -2.9633929e-06,  1.0504203e-05],\n",
      "       [ 3.7602710e-08,  3.8262129e-08, -9.7709119e-10, ...,\n",
      "        -2.0015454e-05,  2.7409755e-05, -3.6194127e-05],\n",
      "       [ 1.6194394e-07, -1.2551301e-08,  9.3772606e-08, ...,\n",
      "         3.7414102e-05, -2.9812323e-05,  2.6415470e-05]], dtype=float32)>, <tf.Tensor: id=887303, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 1.2464835e-06,  1.5340163e-08,  7.3328806e-06, ...,\n",
      "         4.1487312e-04, -7.0499460e-05,  1.6543952e-05],\n",
      "       [-2.5153662e-07, -1.4404093e-06,  1.1594326e-06, ...,\n",
      "        -5.8947899e-05,  2.7796674e-05,  1.2993346e-05],\n",
      "       [ 1.7940198e-07,  1.6733559e-08,  3.2515450e-07, ...,\n",
      "         1.1435329e-04, -2.4932648e-05,  4.3365667e-06],\n",
      "       ...,\n",
      "       [ 5.4487617e-07,  1.0506214e-06,  8.4781880e-07, ...,\n",
      "         8.6110122e-06, -8.0477042e-07, -6.2738025e-07],\n",
      "       [ 4.1063563e-06,  2.4980438e-06,  7.6912547e-06, ...,\n",
      "        -2.2994254e-04,  3.6944333e-05, -1.7630038e-05],\n",
      "       [ 3.4548402e-06,  8.3282448e-06, -1.6460983e-05, ...,\n",
      "        -5.4135371e-04,  6.2768951e-05,  6.9302914e-05]], dtype=float32)>, <tf.Tensor: id=887304, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 4.6261575e-06, -8.1534111e-08, -2.6403691e-06, ...,\n",
      "        9.3100971e-06, -2.3277050e-04, -5.5206136e-04], dtype=float32)>, <tf.Tensor: id=887207, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.2979975e-06, -4.2632564e-06, -1.3618303e-06, ...,\n",
      "        -1.5880725e-06, -1.9969684e-06, -2.7843362e-06],\n",
      "       [-2.4315157e-06, -5.2516036e-05, -2.7185965e-06, ...,\n",
      "        -2.8703705e-06, -4.0029572e-06, -5.5269393e-06],\n",
      "       [-9.0315655e-07, -7.2245835e-05, -1.0400825e-06, ...,\n",
      "        -1.2471087e-06, -1.2455830e-06, -2.0750508e-06],\n",
      "       ...,\n",
      "       [ 1.1658660e-05,  5.7398540e-04,  1.0729479e-05, ...,\n",
      "         1.3913121e-05,  1.6169126e-05,  2.2203818e-05],\n",
      "       [-1.2270702e-06,  7.7377081e-07, -1.2813183e-06, ...,\n",
      "        -1.4769527e-06, -1.8408778e-06, -2.5940165e-06],\n",
      "       [-7.4559125e-07, -3.0884144e-04, -5.1407562e-07, ...,\n",
      "        -8.8588899e-07, -6.0964254e-07, -1.0591608e-06]], dtype=float32)>, <tf.Tensor: id=887199, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.19935767e-05,  5.80377702e-04,  1.09835837e-05, -4.67945531e-04,\n",
      "       -6.62711263e-03,  3.05467438e-05,  2.50600860e-04,  1.54883746e-05,\n",
      "        1.45275753e-05,  1.51170461e-05, -1.41377794e-03,  4.11177578e-04,\n",
      "        4.35219437e-04,  2.19327612e-05,  1.69713257e-05, -5.46885421e-04,\n",
      "       -8.62337693e-05, -3.02722678e-03,  5.08195626e-05, -5.52940728e-05,\n",
      "        6.12840449e-05,  4.25740145e-05,  1.26473242e-04,  2.87198869e-04,\n",
      "        2.46696174e-04, -3.29317627e-05,  3.25948204e-05, -2.93094025e-04,\n",
      "       -2.02859992e-05, -4.10076173e-04,  3.16623424e-04,  1.81973392e-05,\n",
      "        1.78794544e-05,  2.05511187e-05,  3.44116706e-05,  3.41002917e-04,\n",
      "       -1.10751006e-03,  5.15270047e-04, -2.96273385e-04, -1.66018817e-05,\n",
      "        7.21251999e-05,  9.09528462e-05, -1.13997739e-04, -1.12687843e-03,\n",
      "        2.72851408e-04, -8.14256127e-05,  9.18519872e-05,  2.08731697e-04,\n",
      "        2.62557354e-04,  9.60531470e-05,  1.29644686e-04,  4.96197135e-05,\n",
      "       -2.41731148e-04,  2.61365436e-04, -4.37481300e-04, -1.11893525e-04,\n",
      "       -3.58662597e-04,  4.07602434e-04,  6.31147559e-05,  2.92783316e-05,\n",
      "        1.96208002e-05,  3.84122868e-05, -1.59731615e-04,  3.20493054e-05,\n",
      "       -2.90689379e-04,  1.43504015e-03,  1.89871085e-03,  2.38769478e-03,\n",
      "        1.13858061e-03,  6.64046453e-03, -1.00826554e-03,  5.36500849e-03,\n",
      "        2.97432509e-03, -3.44303856e-03,  2.92426557e-04, -3.78269178e-05,\n",
      "       -2.05801707e-03, -1.15791929e-03, -2.08079023e-03,  3.46744200e-03,\n",
      "        3.49391904e-03,  1.22013400e-04, -7.73271255e-04,  1.21351890e-03,\n",
      "       -5.90292783e-03, -4.26030747e-04, -9.75757081e-04,  1.47805305e-03,\n",
      "       -8.57042964e-04, -1.67533383e-03, -5.12146275e-04,  1.45815166e-05,\n",
      "        1.20089508e-05,  1.47670398e-05,  1.72982491e-05,  1.19264841e-05,\n",
      "        1.31180668e-05,  1.46092134e-05,  1.35896980e-05,  1.34340144e-05,\n",
      "        1.34523752e-05,  1.46341335e-05,  1.38591768e-05,  1.22663032e-05,\n",
      "        1.30865028e-05,  1.32780551e-05,  1.30517792e-05,  1.42926319e-05,\n",
      "        1.65586680e-05,  2.27564815e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc550>, <tf.Tensor: id=887663, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 5.72362183e-07, -7.92706430e-07, -1.27954161e-06, ...,\n",
      "        -8.70148215e-05,  5.26358272e-05,  1.48827266e-05],\n",
      "       [-5.61409763e-08,  1.43705464e-07,  5.61481777e-08, ...,\n",
      "         5.14184758e-05, -9.06771220e-06,  3.91385038e-05],\n",
      "       [-4.63483580e-08, -4.18557029e-08,  2.69396434e-08, ...,\n",
      "        -1.18881444e-05,  2.90277294e-05,  7.66618177e-06],\n",
      "       ...,\n",
      "       [-4.28336591e-07, -1.61490078e-07,  1.18654896e-07, ...,\n",
      "         1.91821732e-06,  7.81403378e-06,  1.13739361e-05],\n",
      "       [-2.03627559e-09, -9.14025833e-09,  1.94721341e-08, ...,\n",
      "         4.86155332e-05, -8.15713720e-05, -4.86678728e-05],\n",
      "       [ 7.66067956e-08,  1.83151073e-07,  2.95202170e-07, ...,\n",
      "        -4.07699445e-05, -1.05940644e-05, -2.35117641e-05]], dtype=float32)>, <tf.Tensor: id=887664, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-6.1654612e-07, -3.2260666e-06, -9.7845323e-06, ...,\n",
      "        -3.5199089e-04,  2.5295511e-05, -2.5136160e-05],\n",
      "       [ 3.6050102e-07, -1.8079426e-08,  1.6519536e-07, ...,\n",
      "         7.7604753e-05, -2.3049315e-05,  9.8876917e-06],\n",
      "       [-5.1579889e-07, -1.0660462e-07, -1.6207126e-06, ...,\n",
      "         4.6991857e-04, -6.8836554e-05, -2.9196130e-05],\n",
      "       ...,\n",
      "       [-1.6619045e-07, -5.2343358e-07, -2.1419096e-07, ...,\n",
      "        -1.4535632e-05,  3.5383791e-06, -3.9725802e-07],\n",
      "       [ 5.9320428e-06,  7.9697529e-06,  2.1867007e-05, ...,\n",
      "        -3.8000793e-04,  5.2516538e-05, -1.8374511e-05],\n",
      "       [ 5.4390553e-06,  8.3239074e-06, -1.6760549e-05, ...,\n",
      "         1.4105083e-04, -3.9584047e-05, -1.4309918e-05]], dtype=float32)>, <tf.Tensor: id=887665, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 9.03865384e-06, -1.60317200e-08,  1.85567615e-06, ...,\n",
      "       -1.46154025e-05, -3.84821760e-04,  1.42737044e-04], dtype=float32)>, <tf.Tensor: id=887568, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.8032945e-07, -3.4910983e-05, -1.1426239e-06, ...,\n",
      "        -1.2091380e-06, -1.5922524e-06, -2.2538970e-06],\n",
      "       [-1.9190395e-06, -1.3529151e-04, -2.3245530e-06, ...,\n",
      "        -2.3006432e-06, -3.2642051e-06, -4.4971916e-06],\n",
      "       [-5.5265048e-07,  4.8220085e-05, -7.4142270e-07, ...,\n",
      "        -7.6294759e-07, -8.4911983e-07, -1.4715177e-06],\n",
      "       ...,\n",
      "       [ 6.9154967e-06,  3.6702285e-04,  7.4585773e-06, ...,\n",
      "         8.4012154e-06,  1.0595774e-05,  1.4573376e-05],\n",
      "       [-8.6072566e-07, -3.0394382e-05, -1.0290671e-06, ...,\n",
      "        -1.0584716e-06, -1.4094278e-06, -2.0018128e-06],\n",
      "       [-1.4754067e-07, -7.1420582e-05, -9.8602563e-08, ...,\n",
      "        -1.8250738e-07, -3.3083584e-08, -1.8342729e-07]], dtype=float32)>, <tf.Tensor: id=887560, shape=(110,), dtype=float32, numpy=\n",
      "array([ 7.05545244e-06,  3.70527414e-04,  7.58542910e-06, -5.46867785e-04,\n",
      "       -1.09339291e-02,  2.12845771e-05,  1.16308103e-03,  9.90202807e-06,\n",
      "        8.96590245e-06,  1.04291103e-05,  1.16795651e-03,  1.58161376e-04,\n",
      "        4.31928311e-06,  1.50005726e-05,  1.04721630e-05, -2.43350421e-03,\n",
      "       -9.51808397e-05,  2.59784283e-03,  3.72343602e-05,  1.00014200e-04,\n",
      "        9.39215359e-04,  3.98442440e-04,  4.89655358e-05,  1.13753325e-04,\n",
      "       -6.23115411e-05,  1.76051864e-04, -2.53714534e-05,  4.18907694e-05,\n",
      "       -1.71598454e-04, -3.03536770e-04,  1.28397238e-04,  1.09989205e-05,\n",
      "        1.13797323e-05,  1.31383485e-05,  1.51846980e-04,  3.25144298e-04,\n",
      "        1.44405916e-04, -9.28509980e-05,  2.42274094e-04, -1.99350270e-05,\n",
      "        2.79294385e-04,  1.95011889e-04,  9.36122342e-06, -5.86220995e-04,\n",
      "        5.43956339e-05,  2.55969462e-05,  1.72647837e-04, -2.50009849e-04,\n",
      "        1.61494609e-04,  8.10579877e-05,  2.95601174e-04,  3.63673498e-05,\n",
      "        5.50592027e-04,  2.14967265e-04, -2.10925675e-04,  1.58750845e-04,\n",
      "        8.42385125e-05, -1.04325736e-04, -2.67301366e-04,  1.21815043e-04,\n",
      "        1.35342307e-05, -3.26770794e-04, -3.43116582e-04,  2.00145732e-05,\n",
      "        1.55435391e-05, -1.53274077e-03,  3.81413585e-04, -1.66883518e-03,\n",
      "       -1.28734770e-04,  7.64805917e-03, -3.87782604e-03, -7.53118657e-04,\n",
      "       -5.48941898e-05,  2.74199550e-03,  6.30716560e-04, -4.59294097e-04,\n",
      "        4.32798028e-04, -7.84670177e-04, -6.77882694e-03,  3.25255282e-03,\n",
      "        1.54799328e-03, -2.13842170e-04, -2.06273887e-03, -2.31757481e-03,\n",
      "        2.82398914e-03,  2.99227377e-03,  1.98478252e-03,  5.10575541e-04,\n",
      "        1.67116697e-04,  1.15860673e-03,  4.29521206e-05,  9.89144701e-06,\n",
      "        8.05996933e-06,  8.77829552e-06,  9.45688043e-06,  7.18669708e-06,\n",
      "        8.25074130e-06,  7.55092833e-06,  7.66425546e-06,  7.57831776e-06,\n",
      "        8.04952197e-06,  8.27866643e-06,  8.72916280e-06,  7.46887554e-06,\n",
      "        7.45295165e-06,  8.17373711e-06,  8.26832547e-06,  8.56321549e-06,\n",
      "        1.07790411e-05,  1.48251884e-05], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdae76a0>, <tf.Tensor: id=888024, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-1.2291123e-07, -3.8444131e-07,  8.2069482e-07, ...,\n",
      "        -3.1301683e-05, -1.8692679e-05, -8.4705671e-06],\n",
      "       [-3.4729788e-07, -2.3518345e-07,  3.3864137e-07, ...,\n",
      "         6.9324210e-06, -2.5442589e-05, -3.0589858e-06],\n",
      "       [-5.4353217e-07, -1.3732523e-07,  8.7531467e-09, ...,\n",
      "        -2.1930255e-05,  1.4878005e-05, -7.5094513e-06],\n",
      "       ...,\n",
      "       [ 1.4108635e-07,  1.3637670e-07,  1.7452886e-07, ...,\n",
      "         7.1916552e-06, -2.4998149e-06,  3.4614297e-07],\n",
      "       [ 3.2025220e-08,  1.7245545e-08, -2.3123263e-08, ...,\n",
      "        -4.5908895e-05,  1.8242066e-05, -4.1493884e-05],\n",
      "       [-5.1885144e-07, -1.7108722e-07,  1.2070984e-07, ...,\n",
      "        -2.5062420e-05,  1.0913977e-05,  4.2917407e-05]], dtype=float32)>, <tf.Tensor: id=888025, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-2.56345174e-06, -6.93257152e-06, -1.29555531e-06, ...,\n",
      "         1.02677084e-04, -2.41491343e-05, -2.09215286e-05],\n",
      "       [-7.46774390e-07, -1.54954034e-06, -2.07220921e-07, ...,\n",
      "         1.41321347e-04, -7.17513103e-06,  7.82974985e-06],\n",
      "       [ 3.51656979e-07,  9.31698310e-07, -8.65996640e-07, ...,\n",
      "        -3.09047100e-05,  1.63558252e-05, -4.85300261e-05],\n",
      "       ...,\n",
      "       [ 2.30328681e-07, -6.02257444e-07, -2.48544808e-07, ...,\n",
      "        -3.49418951e-05,  4.15924296e-06, -1.45538479e-06],\n",
      "       [ 6.82069640e-06,  1.08392278e-05,  1.47970986e-05, ...,\n",
      "        -1.94034204e-04,  2.39351648e-05,  1.74252746e-05],\n",
      "       [-3.52880443e-06, -1.01427613e-05,  4.33199830e-06, ...,\n",
      "        -3.54844437e-04,  4.61031741e-05,  3.22792803e-05]], dtype=float32)>, <tf.Tensor: id=888026, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 1.2109818e-05,  5.0765093e-06, -3.1913708e-06, ...,\n",
      "       -3.5957095e-05, -1.9348766e-04, -3.6058063e-04], dtype=float32)>, <tf.Tensor: id=887929, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.17294371e-06, -1.34967049e-05, -1.27038606e-06, ...,\n",
      "        -1.40906991e-06, -1.81693201e-06, -2.53404232e-06],\n",
      "       [-1.96867040e-06, -1.42130593e-04, -2.33738092e-06, ...,\n",
      "        -2.28933277e-06, -3.34692868e-06, -4.62508342e-06],\n",
      "       [-6.89415515e-07, -1.01274200e-05, -8.31134912e-07, ...,\n",
      "        -9.15855139e-07, -9.75510034e-07, -1.60220543e-06],\n",
      "       ...,\n",
      "       [ 9.90907211e-06,  9.95450537e-05,  9.31980958e-06, ...,\n",
      "         1.16879755e-05,  1.36186973e-05,  1.86760335e-05],\n",
      "       [-1.11002464e-06,  1.53351029e-05, -1.17761044e-06, ...,\n",
      "        -1.33717242e-06, -1.63785671e-06, -2.32493471e-06],\n",
      "       [-4.96504299e-07, -7.01121025e-05, -2.90334754e-07, ...,\n",
      "        -5.82084908e-07, -2.85178032e-07, -5.36908885e-07]], dtype=float32)>, <tf.Tensor: id=887921, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.02168351e-05,  9.99978365e-05,  9.55172891e-06,  3.82613245e-04,\n",
      "        5.18258545e-04, -2.86276452e-04,  5.78887877e-04,  1.32751202e-05,\n",
      "        1.24305298e-05,  1.33139092e-05, -5.72194229e-04, -3.32033262e-04,\n",
      "       -4.08176886e-04,  1.88544400e-05,  1.44075548e-05, -2.35074759e-03,\n",
      "       -1.68180279e-03, -1.89194034e-05,  4.36379269e-05,  8.53397069e-04,\n",
      "       -3.07584443e-04, -1.66410187e-04, -2.70542514e-05, -5.41704103e-05,\n",
      "        8.57168052e-05, -1.78354821e-05, -6.68218490e-07,  2.53831695e-06,\n",
      "       -3.42722982e-04,  3.62012797e-04, -1.44179012e-05,  1.53206365e-05,\n",
      "        1.52035273e-05,  1.77510246e-05,  5.52186248e-06, -8.22704169e-06,\n",
      "        3.81791440e-04, -8.40353838e-04, -6.54636649e-04, -1.34539121e-04,\n",
      "        3.45581415e-04,  5.46655065e-05, -1.89905986e-04,  1.09799788e-03,\n",
      "        2.31467740e-04,  5.47022501e-05,  2.03457384e-04,  3.02152941e-04,\n",
      "        2.06509067e-04,  1.74203509e-04,  1.66461832e-04,  4.59759358e-05,\n",
      "       -2.52127211e-05,  2.61175621e-04, -4.72922809e-04,  2.98742216e-05,\n",
      "        1.01362624e-04, -9.21406827e-05,  5.78102226e-05, -4.73875727e-04,\n",
      "        1.67207399e-05,  1.72191663e-04, -5.26056056e-06,  2.72172856e-05,\n",
      "        1.96500405e-05, -1.26131577e-04,  6.11135620e-05, -2.24679592e-03,\n",
      "        2.63816793e-03, -7.04000471e-04, -1.34201557e-03,  1.86824612e-03,\n",
      "        6.38334139e-04, -2.26932298e-03, -1.67927908e-04, -9.13662254e-04,\n",
      "        9.44020343e-04, -1.41845038e-03,  1.94174674e-04,  8.64785630e-03,\n",
      "       -4.09016525e-03,  1.14799972e-04, -3.84951197e-03,  2.85930117e-03,\n",
      "       -2.94554251e-04,  2.49944557e-03,  2.27903575e-03, -2.00121198e-03,\n",
      "        1.04959428e-04, -6.50712405e-04, -5.50534925e-04,  1.27672711e-05,\n",
      "        1.03207731e-05,  1.24989838e-05,  1.45670674e-05,  1.01037303e-05,\n",
      "        1.12452835e-05,  1.22547708e-05,  1.13083361e-05,  1.09634038e-05,\n",
      "        1.11678210e-05,  1.21160510e-05,  1.16995443e-05,  1.02901813e-05,\n",
      "        1.09821785e-05,  1.12166408e-05,  1.10864257e-05,  1.20389350e-05,\n",
      "        1.39630420e-05,  1.91643230e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdae6198>, <tf.Tensor: id=888385, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-1.55022360e-06, -1.56929696e-06,  2.07923108e-07, ...,\n",
      "        -7.08400112e-05, -2.33889623e-05, -2.35672796e-05],\n",
      "       [ 1.45247469e-07,  2.13451415e-07,  4.36902567e-08, ...,\n",
      "        -2.57555985e-05, -2.87549010e-05, -1.00267998e-05],\n",
      "       [-2.24354523e-07, -2.08699613e-09,  5.95594010e-08, ...,\n",
      "         1.25060515e-05,  1.82009280e-05,  2.28615536e-05],\n",
      "       ...,\n",
      "       [-6.46248566e-07, -3.57078335e-07,  3.84192504e-07, ...,\n",
      "         6.63207902e-06, -2.86800127e-06,  9.61120531e-06],\n",
      "       [ 4.11258334e-08,  2.48179326e-08, -1.55771787e-08, ...,\n",
      "         4.34575395e-06, -9.46208365e-06, -7.72813382e-06],\n",
      "       [ 1.81498564e-07,  7.17970323e-08, -5.12874436e-08, ...,\n",
      "        -1.53155215e-05,  1.65289384e-05,  3.47892201e-05]], dtype=float32)>, <tf.Tensor: id=888386, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 2.0775606e-06,  3.9432616e-06,  6.2679769e-06, ...,\n",
      "        -1.2501137e-04,  1.0958679e-05, -2.5630872e-05],\n",
      "       [-5.6399324e-07, -8.9436230e-07,  2.8963572e-07, ...,\n",
      "         2.0933355e-04, -2.6631586e-05,  1.6222752e-05],\n",
      "       [ 2.6218498e-08,  5.6699304e-07, -1.9088000e-06, ...,\n",
      "        -1.1985839e-04,  3.1382093e-05,  4.1241256e-06],\n",
      "       ...,\n",
      "       [ 4.9053455e-07,  6.7175313e-07, -3.3468964e-07, ...,\n",
      "        -2.2381962e-05,  3.4770151e-06, -2.4488095e-06],\n",
      "       [ 1.9140273e-06, -8.5082030e-07,  8.5377878e-06, ...,\n",
      "        -1.9105992e-04,  2.3401341e-05, -4.5541383e-05],\n",
      "       [-1.1212492e-06, -3.5924813e-06, -1.0225172e-05, ...,\n",
      "        -5.5622350e-04,  6.4746127e-05,  3.6052348e-05]], dtype=float32)>, <tf.Tensor: id=888387, shape=(6144,), dtype=float32, numpy=\n",
      "array([-1.1628438e-05,  4.4314729e-06, -1.8618289e-08, ...,\n",
      "       -2.2687796e-05, -1.9104980e-04, -5.6045840e-04], dtype=float32)>, <tf.Tensor: id=888290, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.0548069e-06,  2.1546684e-05, -1.2046337e-06, ...,\n",
      "        -1.2810208e-06, -1.6883430e-06,  1.7885164e-05],\n",
      "       [-1.9082677e-06, -1.2063680e-04, -2.3127068e-06, ...,\n",
      "        -2.2580475e-06, -3.2498328e-06,  4.1992764e-05],\n",
      "       [-5.6976535e-07,  1.5145575e-04, -7.4007465e-07, ...,\n",
      "        -7.3082288e-07, -8.6978139e-07, -5.4547636e-05],\n",
      "       ...,\n",
      "       [ 7.3020328e-06,  2.2990511e-04,  7.6861961e-06, ...,\n",
      "         8.7542530e-06,  1.0989099e-05, -1.4001215e-04],\n",
      "       [-9.2022316e-07, -6.7189594e-06, -1.0549064e-06, ...,\n",
      "        -1.1177963e-06, -1.4497009e-06,  2.7519494e-05],\n",
      "       [-4.3787196e-08,  7.9833089e-06,  5.6908029e-08, ...,\n",
      "        -2.0162616e-08,  1.8430136e-07, -3.1335847e-05]], dtype=float32)>, <tf.Tensor: id=888282, shape=(110,), dtype=float32, numpy=\n",
      "array([ 7.46322712e-06,  2.51062302e-04,  7.82267944e-06, -5.21470269e-04,\n",
      "        1.07042601e-04,  2.27681085e-05,  1.18843443e-03,  1.03164157e-05,\n",
      "        9.42102542e-06,  1.10057417e-05,  7.01068726e-04,  4.23944555e-04,\n",
      "        6.57653145e-04,  1.55212965e-05,  1.11242916e-05, -2.03428534e-03,\n",
      "        4.88549529e-04,  2.05839286e-03,  3.76825919e-05,  3.72795737e-04,\n",
      "        6.38244906e-04,  1.25745501e-04,  2.69667915e-04,  1.42741221e-04,\n",
      "       -1.73609733e-04, -6.73144823e-05, -2.93602614e-04, -8.62207962e-05,\n",
      "        4.15198083e-05, -4.31085355e-04,  4.46847494e-04,  1.16188849e-05,\n",
      "        1.18655325e-05,  1.37346387e-05,  1.48085455e-04, -1.58824885e-04,\n",
      "        3.40932602e-04,  5.31005091e-04,  1.22305748e-04,  1.20515921e-04,\n",
      "        1.58796072e-04,  3.56034347e-04, -3.96831718e-04, -6.21267245e-05,\n",
      "        2.08419980e-04, -1.16097493e-04,  1.77510185e-04,  4.18803538e-04,\n",
      "       -1.42821271e-04,  4.26270330e-04, -4.13945236e-05,  4.02130318e-05,\n",
      "        3.93390161e-04, -5.88496565e-04, -5.56654297e-04, -1.39694952e-04,\n",
      "       -3.78791650e-04, -3.54638178e-05,  4.98976660e-05,  1.29868073e-04,\n",
      "        1.40587690e-05,  6.86103704e-07, -1.30522067e-05,  2.10835733e-05,\n",
      "        1.65566689e-05, -5.77196525e-03,  1.22906198e-03, -2.39079259e-03,\n",
      "        3.56383598e-03,  5.00120083e-03, -2.15366390e-03,  1.27003004e-03,\n",
      "       -4.67044208e-03, -8.81107710e-03,  3.21248866e-04,  1.71242503e-03,\n",
      "       -1.00326235e-03, -1.00405421e-03,  3.58931813e-03,  4.37590573e-03,\n",
      "       -7.00203469e-04, -3.46497371e-04, -8.22958129e-04,  2.09083693e-04,\n",
      "       -1.99653581e-03,  8.99385021e-04,  1.70754222e-03, -3.16225371e-04,\n",
      "       -1.06842563e-04,  1.17033813e-03, -1.80335919e-04,  1.02306867e-05,\n",
      "        8.33179820e-06, -1.46908686e-04,  1.01160731e-05,  7.49575247e-06,\n",
      "       -1.47598199e-04,  8.25343250e-06,  8.15196745e-06,  8.05573382e-06,\n",
      "        8.69975520e-06,  8.75569185e-06,  8.95212997e-06,  7.77890091e-06,\n",
      "        8.00263842e-06,  8.59258580e-06,  8.62389061e-06,  8.93668130e-06,\n",
      "        1.11886256e-05, -1.40833130e-04], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc588>, <tf.Tensor: id=888746, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-2.1959563e-06, -6.7273601e-07,  2.6650949e-07, ...,\n",
      "        -2.6553325e-05, -6.4970259e-06, -4.2368364e-07],\n",
      "       [ 5.7083065e-07,  4.1243135e-07,  1.5383573e-07, ...,\n",
      "        -7.9032643e-06,  1.5911535e-05, -1.2113537e-05],\n",
      "       [-5.5011014e-08,  2.3123390e-07, -1.2613377e-07, ...,\n",
      "        -4.4384637e-05,  2.4537154e-05,  4.3507976e-06],\n",
      "       ...,\n",
      "       [-1.7720998e-07,  1.3399863e-07, -2.9849289e-08, ...,\n",
      "         3.5914925e-07,  5.0370231e-06,  3.6865015e-06],\n",
      "       [ 8.5743572e-09,  3.7261527e-09,  1.8621101e-08, ...,\n",
      "        -4.1697909e-05, -1.3523704e-05, -5.0488732e-05],\n",
      "       [ 3.1176273e-07, -6.0968240e-08,  1.8975155e-07, ...,\n",
      "        -2.2677717e-05, -8.9863879e-06,  1.3463481e-05]], dtype=float32)>, <tf.Tensor: id=888747, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-3.3194112e-06, -6.9164857e-06, -5.3712274e-06, ...,\n",
      "        -3.7762153e-04,  5.4468939e-05, -5.9517457e-05],\n",
      "       [ 3.8508162e-07, -9.9118239e-08,  6.6486155e-07, ...,\n",
      "        -6.6541092e-05,  1.7248194e-05,  9.6327767e-06],\n",
      "       [ 2.8380987e-07,  5.2782212e-07, -1.3784188e-06, ...,\n",
      "        -4.2631110e-04,  7.7227138e-05,  3.8844287e-06],\n",
      "       ...,\n",
      "       [ 2.8245017e-07,  5.7405396e-07, -5.4011798e-07, ...,\n",
      "        -3.8141709e-05,  4.9319187e-06, -1.2026629e-06],\n",
      "       [ 7.9676289e-07, -3.2194168e-06,  4.3747550e-06, ...,\n",
      "         9.8466080e-05, -1.5749794e-05, -2.2345595e-05],\n",
      "       [-3.9952442e-06, -1.4274954e-05, -4.6231007e-06, ...,\n",
      "        -9.9044464e-06, -1.7994331e-05,  4.2403200e-05]], dtype=float32)>, <tf.Tensor: id=888748, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 9.0422527e-06, -8.5699116e-07, -5.2412793e-07, ...,\n",
      "       -3.9237988e-05,  1.0028278e-04, -7.5579192e-06], dtype=float32)>, <tf.Tensor: id=888651, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.0343525e-06, -5.9729158e-05, -1.1825991e-06, ...,\n",
      "        -1.2418673e-06, -1.6587379e-06, -2.3390648e-06],\n",
      "       [-1.9882134e-06, -1.3254405e-04, -2.3371801e-06, ...,\n",
      "        -2.2936970e-06, -3.3190540e-06, -4.5843835e-06],\n",
      "       [-5.8509170e-07, -4.1989842e-06, -7.6076935e-07, ...,\n",
      "        -7.9887866e-07, -8.9631715e-07, -1.5068842e-06],\n",
      "       ...,\n",
      "       [ 7.1801314e-06,  7.1660767e-04,  7.5659063e-06, ...,\n",
      "         8.4873063e-06,  1.0880222e-05,  1.5033232e-05],\n",
      "       [-8.7315323e-07, -6.6923414e-05, -1.0135693e-06, ...,\n",
      "        -1.0395675e-06, -1.3958074e-06, -1.9843728e-06],\n",
      "       [ 7.1614231e-08, -1.1739922e-04,  1.3860947e-07, ...,\n",
      "         8.3155172e-08,  3.0484023e-07,  2.9581958e-07]], dtype=float32)>, <tf.Tensor: id=888643, shape=(110,), dtype=float32, numpy=\n",
      "array([ 7.33302932e-06,  7.28177140e-04,  7.70637598e-06, -2.04743468e-04,\n",
      "        1.04672601e-02, -1.33026202e-04,  9.88222964e-05,  1.01578344e-05,\n",
      "        9.16482804e-06,  1.09069888e-05, -4.36267146e-04, -7.78614412e-05,\n",
      "       -2.68663309e-04,  1.54053923e-05,  1.10540059e-05, -9.55809257e-04,\n",
      "       -7.76713190e-04,  9.41842329e-04, -1.18603406e-04, -7.08407242e-05,\n",
      "        6.53050782e-04,  5.68201882e-04,  5.88401381e-05,  1.07377600e-04,\n",
      "       -2.01209783e-04,  2.23203315e-05,  2.92678305e-04,  4.51414235e-05,\n",
      "       -1.97506597e-04,  7.58103779e-05, -1.86352263e-05,  1.14208933e-05,\n",
      "        1.16619067e-05,  1.33136537e-05,  1.46614417e-04,  1.80272273e-05,\n",
      "       -6.10472460e-04, -1.16700721e-04,  2.76061706e-04,  9.98047326e-05,\n",
      "       -3.17522354e-04,  2.83267709e-05,  2.49087519e-04, -4.50547785e-04,\n",
      "        4.21992081e-05,  1.96178837e-04, -1.47833693e-04,  5.50621597e-04,\n",
      "        3.20886553e-04,  1.03040176e-04,  8.84307083e-05,  4.03443846e-05,\n",
      "        2.09795937e-04, -4.56026581e-04, -3.78812925e-04,  1.79248855e-05,\n",
      "       -6.55780896e-05, -2.00769849e-04,  5.02269977e-05, -2.73652895e-05,\n",
      "        1.35973796e-05,  5.28609235e-07, -7.16864452e-06,  2.07075245e-05,\n",
      "        1.65514648e-05, -7.54855806e-03,  9.26866895e-04, -7.43873464e-03,\n",
      "        2.64866068e-03, -7.48722535e-03,  1.62386394e-03, -3.46488872e-04,\n",
      "       -4.99291578e-03,  1.65946456e-03, -1.42072066e-04,  1.79486768e-03,\n",
      "       -2.85156234e-03,  3.03153973e-03,  2.29210797e-04,  2.00351747e-03,\n",
      "       -8.10525788e-04, -1.79464376e-04, -7.04581663e-03,  3.78608983e-03,\n",
      "        6.09117746e-03,  1.72145537e-03, -1.38236582e-03,  1.95201649e-03,\n",
      "        8.02410883e-04,  1.09949650e-03,  2.66644522e-04,  9.83228620e-06,\n",
      "        8.26259020e-06,  9.22176514e-06,  9.73478382e-06,  7.34570267e-06,\n",
      "        8.67475137e-06,  7.94859716e-06,  7.97558550e-06,  7.99470126e-06,\n",
      "        8.86442467e-06,  8.59203647e-06,  8.79921299e-06,  7.51032576e-06,\n",
      "        7.85569955e-06,  8.37094558e-06,  8.48110085e-06,  8.65708353e-06,\n",
      "        1.10836409e-05,  1.53136534e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc9b0>, <tf.Tensor: id=889107, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 1.22855351e-06, -1.43631519e-06, -4.29735536e-07, ...,\n",
      "         6.07925085e-06, -2.82958772e-05,  2.91313900e-05],\n",
      "       [ 2.48119022e-07,  1.70482153e-07, -8.13995271e-09, ...,\n",
      "         5.69717986e-06,  8.54023165e-06, -2.86391460e-05],\n",
      "       [-2.87675221e-07, -7.16487918e-08,  1.88642886e-07, ...,\n",
      "        -1.12232528e-05, -6.81646497e-06,  5.38398444e-06],\n",
      "       ...,\n",
      "       [-6.75200965e-07, -2.39766024e-07,  1.69023338e-07, ...,\n",
      "        -1.10286965e-05, -4.59080502e-06, -4.42042801e-06],\n",
      "       [-2.50233132e-08,  1.57910822e-08,  2.18081109e-08, ...,\n",
      "        -1.92691568e-05,  2.56010371e-05, -1.70845378e-05],\n",
      "       [ 4.43447561e-07,  2.22481802e-07, -1.70856268e-07, ...,\n",
      "        -2.13131843e-05, -1.25544311e-05,  1.67807884e-05]], dtype=float32)>, <tf.Tensor: id=889108, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-3.2437790e-06, -4.3337518e-06, -4.5075049e-06, ...,\n",
      "        -2.3438965e-04,  1.7842236e-05, -6.2636675e-05],\n",
      "       [ 3.3920983e-07, -1.3738752e-07,  1.0202231e-07, ...,\n",
      "        -1.5598726e-04,  2.5129993e-05,  2.5680372e-05],\n",
      "       [-1.1677803e-07,  5.3900072e-07, -1.0812860e-06, ...,\n",
      "        -1.1820538e-04,  2.7228765e-05, -1.8142984e-05],\n",
      "       ...,\n",
      "       [-2.3384273e-07,  1.8388178e-08,  5.4312534e-07, ...,\n",
      "        -9.6460462e-06,  1.3283698e-06, -1.7992927e-06],\n",
      "       [-2.4036194e-07, -6.0356751e-06, -2.0954767e-06, ...,\n",
      "         1.3272796e-04, -1.7746588e-05,  1.5628775e-05],\n",
      "       [-1.4935851e-05, -3.7206493e-05,  2.9680024e-05, ...,\n",
      "         4.3772586e-04, -6.0762981e-05,  1.5090764e-05]], dtype=float32)>, <tf.Tensor: id=889109, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 1.0964497e-05,  9.2336563e-08,  1.9510776e-06, ...,\n",
      "       -9.4266943e-06,  1.3373914e-04,  4.4016427e-04], dtype=float32)>, <tf.Tensor: id=889012, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.15764078e-06,  2.63263646e-04, -1.23405152e-06, ...,\n",
      "        -1.34366132e-06, -1.76995786e-06, -2.53049075e-06],\n",
      "       [-2.03617128e-06,  3.98271717e-04, -2.29966281e-06, ...,\n",
      "        -2.28827230e-06, -3.30453076e-06, -4.67640166e-06],\n",
      "       [-6.70478471e-07, -1.85458441e-04, -8.26700330e-07, ...,\n",
      "        -8.86973169e-07, -9.90680064e-07, -1.57304180e-06],\n",
      "       ...,\n",
      "       [ 9.59369117e-06, -3.19192861e-03,  8.86911221e-06, ...,\n",
      "         1.09691118e-05,  1.31445795e-05,  1.85555182e-05],\n",
      "       [-9.87713861e-07,  3.14126315e-04, -1.03440539e-06, ...,\n",
      "        -1.14331374e-06, -1.45393130e-06, -2.12704322e-06],\n",
      "       [-3.25189603e-07,  7.14089430e-04, -8.06835274e-08, ...,\n",
      "        -3.62546359e-07, -4.53367086e-08, -2.55148962e-07]], dtype=float32)>, <tf.Tensor: id=889004, shape=(110,), dtype=float32, numpy=\n",
      "array([ 9.90864555e-06, -3.42214014e-03,  9.10054223e-06, -3.52179748e-04,\n",
      "        2.51202215e-03,  2.68875810e-05,  6.89201872e-04,  1.30788594e-05,\n",
      "        1.20619961e-05,  1.33073827e-05, -1.13878096e-03,  2.27974509e-04,\n",
      "        1.99885370e-04,  1.84232485e-05,  1.44861842e-05,  2.17950111e-03,\n",
      "        3.13401979e-04,  1.46317878e-04,  4.19121716e-05,  1.58764044e-04,\n",
      "        3.37559846e-04,  2.40971596e-04,  4.53327302e-05,  2.70439079e-04,\n",
      "        4.27460065e-04,  1.75449357e-04,  1.80955321e-04,  1.59305964e-05,\n",
      "       -4.03370155e-04, -3.97062860e-04,  1.14474140e-04,  1.51565891e-05,\n",
      "        1.46698540e-05,  1.70313979e-05,  1.34304049e-04,  3.57669400e-04,\n",
      "       -7.80874398e-05,  3.95145704e-04, -1.44199967e-05, -8.39699169e-07,\n",
      "        3.59578815e-04,  3.37465681e-05, -6.03579902e-05, -4.33439505e-04,\n",
      "        5.45899602e-05, -2.48891942e-04, -1.26928615e-04,  4.41507029e-04,\n",
      "       -1.15991046e-04,  3.75812466e-04,  2.75817642e-04,  4.84182165e-05,\n",
      "        2.59860419e-04, -8.48754862e-05, -1.07848621e-03,  4.37973831e-05,\n",
      "       -2.01736839e-04,  3.16748832e-04, -9.42289262e-05,  1.57239876e-04,\n",
      "        1.60940799e-05,  1.79861541e-04,  1.61330143e-04,  2.63461006e-05,\n",
      "        1.97178324e-05, -5.03732916e-03,  3.21989181e-03, -2.83969869e-03,\n",
      "        2.55855126e-03, -1.12535991e-02,  8.93592602e-04, -2.30319449e-03,\n",
      "       -1.55194371e-03,  2.92547978e-04, -1.34699716e-04,  1.64960953e-03,\n",
      "       -3.76674463e-03,  3.48505739e-04,  4.24607424e-04, -3.12962756e-03,\n",
      "       -5.48604759e-04, -4.82620642e-04,  3.96131584e-03,  5.18477289e-04,\n",
      "        5.03993360e-03,  1.73175056e-03, -6.41434977e-04,  1.53396232e-03,\n",
      "        6.79641496e-04,  4.78379522e-03, -1.74677407e-05,  1.19570859e-05,\n",
      "        1.02317763e-05,  1.24548806e-05,  1.38028154e-05,  9.97149709e-06,\n",
      "        1.10543206e-05,  1.16389110e-05,  1.09403854e-05,  1.09433486e-05,\n",
      "        1.11679137e-05,  1.16696656e-05,  1.11469635e-05,  1.00306979e-05,\n",
      "        1.07527248e-05,  1.10809860e-05,  1.05991885e-05,  1.13124161e-05,\n",
      "        1.35016171e-05,  1.90879873e-05], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc518>, <tf.Tensor: id=889468, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 2.12514942e-06, -4.30143075e-07, -5.52845677e-07, ...,\n",
      "        -1.90090341e-05, -4.54943984e-06,  2.65667186e-05],\n",
      "       [-8.07125389e-08, -2.98955136e-08, -1.38342102e-07, ...,\n",
      "         4.41361590e-06,  4.56988637e-05, -5.50793175e-06],\n",
      "       [-6.61572869e-07,  3.82869132e-08,  1.87039817e-07, ...,\n",
      "        -2.17803317e-05,  2.97595834e-05, -1.71621887e-05],\n",
      "       ...,\n",
      "       [ 5.35297545e-07,  1.91635330e-08, -9.75905508e-08, ...,\n",
      "         8.15444673e-06, -4.97278643e-06,  3.25660039e-06],\n",
      "       [-2.12612132e-08,  4.91199721e-08,  7.69868080e-09, ...,\n",
      "        -9.18863298e-05,  5.65555201e-05,  1.87585056e-06],\n",
      "       [ 1.30558050e-07,  1.16622076e-07,  3.21251832e-07, ...,\n",
      "        -1.73448898e-05, -3.82582075e-05, -4.30033697e-06]], dtype=float32)>, <tf.Tensor: id=889469, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-2.9881985e-06, -6.5921690e-06,  1.1328561e-07, ...,\n",
      "        -1.4117279e-04, -3.5264943e-06, -2.5158017e-05],\n",
      "       [-8.3786901e-08, -8.5731642e-07, -5.0962893e-07, ...,\n",
      "        -7.0416403e-04,  8.2747523e-05, -2.2888940e-05],\n",
      "       [-3.0780473e-07, -7.5606289e-07,  1.6661440e-06, ...,\n",
      "         4.5389857e-04, -7.0063965e-05,  2.0305773e-05],\n",
      "       ...,\n",
      "       [-1.1043609e-06, -1.3940974e-06, -1.3225426e-06, ...,\n",
      "         7.6194003e-05, -7.8400544e-06, -1.0821131e-07],\n",
      "       [ 3.0807371e-06, -4.0119494e-06,  1.3518679e-05, ...,\n",
      "        -6.4788514e-04,  7.2039809e-05, -3.1030497e-05],\n",
      "       [ 3.9492451e-07,  2.9845705e-07,  7.8145433e-07, ...,\n",
      "         3.3383019e-04, -4.9018839e-05, -2.2213379e-05]], dtype=float32)>, <tf.Tensor: id=889470, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 1.8476168e-05, -2.0180071e-06,  2.1771880e-06, ...,\n",
      "        8.3566047e-05, -6.8237231e-04,  3.3987546e-04], dtype=float32)>, <tf.Tensor: id=889373, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.3305865e-06,  1.6486624e-04, -1.3895544e-06, ...,\n",
      "        -1.5037149e-06, -1.9198387e-06, -2.7587876e-06],\n",
      "       [-2.1745720e-06,  3.0424024e-04, -2.4468054e-06, ...,\n",
      "        -2.3639009e-06, -3.4431084e-06, -4.9521268e-06],\n",
      "       [-8.2014469e-07, -9.2530550e-05, -1.0166532e-06, ...,\n",
      "        -1.0564136e-06, -1.1565592e-06, -1.7771101e-06],\n",
      "       ...,\n",
      "       [ 1.2560671e-05, -1.5921658e-03,  1.1372731e-05, ...,\n",
      "         1.4008555e-05,  1.6074579e-05,  2.2496621e-05],\n",
      "       [-1.2317985e-06,  1.7294096e-04, -1.2366434e-06, ...,\n",
      "        -1.3813756e-06, -1.6928950e-06, -2.4820599e-06],\n",
      "       [-5.4297880e-07,  1.3869061e-04, -3.1261786e-07, ...,\n",
      "        -6.5166842e-07, -2.8070431e-07, -4.7894952e-07]], dtype=float32)>, <tf.Tensor: id=889365, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.36550616e-05, -1.63171720e-03,  1.22132060e-05, -1.03231624e-03,\n",
      "       -1.12372488e-02,  3.33375792e-05,  2.73376587e-04,  1.71729371e-05,\n",
      "        1.59261417e-05,  1.75309524e-05, -2.38397231e-04,  7.27637816e-05,\n",
      "        3.59623868e-04,  2.30520600e-05,  1.85367280e-05,  2.34224088e-03,\n",
      "        8.22520815e-04,  4.94313645e-05,  4.83767071e-05,  5.90027950e-04,\n",
      "        1.13212838e-04,  9.12889009e-05,  2.21314593e-04,  4.28853731e-04,\n",
      "       -1.81064170e-05,  5.00364171e-04,  5.16083906e-04, -1.15429677e-04,\n",
      "       -3.91340494e-04,  5.55817038e-04,  2.57724634e-04,  1.98973612e-05,\n",
      "        1.91715335e-05,  2.15802229e-05,  2.87044561e-04, -1.04099757e-03,\n",
      "       -8.28452001e-04, -5.59077016e-05, -1.56739334e-04, -7.68688624e-04,\n",
      "       -2.44897528e-04, -4.28215222e-04, -1.64674784e-04, -1.26794702e-03,\n",
      "       -4.05069994e-04,  8.24565504e-05, -4.23101010e-04, -1.97686779e-04,\n",
      "       -1.19508314e-03, -1.33863848e-03,  1.21930541e-04,  5.66967610e-05,\n",
      "        1.12175774e-04, -1.00443384e-03, -1.64803409e-03, -2.47123506e-04,\n",
      "        1.27595471e-04,  4.84314223e-04,  7.24794954e-05, -2.95836478e-04,\n",
      "        2.01622643e-05, -2.74416845e-04, -1.36010989e-04,  3.19046849e-05,\n",
      "        2.51364490e-05,  6.84433756e-03,  1.31256948e-03, -3.76137393e-03,\n",
      "        3.86734027e-03, -1.74235774e-03,  2.73229554e-03,  2.27056281e-03,\n",
      "        2.66018137e-03,  3.01143038e-03,  9.75101066e-05,  2.12683901e-03,\n",
      "        1.14223757e-03, -1.50867330e-03,  1.04304578e-03, -1.68095308e-03,\n",
      "       -1.57518964e-03, -6.06826565e-04,  1.70217711e-03, -2.76571698e-03,\n",
      "        1.71132246e-03, -1.86578080e-03, -2.17918586e-03,  3.14303837e-03,\n",
      "        1.21574884e-03,  1.08711701e-03, -6.69951492e-04,  1.60444943e-05,\n",
      "        1.36895533e-05,  1.65482561e-05,  1.90528935e-05,  1.34709480e-05,\n",
      "        1.51048553e-05,  1.56403257e-05,  1.49254929e-05,  1.45474532e-05,\n",
      "        1.46456641e-05,  1.55357993e-05,  1.48532172e-05,  1.40475722e-05,\n",
      "        1.47476803e-05,  1.43741345e-05,  1.40457632e-05,  1.51996001e-05,\n",
      "        1.72380387e-05,  2.40596455e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafcc18>, <tf.Tensor: id=889829, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 1.53871831e-08,  1.23916550e-06, -9.64822561e-07, ...,\n",
      "        -4.78494803e-05,  1.90392766e-05,  4.87516445e-05],\n",
      "       [ 5.69777008e-08, -8.97077328e-08,  1.15954668e-09, ...,\n",
      "        -3.80029587e-06, -1.13792166e-05,  8.59251031e-06],\n",
      "       [-3.50764076e-08, -4.83991975e-08,  3.13779474e-08, ...,\n",
      "        -5.13650630e-05,  2.51243673e-05,  2.15707605e-05],\n",
      "       ...,\n",
      "       [ 2.88988531e-07,  5.02179603e-07,  1.51968766e-08, ...,\n",
      "        -1.54454810e-05,  7.91313937e-07, -3.39092168e-07],\n",
      "       [ 1.03934681e-07,  3.53464706e-08, -3.26831220e-08, ...,\n",
      "        -2.36045926e-05,  1.14495024e-05, -2.35956959e-05],\n",
      "       [ 1.18350961e-07,  1.04454195e-07,  1.51270243e-07, ...,\n",
      "         2.41279213e-05,  1.08555396e-05,  3.05648100e-05]], dtype=float32)>, <tf.Tensor: id=889830, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 2.12506461e-06,  2.08709685e-06,  2.23452025e-06, ...,\n",
      "         1.99350645e-04, -3.25409856e-05, -4.55812869e-06],\n",
      "       [ 3.02724246e-08, -4.53656185e-07, -2.14650231e-06, ...,\n",
      "        -1.77895170e-04,  4.37358212e-05, -7.49999817e-05],\n",
      "       [-5.98088718e-07, -6.60206354e-07, -1.32653417e-06, ...,\n",
      "        -1.10349865e-05,  9.56884287e-06,  3.56473743e-06],\n",
      "       ...,\n",
      "       [ 3.28325370e-07,  3.14931185e-07,  1.80266113e-07, ...,\n",
      "        -1.35126299e-06, -1.41931977e-07, -2.39243559e-06],\n",
      "       [ 5.45684179e-06,  8.34015918e-06,  1.41718801e-05, ...,\n",
      "         1.16327770e-04, -2.70968885e-05,  6.73154273e-06],\n",
      "       [-9.65281470e-06, -2.21717855e-05,  1.42832789e-06, ...,\n",
      "         2.18631714e-04, -3.50626287e-05,  1.36986328e-05]], dtype=float32)>, <tf.Tensor: id=889831, shape=(6144,), dtype=float32, numpy=\n",
      "array([-7.26261351e-06, -7.19628133e-07,  6.07113543e-06, ...,\n",
      "       -8.78676474e-07,  1.18361946e-04,  2.17646390e-04], dtype=float32)>, <tf.Tensor: id=889734, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.1558368e-06, -8.2640297e-05, -1.2028526e-06, ...,\n",
      "        -1.3025870e-06, -1.7270879e-06, -2.5024387e-06],\n",
      "       [-1.7934242e-06, -1.5384342e-04, -2.0787784e-06, ...,\n",
      "        -1.9738548e-06, -2.9775581e-06, -4.3850982e-06],\n",
      "       [-7.0074123e-07, -1.2931843e-05, -7.9580218e-07, ...,\n",
      "        -8.9639155e-07, -9.3611976e-07, -1.5357763e-06],\n",
      "       ...,\n",
      "       [ 1.1415537e-05,  1.2223317e-03,  9.8765886e-06, ...,\n",
      "         1.2571992e-05,  1.4557720e-05,  2.0299094e-05],\n",
      "       [-1.1937956e-06, -1.3575741e-04, -1.1371997e-06, ...,\n",
      "        -1.3416560e-06, -1.6000424e-06, -2.3503612e-06],\n",
      "       [-4.1186718e-07, -3.2102672e-04, -1.1968859e-07, ...,\n",
      "        -4.6998261e-07, -3.0225038e-08, -8.3705856e-08]], dtype=float32)>, <tf.Tensor: id=889726, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.20426994e-05,  1.29527156e-03,  1.03359553e-05, -6.73127652e-04,\n",
      "       -3.08228866e-03,  3.02977533e-05,  6.63060171e-04,  1.49968128e-05,\n",
      "        1.38977603e-05,  1.50558089e-05,  7.56801746e-04, -2.62776506e-04,\n",
      "       -1.18260497e-04,  2.05108008e-05,  1.64527246e-05,  2.13753548e-03,\n",
      "        1.99655842e-04, -3.09145777e-03, -2.66969233e-04,  7.25722173e-04,\n",
      "       -1.28471013e-03,  1.31981302e-04, -2.85503746e-04,  4.92945837e-04,\n",
      "       -5.48825948e-04, -1.01688136e-04, -1.84313336e-04,  1.00922960e-04,\n",
      "       -3.14298552e-04,  8.06173921e-05,  8.09673948e-05,  1.74803827e-05,\n",
      "        1.70834592e-05,  1.97267618e-05, -4.45540572e-05,  9.73007525e-04,\n",
      "       -4.58649214e-04,  7.69311955e-05,  1.70783198e-04,  1.49015934e-04,\n",
      "       -2.26977514e-04, -5.92615921e-04,  6.76705735e-04,  4.15823219e-04,\n",
      "       -1.01416750e-04, -7.13035188e-05, -1.13061578e-04,  2.44346214e-04,\n",
      "        3.65089683e-04,  6.95957278e-05,  2.37339409e-04,  5.33379207e-05,\n",
      "        2.40871639e-04, -3.48904927e-04, -8.97999562e-04,  2.24316012e-04,\n",
      "       -3.26111658e-05,  4.82496398e-05,  6.86210260e-05, -1.35875365e-04,\n",
      "        1.76025096e-05,  3.95267234e-05,  1.75829373e-05,  2.98213290e-05,\n",
      "        2.24219566e-05,  2.54164473e-03,  6.66552573e-04,  2.17109511e-04,\n",
      "       -1.85244670e-03, -2.00598314e-03,  2.99500627e-03, -2.09824706e-04,\n",
      "        3.27926059e-03,  5.26090525e-03,  3.67640750e-04,  7.81190756e-04,\n",
      "        9.17625497e-04,  1.66355050e-03, -5.02654468e-04, -2.12339195e-03,\n",
      "        1.73650711e-04, -7.50491454e-04, -2.69299629e-03, -2.14353460e-03,\n",
      "       -8.52190773e-04, -3.38791381e-03, -2.21588649e-03,  2.11170805e-03,\n",
      "       -6.12952688e-04,  7.16908107e-05,  2.86425726e-04,  1.38171417e-05,\n",
      "        1.13788265e-05,  1.49145772e-05,  1.69153500e-05,  1.14930790e-05,\n",
      "        1.30584931e-05,  1.44232472e-05,  1.29890950e-05,  1.25285496e-05,\n",
      "        1.26690602e-05,  1.40310376e-05,  1.29087548e-05,  1.15358771e-05,\n",
      "        1.28657648e-05,  1.27290868e-05,  1.22286292e-05,  1.32422801e-05,\n",
      "        1.52144366e-05,  2.11945098e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafcd68>, <tf.Tensor: id=890190, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-2.0587868e-06, -2.0401451e-06,  4.8741060e-07, ...,\n",
      "        -2.0270336e-05, -1.0635642e-05,  8.8183158e-05],\n",
      "       [ 2.9233271e-07,  6.2959181e-08,  2.2286509e-08, ...,\n",
      "        -3.6647860e-05,  1.8706243e-05,  1.1478850e-05],\n",
      "       [ 1.2598454e-07, -2.1742410e-07,  3.3266767e-07, ...,\n",
      "        -2.7785327e-05,  4.2842785e-05,  1.4021356e-05],\n",
      "       ...,\n",
      "       [ 1.3090594e-06,  5.0803288e-07, -2.5560919e-07, ...,\n",
      "        -5.5003227e-07, -9.1283910e-06,  9.8801297e-07],\n",
      "       [ 2.4194915e-08, -1.0881577e-10,  2.3093671e-08, ...,\n",
      "         1.5066813e-05, -3.2321932e-05, -5.8639489e-06],\n",
      "       [ 6.3668961e-07,  5.5738944e-08, -1.8547875e-07, ...,\n",
      "         1.2037525e-05, -1.9704465e-05, -1.8434329e-06]], dtype=float32)>, <tf.Tensor: id=890191, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-4.0564032e-07, -7.0323026e-06,  4.4898270e-06, ...,\n",
      "         3.1194527e-04, -5.2639774e-05,  3.9424172e-05],\n",
      "       [-5.6864826e-07, -1.5724646e-06, -1.8046338e-06, ...,\n",
      "        -2.4334765e-04,  4.6069319e-05, -3.6001773e-05],\n",
      "       [-3.3267034e-07, -1.0490430e-06,  2.1826936e-06, ...,\n",
      "         5.0390954e-04, -5.9456655e-05,  2.5850341e-05],\n",
      "       ...,\n",
      "       [ 4.8682637e-07,  6.9430735e-07, -1.0055575e-06, ...,\n",
      "        -6.0107977e-06, -3.8835574e-07,  8.0943659e-07],\n",
      "       [ 1.0424411e-05,  1.9055462e-05,  2.7463011e-05, ...,\n",
      "        -3.2995976e-04,  3.4181452e-05, -2.4211364e-05],\n",
      "       [-3.4343159e-06, -8.0417067e-06, -3.8033641e-06, ...,\n",
      "        -6.9093257e-05, -1.5715257e-05,  5.9991904e-05]], dtype=float32)>, <tf.Tensor: id=890192, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 1.7159748e-05,  4.1287435e-06,  2.4245685e-06, ...,\n",
      "       -6.8731583e-06, -3.3114606e-04, -6.9487040e-05], dtype=float32)>, <tf.Tensor: id=890095, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.2618127e-06,  7.4986674e-05, -1.2778297e-06, ...,\n",
      "        -1.4074886e-06, -1.8419425e-06, -2.6803605e-06],\n",
      "       [-1.8603541e-06,  6.0410137e-05, -2.1092337e-06, ...,\n",
      "        -1.9753168e-06, -3.0469207e-06, -4.4946223e-06],\n",
      "       [-6.8960730e-07, -2.3706718e-05, -8.0240909e-07, ...,\n",
      "        -9.3362848e-07, -9.7325415e-07, -1.5200576e-06],\n",
      "       ...,\n",
      "       [ 1.2707595e-05, -8.7178499e-04,  1.0665541e-05, ...,\n",
      "         1.3705148e-05,  1.5821088e-05,  2.2261249e-05],\n",
      "       [-1.2532764e-06,  6.5993234e-05, -1.1618163e-06, ...,\n",
      "        -1.3763697e-06, -1.6416437e-06, -2.4319165e-06],\n",
      "       [-8.5260149e-07,  3.2539468e-04, -3.7494397e-07, ...,\n",
      "        -9.2159769e-07, -5.0828999e-07, -8.0903595e-07]], dtype=float32)>, <tf.Tensor: id=890087, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.33376898e-05, -8.76029371e-04,  1.10979117e-05, -8.37125233e-04,\n",
      "       -8.65070429e-03,  3.27006383e-05, -4.91109444e-04,  1.66513473e-05,\n",
      "        1.52722841e-05,  1.63506793e-05,  3.11226584e-04, -7.70243059e-05,\n",
      "        4.74363369e-05,  2.19888643e-05,  1.82672793e-05,  2.11447920e-03,\n",
      "       -6.79048477e-04, -5.49944816e-04,  4.87993166e-05, -1.14756593e-04,\n",
      "        4.38504561e-04, -8.25325376e-04,  3.15858255e-04, -2.45437725e-04,\n",
      "       -1.57255170e-04,  1.61091608e-04, -4.24653233e-04,  1.71891043e-06,\n",
      "       -8.66246701e-04,  9.34186246e-05, -2.39845685e-04,  1.93392334e-05,\n",
      "        1.83854008e-05,  2.13743879e-05,  2.64932489e-04,  7.43951416e-04,\n",
      "       -8.21983049e-05, -2.02129508e-04,  1.77815178e-04,  4.68545186e-05,\n",
      "       -4.31233675e-05,  2.19428315e-04,  5.82730398e-04,  7.95064378e-04,\n",
      "       -8.87722781e-05, -5.32099075e-05, -8.29794008e-05,  2.64794333e-04,\n",
      "        2.29604018e-04,  4.65997146e-04, -2.09427235e-04, -2.54376413e-04,\n",
      "       -2.04801559e-04, -4.40870033e-04, -5.49325268e-05, -7.10056484e-05,\n",
      "        1.40456556e-04,  2.51764839e-04,  7.55629735e-05,  1.93641972e-04,\n",
      "       -1.37401948e-04, -1.04199222e-04, -1.27903302e-04,  3.28766109e-05,\n",
      "        2.43528048e-05,  7.64631014e-03,  1.43903820e-03,  1.80624542e-03,\n",
      "       -3.59132839e-03,  2.02800520e-03,  2.03474727e-03, -3.13195109e-04,\n",
      "        2.72355927e-03,  2.32109544e-03, -7.31570399e-05,  1.90045437e-04,\n",
      "       -3.58805817e-04, -6.01773267e-04,  2.77104811e-03,  1.08635647e-03,\n",
      "       -1.91821856e-03, -6.50753500e-05,  6.46079704e-03, -8.22904333e-03,\n",
      "       -4.71279258e-03, -2.39007175e-04, -1.40913192e-03, -4.90801351e-04,\n",
      "        5.36119391e-04, -5.18978806e-04,  1.38279094e-04,  1.49678763e-05,\n",
      "        1.22833126e-05,  1.64278208e-05,  1.87234837e-05,  1.27311096e-05,\n",
      "        1.43519983e-05,  1.60256623e-05,  1.39574222e-05,  1.38550695e-05,\n",
      "        1.37044344e-05,  1.52510147e-05,  1.38649966e-05,  1.27040885e-05,\n",
      "        1.43120051e-05,  1.40047332e-05,  1.31861880e-05,  1.43635025e-05,\n",
      "        1.64769335e-05,  2.31779304e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdae7898>, <tf.Tensor: id=890551, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-2.9394448e-07, -1.2533843e-06, -5.4072177e-07, ...,\n",
      "         1.4597919e-05, -2.7152990e-05,  3.9875926e-05],\n",
      "       [ 1.8502514e-07,  2.0617763e-07, -2.0713546e-07, ...,\n",
      "         2.2258717e-05, -1.7916900e-05,  1.2663344e-05],\n",
      "       [-3.4814781e-07, -1.5259917e-07,  2.1511902e-07, ...,\n",
      "         1.8524894e-05, -2.3308348e-05,  2.4577605e-05],\n",
      "       ...,\n",
      "       [ 5.8447893e-07,  7.8106822e-08, -9.2690044e-08, ...,\n",
      "        -6.8416707e-06, -1.1421561e-05, -6.3827956e-06],\n",
      "       [ 8.9703143e-08,  2.9102628e-08, -2.0730624e-08, ...,\n",
      "         5.1835006e-05,  4.3402397e-05,  3.5172558e-05],\n",
      "       [ 3.9965386e-07, -6.4765509e-08,  1.8674740e-07, ...,\n",
      "         2.6194220e-05, -5.5638779e-05,  3.4736445e-08]], dtype=float32)>, <tf.Tensor: id=890552, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 5.3158101e-06,  3.1839227e-06,  4.7547337e-06, ...,\n",
      "        -3.9783068e-04,  4.1598316e-05, -1.0371586e-04],\n",
      "       [-7.0090277e-07, -7.1687293e-07,  2.0256316e-06, ...,\n",
      "         5.1878527e-04, -6.7095796e-05,  5.7563513e-05],\n",
      "       [-1.9771369e-07,  3.8455400e-07, -1.2338319e-06, ...,\n",
      "        -1.9247067e-04,  2.1979498e-05,  2.7477979e-06],\n",
      "       ...,\n",
      "       [ 5.1757769e-07,  5.6030115e-07,  1.5657853e-06, ...,\n",
      "         3.7316440e-05, -4.2164202e-06,  3.7036475e-06],\n",
      "       [-3.7688676e-07, -5.1450729e-06, -2.5494248e-06, ...,\n",
      "         4.6812274e-04, -7.1396476e-05,  7.3856958e-05],\n",
      "       [-7.4116820e-06, -2.1486574e-05,  6.2035424e-06, ...,\n",
      "         1.7206128e-04, -4.3131884e-05,  7.4867014e-05]], dtype=float32)>, <tf.Tensor: id=890553, shape=(6144,), dtype=float32, numpy=\n",
      "array([-1.9808498e-05,  7.9887641e-06,  1.6366073e-06, ...,\n",
      "        3.7259946e-05,  4.7284018e-04,  1.7037506e-04], dtype=float32)>, <tf.Tensor: id=890456, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.14965269e-06, -2.49328696e-05, -1.24145038e-06, ...,\n",
      "        -1.33389369e-06, -1.75804075e-06, -2.56361318e-06],\n",
      "       [-1.91267009e-06, -9.61467595e-05, -2.20809443e-06, ...,\n",
      "        -2.13166595e-06, -3.15573061e-06, -4.56535099e-06],\n",
      "       [-5.72037266e-07,  7.07032450e-05, -7.30343459e-07, ...,\n",
      "        -7.67346478e-07, -8.45171030e-07, -1.45458341e-06],\n",
      "       ...,\n",
      "       [ 9.86079976e-06,  6.39139849e-04,  9.09886967e-06, ...,\n",
      "         1.09694392e-05,  1.33139365e-05,  1.87100668e-05],\n",
      "       [-9.99273084e-07, -6.04795314e-05, -1.01010744e-06, ...,\n",
      "        -1.13781664e-06, -1.41136138e-06, -2.10922713e-06],\n",
      "       [-2.93224609e-08, -1.87214071e-04,  1.75770097e-07, ...,\n",
      "        -5.11784037e-09,  3.43255351e-07,  4.13790417e-07]], dtype=float32)>, <tf.Tensor: id=890448, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.02702015e-05,  6.85388979e-04,  9.39908841e-06,  5.46891941e-04,\n",
      "        1.90414127e-03,  2.95032805e-05, -2.42390510e-04,  1.35115542e-05,\n",
      "        1.19718770e-05,  1.40569928e-05,  4.92206251e-04, -2.34817577e-04,\n",
      "       -1.51213433e-04,  1.89631846e-05,  1.46282509e-05,  1.89595507e-03,\n",
      "        9.51256487e-04, -1.21698529e-03, -1.12321439e-04,  3.55733238e-04,\n",
      "        4.03932761e-04,  3.03191628e-04,  3.13888188e-04, -2.65425333e-04,\n",
      "       -1.33787098e-05,  1.35648035e-04,  1.53622110e-04,  1.71329055e-04,\n",
      "       -5.41052723e-04,  1.10573237e-04, -6.78789074e-05,  1.53713154e-05,\n",
      "        1.53354558e-05,  1.75010573e-05,  2.58547690e-04,  5.57979394e-04,\n",
      "       -2.00098206e-04, -3.90391651e-04,  3.44215659e-04,  1.45990169e-04,\n",
      "       -1.84750039e-04, -2.43169430e-04,  1.98744910e-05, -1.16704928e-03,\n",
      "        2.17028806e-04,  2.55760155e-04,  7.15182032e-05,  1.87153695e-04,\n",
      "        4.93724510e-05, -1.02999053e-04,  6.41689840e-05,  5.33565326e-05,\n",
      "        5.26338350e-04, -8.88879295e-05,  8.44017952e-04, -7.06463834e-05,\n",
      "       -2.72363104e-05,  3.12936696e-04, -9.05462075e-05,  2.34338295e-05,\n",
      "        1.67614253e-05,  2.03086573e-04,  1.76010435e-04,  2.58225427e-05,\n",
      "        2.10417784e-05, -3.36793577e-03, -7.59280869e-04,  4.40294202e-03,\n",
      "       -1.04675116e-03,  8.20927974e-03,  1.23343329e-04, -2.38302653e-03,\n",
      "        3.31674959e-03,  2.70386809e-03,  1.15138319e-04, -1.24126778e-03,\n",
      "        5.77541417e-04,  1.54360954e-03, -5.82827860e-03, -1.04684141e-02,\n",
      "       -2.32522935e-03, -4.53614135e-04,  5.81058022e-03,  5.64204343e-03,\n",
      "       -8.15005042e-03,  9.46059707e-04, -1.51805452e-03, -3.05341207e-03,\n",
      "       -1.02525402e-03,  4.81150491e-05,  3.97196040e-04,  1.22290739e-05,\n",
      "        1.04503451e-05,  1.30194658e-05,  1.39399826e-05,  1.00007182e-05,\n",
      "        1.16611855e-05,  1.16047959e-05,  1.10801429e-05,  1.08319318e-05,\n",
      "        1.14447657e-05,  1.18620501e-05,  1.12784337e-05,  1.00598809e-05,\n",
      "        1.10096180e-05,  1.09354423e-05,  1.07708875e-05,  1.13927154e-05,\n",
      "        1.37637553e-05,  1.93234664e-05], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdae67f0>, <tf.Tensor: id=890912, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 9.3815788e-07,  4.4342707e-07,  4.0694026e-08, ...,\n",
      "         6.6178122e-06, -6.6279092e-08,  1.8868455e-05],\n",
      "       [ 2.2335905e-07,  1.2740527e-07, -2.7431086e-09, ...,\n",
      "        -3.3592012e-05, -2.1295617e-05,  1.7993127e-05],\n",
      "       [ 1.1920438e-08,  2.9128913e-08,  9.7974279e-08, ...,\n",
      "        -2.5083416e-05,  1.5008587e-05,  1.8556193e-05],\n",
      "       ...,\n",
      "       [ 1.0266696e-07,  2.1397723e-07, -7.7168487e-08, ...,\n",
      "         3.0482549e-06, -1.5211483e-05, -7.2969742e-06],\n",
      "       [ 8.3346183e-09,  1.8052425e-08, -1.8987951e-09, ...,\n",
      "        -4.2425854e-05,  9.2434002e-06, -3.0452407e-05],\n",
      "       [ 4.4022411e-07,  8.7269186e-08,  1.3780405e-07, ...,\n",
      "         4.7237059e-05, -8.5378606e-06, -9.5124487e-06]], dtype=float32)>, <tf.Tensor: id=890913, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 2.2410745e-06,  6.8479517e-06,  9.3553526e-06, ...,\n",
      "        -3.9434497e-04,  3.8413509e-05, -1.8396664e-05],\n",
      "       [-4.4800510e-07, -4.8741464e-07, -6.4401985e-07, ...,\n",
      "         2.9447529e-04, -5.0970615e-05, -1.5549389e-05],\n",
      "       [ 4.1894191e-07,  8.7663057e-07,  1.3808076e-06, ...,\n",
      "         2.8425922e-05, -8.7766011e-06, -3.9194325e-05],\n",
      "       ...,\n",
      "       [ 3.4161737e-08,  5.3993449e-07,  5.9488315e-07, ...,\n",
      "         2.1234751e-05, -2.0475352e-06,  1.4027030e-06],\n",
      "       [ 3.1841503e-06,  1.2881453e-06,  5.1961456e-06, ...,\n",
      "         4.4150773e-04, -7.5706928e-05,  2.4571429e-05],\n",
      "       [ 6.6016514e-06,  8.1112057e-06, -1.5513120e-05, ...,\n",
      "         1.2172511e-04, -1.7568760e-05,  7.1415134e-06]], dtype=float32)>, <tf.Tensor: id=890914, shape=(6144,), dtype=float32, numpy=\n",
      "array([-1.2164440e-05,  2.2203794e-06, -3.2658638e-06, ...,\n",
      "        2.1685832e-05,  4.4465929e-04,  1.2288909e-04], dtype=float32)>, <tf.Tensor: id=890817, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.18848175e-06,  5.62107198e-05, -1.27108524e-06, ...,\n",
      "        -1.37237669e-06, -1.81784958e-06, -2.61596961e-06],\n",
      "       [-2.03404488e-06,  1.18026837e-05, -2.30045680e-06, ...,\n",
      "        -2.25561826e-06, -3.32956051e-06, -4.70408031e-06],\n",
      "       [-6.69416067e-07, -1.02737358e-05, -7.95583674e-07, ...,\n",
      "        -9.09180130e-07, -9.77607215e-07, -1.60549098e-06],\n",
      "       ...,\n",
      "       [ 1.01847163e-05, -8.99048988e-04,  9.36286051e-06, ...,\n",
      "         1.15240373e-05,  1.39757212e-05,  1.94562235e-05],\n",
      "       [-1.01369028e-06,  9.16754652e-05, -1.01977980e-06, ...,\n",
      "        -1.16222463e-06, -1.44410546e-06, -2.12144641e-06],\n",
      "       [-1.36460216e-07,  2.07984151e-04,  1.14735315e-07, ...,\n",
      "        -1.61602969e-07,  2.22761784e-07,  1.88210819e-07]], dtype=float32)>, <tf.Tensor: id=890809, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.04633718e-05, -9.51645547e-04,  9.57389148e-06,  1.27329887e-03,\n",
      "        5.33162151e-03,  3.13929631e-05, -9.74671508e-04,  1.37403840e-05,\n",
      "        1.22105967e-05,  1.42755616e-05,  9.14417760e-05, -5.10736951e-04,\n",
      "       -2.81087821e-04,  1.95160537e-05,  1.51955528e-05, -1.38211576e-03,\n",
      "       -1.94902881e-04,  1.26980897e-03,  4.63440592e-05, -3.59680096e-04,\n",
      "        9.61947953e-04,  4.72051543e-05, -1.14037146e-04,  5.97424041e-05,\n",
      "        2.77215840e-05, -1.36572344e-05, -4.28176165e-04, -1.09964858e-04,\n",
      "       -3.41702194e-04,  2.94046273e-04, -5.67776697e-05,  1.57402792e-05,\n",
      "        1.56275219e-05,  1.78446935e-05,  9.82877536e-05,  1.41290657e-04,\n",
      "        6.73368806e-04, -3.64335137e-04,  3.60894919e-04, -1.41908327e-04,\n",
      "        6.48137822e-04,  9.98071773e-05,  2.19833397e-04,  2.65084498e-04,\n",
      "       -8.89342118e-05, -1.97890200e-04, -5.34116625e-05,  3.39274993e-04,\n",
      "       -8.82227323e-05,  9.25214044e-05, -3.89830035e-04,  5.66351555e-05,\n",
      "        3.74480209e-04,  4.79843060e-04, -2.83049303e-06,  2.61197943e-04,\n",
      "        1.38103715e-04,  6.57447032e-04,  7.02231628e-05, -2.76560313e-04,\n",
      "        1.73648859e-05,  6.07194597e-05,  3.49159527e-05,  2.70317159e-05,\n",
      "        2.20008187e-05, -4.49798256e-03, -1.14962610e-03,  4.84886533e-03,\n",
      "       -1.75401525e-04,  5.79875428e-03,  1.60662457e-03, -2.66301376e-03,\n",
      "       -7.07560685e-04,  2.15664436e-03,  1.09979308e-04, -1.37151137e-03,\n",
      "       -2.19168141e-05, -1.81382999e-03,  4.48618084e-04, -9.01293568e-03,\n",
      "        4.62948857e-03,  1.06815161e-04, -3.08036245e-03,  4.64522233e-03,\n",
      "       -3.12666316e-03, -3.75714299e-05, -1.13845640e-03, -1.72500045e-03,\n",
      "        2.35322237e-04, -2.07770406e-03,  3.86169821e-04,  1.24997296e-05,\n",
      "        1.07327851e-05,  1.31279330e-05,  1.44033284e-05,  1.01932683e-05,\n",
      "        1.18953512e-05,  1.18864937e-05,  1.14253062e-05,  1.12298885e-05,\n",
      "        1.19801161e-05,  1.24041626e-05,  1.15517623e-05,  1.02076792e-05,\n",
      "        1.11648751e-05,  1.10701940e-05,  1.10647179e-05,  1.18273383e-05,\n",
      "        1.43031511e-05,  1.99125789e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc4a8>, <tf.Tensor: id=891273, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-1.44074718e-07, -2.27581026e-07, -6.35252093e-07, ...,\n",
      "        -1.18090766e-05, -2.27620421e-05,  4.13057969e-05],\n",
      "       [ 2.61577867e-07,  3.46724818e-08,  2.69088440e-09, ...,\n",
      "         4.53136545e-06, -1.76091926e-05, -2.33083501e-05],\n",
      "       [-4.00432612e-07, -1.18255507e-07,  1.67298012e-07, ...,\n",
      "        -2.04029539e-06,  3.06966976e-05,  3.63533254e-05],\n",
      "       ...,\n",
      "       [ 1.78374293e-07,  1.32343189e-07, -2.06408615e-07, ...,\n",
      "         2.26944371e-06, -6.20881519e-06,  2.31683157e-06],\n",
      "       [ 7.06236918e-08,  4.16867572e-08,  9.02286779e-09, ...,\n",
      "        -3.71969727e-05, -1.26933555e-05, -9.64343144e-06],\n",
      "       [ 5.06957463e-07,  2.55260346e-07, -7.85445025e-08, ...,\n",
      "         4.89533704e-05, -1.84691780e-05,  2.95855189e-05]], dtype=float32)>, <tf.Tensor: id=891274, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-2.4299624e-07, -2.7808774e-06,  7.9893596e-08, ...,\n",
      "         1.8005073e-04, -4.8222377e-05,  2.6860733e-05],\n",
      "       [ 2.5936458e-07,  1.1786325e-06,  1.1622781e-06, ...,\n",
      "         3.4084124e-04, -5.0771672e-05,  1.1567996e-05],\n",
      "       [ 5.5834272e-07,  1.4110308e-06,  4.9395356e-07, ...,\n",
      "        -1.7664141e-04,  3.0617342e-05, -3.0900243e-05],\n",
      "       ...,\n",
      "       [ 2.2896891e-07,  3.4659939e-07,  1.2913043e-07, ...,\n",
      "         6.5662034e-06, -1.6773753e-06, -9.6750784e-07],\n",
      "       [ 1.0920783e-06, -5.1942106e-06, -5.4037723e-06, ...,\n",
      "         3.9361883e-04, -6.4407497e-05,  4.5429195e-05],\n",
      "       [ 6.7509509e-06,  1.4769785e-05, -1.6877979e-05, ...,\n",
      "        -1.5139028e-04,  8.6967502e-06, -3.8647136e-07]], dtype=float32)>, <tf.Tensor: id=891275, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 7.9817783e-06, -1.6804609e-06, -4.0741115e-06, ...,\n",
      "        6.4474361e-06,  3.9960881e-04, -1.5260823e-04], dtype=float32)>, <tf.Tensor: id=891178, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.0309982e-06,  9.9635705e-05, -1.1438170e-06, ...,\n",
      "        -1.2129011e-06, -1.6115935e-06, -2.3534972e-06],\n",
      "       [-1.8167859e-06,  5.4788688e-06, -2.1396204e-06, ...,\n",
      "        -2.0648395e-06, -3.0405918e-06, -4.4040244e-06],\n",
      "       [-4.6483785e-07,  2.2509113e-04, -6.8102941e-07, ...,\n",
      "        -6.7288579e-07, -7.9916504e-07, -1.3421411e-06],\n",
      "       ...,\n",
      "       [ 8.8578645e-06, -1.4140132e-03,  8.2265051e-06, ...,\n",
      "         9.9893059e-06,  1.2137636e-05,  1.7125334e-05],\n",
      "       [-9.4923354e-07,  1.6034793e-04, -9.5646828e-07, ...,\n",
      "        -1.0859817e-06, -1.3412113e-06, -2.0067507e-06],\n",
      "       [-1.7011951e-07,  4.8215684e-04,  9.5779001e-08, ...,\n",
      "        -1.4523866e-07,  1.4778473e-07,  1.0933256e-07]], dtype=float32)>, <tf.Tensor: id=891170, shape=(110,), dtype=float32, numpy=\n",
      "array([ 9.14766861e-06, -1.44423172e-03,  8.43084490e-06,  8.20967252e-04,\n",
      "        7.69700436e-03,  2.77210584e-05, -3.37631500e-04,  1.22439251e-05,\n",
      "        1.07093492e-05,  1.27328249e-05,  3.40764236e-04,  9.97441530e-05,\n",
      "        2.88021547e-04,  1.72990331e-05,  1.35562850e-05,  8.79884639e-04,\n",
      "        2.04843702e-03,  2.20302865e-03, -2.70781835e-04,  1.21783058e-04,\n",
      "        4.32092987e-04, -3.26517242e-04,  3.05009627e-04,  1.65271085e-05,\n",
      "        1.37670242e-04,  4.01955011e-04,  3.08623508e-04,  3.15472746e-04,\n",
      "        2.54362298e-04,  1.18721226e-04,  9.20255043e-05,  1.37722518e-05,\n",
      "        1.38567320e-05,  1.55253456e-05, -3.92538583e-04,  5.80803666e-04,\n",
      "        1.36992137e-04,  6.73501490e-05,  4.58512164e-04,  2.89227755e-04,\n",
      "        4.23123012e-04,  2.29178040e-04,  4.49349900e-04,  1.15938741e-03,\n",
      "        4.46528793e-05,  9.76353986e-05, -2.34753694e-04,  9.45831343e-05,\n",
      "        3.07685696e-05,  2.01648960e-04,  3.42129701e-04,  5.22545961e-05,\n",
      "       -3.34553566e-04,  7.34455942e-04,  8.42136331e-04, -7.14196576e-05,\n",
      "        1.25053630e-04, -5.65745169e-04,  6.35424876e-05, -1.32974354e-04,\n",
      "        1.55690286e-05,  4.29440552e-05,  1.79995332e-04,  2.32592829e-05,\n",
      "        1.94175100e-05, -2.10508297e-04, -2.62275990e-03,  5.42058423e-03,\n",
      "       -5.21751004e-04,  2.64973450e-03,  1.61068863e-03, -7.54258246e-04,\n",
      "       -1.66521303e-03, -1.04610333e-02, -3.83676786e-04, -1.70563610e-04,\n",
      "       -3.94532550e-03, -2.85862200e-03, -3.67843150e-03, -1.25923048e-04,\n",
      "        4.01463732e-03,  3.22743697e-04, -7.02482369e-03,  5.56014245e-04,\n",
      "        4.33899136e-03, -5.39527333e-04, -2.40077358e-03, -1.24626560e-03,\n",
      "        9.25571716e-04, -7.87721830e-04, -2.75031314e-04,  1.09498105e-05,\n",
      "        9.61388741e-06,  1.15308885e-05,  1.25742463e-05,  9.02222928e-06,\n",
      "        1.04016181e-05,  1.01655805e-05,  1.00202315e-05,  9.82555503e-06,\n",
      "        1.02976055e-05,  1.06449297e-05,  9.89307591e-06,  9.05411980e-06,\n",
      "        9.68928271e-06,  9.68064887e-06,  9.64667834e-06,  1.02923677e-05,\n",
      "        1.24554699e-05,  1.75687510e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafcac8>, <tf.Tensor: id=891634, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 2.5386805e-06, -3.9712097e-07, -5.5299262e-07, ...,\n",
      "        -1.2351726e-05,  4.3884953e-05,  5.9827664e-05],\n",
      "       [ 2.0048121e-08, -4.2712458e-08,  2.0808072e-07, ...,\n",
      "        -4.3103195e-05,  1.9358016e-05, -1.6605719e-06],\n",
      "       [-5.1136681e-07, -6.2902153e-08,  1.6828386e-07, ...,\n",
      "         1.7709632e-05,  5.1006759e-06,  1.7577528e-05],\n",
      "       ...,\n",
      "       [-2.7773976e-07, -5.4349606e-08,  3.9873004e-08, ...,\n",
      "        -4.0357568e-06, -9.4085162e-06, -6.7976148e-06],\n",
      "       [ 2.2020380e-08,  2.4905844e-08,  2.7547371e-09, ...,\n",
      "        -4.2429288e-05, -2.6377704e-05,  7.2142948e-06],\n",
      "       [-2.7169786e-08,  7.7979415e-08,  2.6955681e-07, ...,\n",
      "         2.5925763e-05,  1.7650053e-05,  2.1653277e-05]], dtype=float32)>, <tf.Tensor: id=891635, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 1.28514148e-06, -5.61757633e-06,  4.81891720e-06, ...,\n",
      "         5.31345722e-04, -7.53019922e-05,  7.97775137e-05],\n",
      "       [-2.47939710e-07, -4.60273327e-07,  2.45857962e-07, ...,\n",
      "        -1.19692595e-04,  1.56282276e-05, -2.24100240e-05],\n",
      "       [-1.51548875e-07, -6.97143719e-07, -2.44685481e-08, ...,\n",
      "         5.05105272e-05, -1.29327464e-05,  3.87514519e-05],\n",
      "       ...,\n",
      "       [-5.92113736e-09, -2.17694378e-07,  4.73491724e-07, ...,\n",
      "         2.51405727e-05, -2.94752772e-06,  2.24738005e-06],\n",
      "       [ 1.05683000e-07, -5.13439863e-06, -2.99678823e-06, ...,\n",
      "         2.61848414e-04, -3.00513129e-05, -7.16348495e-07],\n",
      "       [ 7.16580053e-06,  1.21855528e-05, -2.10688468e-05, ...,\n",
      "        -1.35720897e-04,  2.39489855e-05,  1.05067775e-05]], dtype=float32)>, <tf.Tensor: id=891636, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 9.9676272e-06, -2.9612183e-06,  8.6858995e-07, ...,\n",
      "        2.5736856e-05,  2.6345521e-04, -1.4033334e-04], dtype=float32)>, <tf.Tensor: id=891539, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.7465875e-07, -5.2391417e-05, -1.0808074e-06, ...,\n",
      "        -1.1329155e-06, -1.5179535e-06, -2.1815524e-06],\n",
      "       [-1.6910802e-06,  1.1702723e-05, -2.0224286e-06, ...,\n",
      "        -1.9184881e-06, -2.8808829e-06, -4.1104577e-06],\n",
      "       [-4.2806465e-07, -3.7171075e-04, -6.3872221e-07, ...,\n",
      "        -6.3266873e-07, -7.4332439e-07, -1.2488266e-06],\n",
      "       ...,\n",
      "       [ 8.6783784e-06, -1.9339722e-04,  8.0500376e-06, ...,\n",
      "         9.6714211e-06,  1.1779972e-05,  1.6259854e-05],\n",
      "       [-9.1253293e-07,  3.3969736e-05, -9.1910204e-07, ...,\n",
      "        -1.0281308e-06, -1.2797115e-06, -1.8717458e-06],\n",
      "       [-1.3992408e-07, -1.5830949e-04,  6.6393810e-08, ...,\n",
      "        -1.4154301e-07,  1.5273022e-07,  1.5546493e-07]], dtype=float32)>, <tf.Tensor: id=891531, shape=(110,), dtype=float32, numpy=\n",
      "array([ 9.03074761e-06, -2.74720922e-04,  8.29847977e-06,  1.22356368e-03,\n",
      "        1.17053930e-03,  2.65160015e-05, -2.60373781e-04,  1.19543911e-05,\n",
      "        1.06075759e-05, -3.00103799e-04,  3.42377229e-04, -5.25292824e-04,\n",
      "       -6.07005932e-05, -9.20921506e-04,  1.32232826e-05,  2.65721115e-04,\n",
      "       -7.48843653e-04,  1.77108322e-03,  4.06789877e-05, -8.31100988e-05,\n",
      "        3.88885848e-04,  2.78348103e-04,  4.12638910e-04, -4.30252749e-06,\n",
      "       -1.83972836e-04,  2.05186734e-04,  1.32913658e-04,  1.41874203e-04,\n",
      "        1.06376006e-04, -2.34534906e-04,  6.80417870e-05,  1.31448678e-05,\n",
      "        1.34507472e-05,  1.48764957e-05,  6.29957503e-05, -3.67076107e-04,\n",
      "        5.68575109e-04,  5.24350733e-04,  4.13927919e-04,  1.03330582e-04,\n",
      "       -9.23400949e-05, -5.63571812e-04,  5.34126768e-04,  1.22010557e-03,\n",
      "       -2.80110922e-04, -6.86598796e-05, -7.08388456e-04,  3.71675778e-05,\n",
      "        1.69010455e-04,  3.23763699e-04, -1.66309284e-04,  5.01150425e-05,\n",
      "       -6.88260654e-04, -2.57236446e-04,  8.11619160e-04,  7.65300429e-05,\n",
      "       -1.91643281e-04, -3.20831823e-05,  6.22345033e-05,  2.14038064e-05,\n",
      "       -2.97653576e-04,  1.89075217e-04,  1.74313143e-04,  2.24460364e-05,\n",
      "        1.91128129e-05,  2.25160061e-03, -2.48169689e-03,  3.89463757e-03,\n",
      "       -5.14675584e-03, -6.16784859e-03,  1.33668911e-03,  5.59247215e-04,\n",
      "       -3.07030918e-04, -4.40953812e-03,  6.84577390e-04, -1.65317231e-03,\n",
      "        1.33642135e-03, -9.29070171e-04, -8.95302452e-04,  5.80817880e-03,\n",
      "        7.99070951e-03,  4.91377432e-05, -2.94199982e-03,  5.91192860e-03,\n",
      "        8.52418365e-04, -4.31175390e-03, -3.37720080e-03, -2.37136940e-03,\n",
      "        8.72249773e-04, -1.78632129e-03,  6.02716871e-04, -1.45357713e-04,\n",
      "       -1.46899256e-04,  1.14141803e-05,  1.20010018e-05,  8.75513979e-06,\n",
      "        1.03963230e-05,  9.86359646e-06,  9.73603164e-06,  9.45506417e-06,\n",
      "        9.79124525e-06,  1.02048425e-05,  9.71000190e-06,  8.87431088e-06,\n",
      "        9.44775911e-06,  9.52739174e-06,  9.35779190e-06,  1.00354018e-05,\n",
      "        1.21531020e-05,  1.67606140e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc3c8>, <tf.Tensor: id=891995, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 1.1396944e-06,  3.3895375e-07, -4.6117191e-07, ...,\n",
      "         2.1925105e-05, -4.4121143e-06,  6.5243657e-05],\n",
      "       [-2.3624189e-07,  1.6814361e-07,  2.4908380e-07, ...,\n",
      "        -1.2962654e-05,  1.5690644e-05,  2.8849636e-06],\n",
      "       [-2.8214631e-08, -7.8425177e-09,  4.6517666e-08, ...,\n",
      "        -4.5761713e-05,  2.2135386e-05, -8.7195822e-06],\n",
      "       ...,\n",
      "       [-9.3159525e-09,  1.5377964e-07, -1.3934515e-07, ...,\n",
      "         6.2701470e-06, -5.5176552e-06,  5.0938420e-06],\n",
      "       [ 6.4412204e-09,  2.5876471e-08, -2.1233513e-08, ...,\n",
      "        -9.5522046e-05,  9.8569581e-06, -6.8636451e-05],\n",
      "       [-8.8136176e-08, -1.2507317e-07,  2.0208387e-07, ...,\n",
      "         5.5609785e-06, -2.2831340e-05, -1.3247743e-05]], dtype=float32)>, <tf.Tensor: id=891996, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 2.1482804e-06,  4.5186587e-07,  1.0604819e-05, ...,\n",
      "         6.2316313e-04, -9.5406816e-05,  3.6499812e-05],\n",
      "       [-7.0359471e-07, -2.0183911e-06, -1.7511447e-06, ...,\n",
      "        -4.6405935e-04,  7.3499876e-05, -7.3821002e-06],\n",
      "       [-9.8540966e-07, -4.5049899e-07, -1.6475000e-06, ...,\n",
      "         6.5723597e-04, -9.5267314e-05,  5.1650829e-08],\n",
      "       ...,\n",
      "       [ 4.8144881e-07,  9.2566813e-07, -6.1902944e-07, ...,\n",
      "        -2.9325847e-05,  3.6929644e-06, -2.5932741e-06],\n",
      "       [ 1.0990777e-05,  1.5302456e-05,  2.9775387e-05, ...,\n",
      "        -4.3536001e-04,  6.8332702e-05, -7.4468662e-05],\n",
      "       [ 9.0044468e-06,  1.7101791e-05, -2.0016827e-05, ...,\n",
      "         3.6260426e-05,  4.6911928e-06,  1.3161511e-05]], dtype=float32)>, <tf.Tensor: id=891997, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 8.2138058e-08, -4.4169326e-07,  6.9693037e-06, ...,\n",
      "       -2.9795621e-05, -4.3943638e-04,  3.4744637e-05], dtype=float32)>, <tf.Tensor: id=891900, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.6367421e-07, -4.5317309e-05, -1.0502839e-06, ...,\n",
      "        -1.1147833e-06, -1.4924177e-06, -2.1135515e-06],\n",
      "       [-1.6598331e-06, -2.1834487e-05, -1.9235888e-06, ...,\n",
      "        -1.8435192e-06, -2.7584283e-06, -3.8785938e-06],\n",
      "       [-5.3204155e-07, -9.7324701e-06, -6.6962775e-07, ...,\n",
      "        -7.4331342e-07, -7.9679262e-07, -1.3383433e-06],\n",
      "       ...,\n",
      "       [ 8.2505048e-06,  4.9396831e-04,  7.6027491e-06, ...,\n",
      "         9.1672118e-06,  1.1209071e-05,  1.5467975e-05],\n",
      "       [-8.7820092e-07, -4.1540919e-05, -8.7175124e-07, ...,\n",
      "        -9.9151305e-07, -1.2299182e-06, -1.7865538e-06],\n",
      "       [-1.5433662e-07, -6.1594910e-06,  5.9583058e-08, ...,\n",
      "        -1.7227308e-07,  1.5414564e-07,  7.0113401e-08]], dtype=float32)>, <tf.Tensor: id=891892, shape=(110,), dtype=float32, numpy=\n",
      "array([ 8.49437311e-06,  4.95894463e-04,  7.77877722e-06, -2.88419469e-05,\n",
      "       -7.31916074e-03, -1.30907050e-04, -2.14630133e-03,  1.09084804e-05,\n",
      "        1.00655388e-05,  1.19737497e-05, -1.43381511e-03,  1.07988402e-04,\n",
      "       -9.46010186e-05,  1.62685446e-05,  1.24981698e-05, -4.27467935e-03,\n",
      "        3.34798067e-04,  7.45549798e-04,  3.97081676e-05,  2.46255113e-05,\n",
      "        6.57483586e-04, -5.04678872e-04, -3.28422844e-04,  1.86070829e-05,\n",
      "       -1.34371861e-04,  3.91728448e-04,  1.70214858e-04, -1.44805963e-04,\n",
      "       -2.59385619e-04,  3.23821150e-05,  3.56881828e-05,  1.23681293e-05,\n",
      "        1.25886445e-05,  1.42440822e-05, -5.79789456e-04,  7.61797346e-05,\n",
      "        2.11176943e-04, -2.76345323e-04, -4.08302498e-04,  2.44132854e-04,\n",
      "        2.86688010e-05,  2.14679720e-04,  1.84874967e-04, -6.64589519e-04,\n",
      "        1.80822375e-04, -7.31782129e-05, -8.16200700e-05, -3.22530628e-04,\n",
      "       -1.53207526e-04,  1.37758434e-07,  1.16335767e-04,  4.74269364e-05,\n",
      "        3.79271980e-04, -1.17123411e-04, -4.40217555e-04,  2.20006477e-04,\n",
      "       -3.86063621e-05, -5.33786253e-04,  6.01257780e-05,  2.47934804e-05,\n",
      "        1.40810107e-05,  2.29803445e-05,  1.30645130e-05,  2.16312255e-05,\n",
      "        1.82924814e-05,  8.89579020e-03, -1.62308861e-05,  1.84936402e-03,\n",
      "        7.03703845e-04, -5.97951934e-03,  5.10172511e-04,  7.91659229e-04,\n",
      "       -4.01791371e-03, -1.97486719e-03,  1.66227386e-04, -1.54085027e-03,\n",
      "        2.70296214e-03, -2.92660110e-03, -1.69515610e-04,  1.13112293e-02,\n",
      "        6.17062673e-04,  6.76863303e-04,  2.17874767e-03, -1.54135539e-03,\n",
      "        6.99719321e-03, -2.53274315e-03, -1.80198374e-04, -1.01848156e-03,\n",
      "        3.21526895e-05, -2.15057138e-04, -2.67432188e-04,  1.03322300e-05,\n",
      "        8.75147180e-06,  1.08848008e-05,  1.13038623e-05,  8.20985315e-06,\n",
      "        9.60800662e-06,  9.49189962e-06,  8.95782432e-06,  8.83349185e-06,\n",
      "        9.20336151e-06,  9.79390461e-06,  9.15198962e-06,  8.32998467e-06,\n",
      "        8.81703636e-06,  9.08932634e-06,  8.86309408e-06,  9.42255792e-06,\n",
      "        1.14800778e-05,  1.58430157e-05], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafcc88>, <tf.Tensor: id=892356, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 2.0854436e-06, -2.3722937e-06,  2.9363164e-07, ...,\n",
      "         5.6466510e-05, -9.7464246e-05,  1.5174502e-05],\n",
      "       [ 5.0778078e-07,  1.0552343e-07, -1.1320821e-07, ...,\n",
      "        -2.5428411e-05,  2.0415544e-05,  1.0145763e-05],\n",
      "       [-1.8163225e-07, -3.8913772e-08, -8.4669779e-08, ...,\n",
      "        -4.7610902e-05, -5.4320171e-06, -3.7915968e-05],\n",
      "       ...,\n",
      "       [ 7.8393549e-07,  3.3941316e-07, -3.2225799e-07, ...,\n",
      "         3.5870144e-07, -7.3804213e-06, -8.0287473e-06],\n",
      "       [ 3.4942484e-08,  6.1075085e-08, -6.0165988e-08, ...,\n",
      "        -1.4957894e-05,  1.6385178e-05,  1.3516315e-05],\n",
      "       [-5.1735910e-08, -1.7358937e-07,  9.2975640e-08, ...,\n",
      "        -2.1987780e-05,  1.3830903e-05,  1.1206741e-05]], dtype=float32)>, <tf.Tensor: id=892357, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-7.9040892e-06, -2.1869140e-05, -8.0197378e-06, ...,\n",
      "         3.1857792e-04, -4.9763610e-05, -1.8088753e-05],\n",
      "       [ 2.6248790e-07, -1.2529524e-07,  1.1413463e-06, ...,\n",
      "         1.7597309e-05,  3.1272968e-06, -5.2880787e-06],\n",
      "       [-6.9731328e-08, -1.4550159e-06,  3.0430549e-06, ...,\n",
      "         2.2912736e-04, -1.9160792e-05,  5.9820733e-05],\n",
      "       ...,\n",
      "       [ 1.1045032e-06,  1.7675867e-06, -2.7742158e-07, ...,\n",
      "        -1.9547504e-05,  1.4548907e-06, -2.9283572e-06],\n",
      "       [ 6.7267038e-06,  7.5360408e-06,  1.0874369e-05, ...,\n",
      "        -1.0718028e-05, -5.4599227e-06, -4.2534451e-05],\n",
      "       [ 5.2343626e-06,  4.3207110e-06, -1.3786041e-05, ...,\n",
      "        -1.1528285e-04,  3.5266778e-05,  6.3805943e-05]], dtype=float32)>, <tf.Tensor: id=892358, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 4.81769930e-05, -3.63065374e-06, -3.17024842e-06, ...,\n",
      "       -2.03000272e-05, -9.09764094e-06, -1.12458954e-04], dtype=float32)>, <tf.Tensor: id=892261, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.0446470e-06,  9.7701079e-05, -1.1109685e-06, ...,\n",
      "        -1.1853136e-06, -1.6028384e-06, -2.2461036e-06],\n",
      "       [-1.7272707e-06,  1.0796924e-04, -1.9608324e-06, ...,\n",
      "        -1.8691160e-06, -2.8586933e-06, -4.0442433e-06],\n",
      "       [-4.9081223e-07, -1.0804031e-04, -6.4596026e-07, ...,\n",
      "        -7.2736327e-07, -7.4828836e-07, -1.2341297e-06],\n",
      "       ...,\n",
      "       [ 1.0104909e-05, -2.7572567e-04,  8.5794854e-06, ...,\n",
      "         1.0971008e-05,  1.3086964e-05,  1.7873184e-05],\n",
      "       [-1.0590213e-06,  8.3031940e-05, -9.5933069e-07, ...,\n",
      "        -1.1610896e-06, -1.3977087e-06, -2.0044881e-06],\n",
      "       [-3.3493694e-07, -3.6752471e-04, -5.6685918e-09, ...,\n",
      "        -3.8917068e-07,  3.9990706e-08, -6.7328955e-08]], dtype=float32)>, <tf.Tensor: id=892253, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.04697447e-05, -2.30163510e-04,  8.82481709e-06, -6.94789807e-04,\n",
      "       -9.09685250e-03,  2.80093809e-05,  4.53411980e-04,  1.31654842e-05,\n",
      "        1.20413897e-05,  1.41851724e-05, -4.45841055e-04,  1.76402638e-04,\n",
      "       -8.94090335e-05,  1.92261286e-05,  1.52169314e-05,  1.22366031e-03,\n",
      "       -3.49392009e-04, -1.46074127e-03, -1.13318347e-04,  5.09326987e-04,\n",
      "        5.36653446e-04, -5.72599005e-04, -1.31865905e-04,  1.01545760e-04,\n",
      "        1.84854493e-04, -8.40036009e-05, -3.05319438e-04,  1.01279868e-04,\n",
      "        2.81944114e-04, -4.27815015e-04,  1.99794245e-04,  1.50589221e-05,\n",
      "        1.48735526e-05,  1.69117739e-05, -5.71694749e-04, -6.31201256e-04,\n",
      "        2.93382298e-04, -3.65236891e-04, -3.85735068e-04, -2.93934136e-06,\n",
      "       -6.65276748e-05, -5.28056815e-04,  1.34270827e-04, -8.18372704e-04,\n",
      "        2.04220283e-04, -3.84376290e-05,  1.39854557e-04,  3.50750604e-04,\n",
      "        4.86506469e-05,  4.07415937e-04, -3.02441244e-04,  5.38424756e-05,\n",
      "        2.78370164e-04,  1.41415348e-05,  6.81159261e-04,  2.43162329e-04,\n",
      "       -3.32973141e-04, -5.82904067e-05,  6.92571193e-05,  2.23610114e-04,\n",
      "       -1.39661148e-04, -1.10408357e-04, -1.38576783e-04,  2.51967831e-05,\n",
      "        2.04277367e-05,  3.14757996e-03,  9.26345179e-04, -2.04232661e-03,\n",
      "       -3.16642923e-03, -3.27175274e-03, -2.19802343e-04,  1.25208846e-03,\n",
      "        2.94300611e-03,  1.66401966e-03,  3.27167392e-04, -2.28772653e-04,\n",
      "        5.51648950e-03,  7.13058049e-04,  5.93726942e-03,  1.19374157e-03,\n",
      "       -1.40714063e-03,  2.29427067e-04,  1.04355137e-03,  2.86198757e-03,\n",
      "       -6.29293639e-03, -2.53435737e-03,  6.35398086e-04,  1.73763908e-03,\n",
      "        5.38284250e-04, -7.34043890e-04,  3.73395538e-04,  1.17243790e-05,\n",
      "        1.04222008e-05,  1.33674012e-05,  1.42128702e-05,  1.02269105e-05,\n",
      "        1.13850765e-05,  1.18571334e-05,  1.14773356e-05,  1.09927314e-05,\n",
      "        1.11238660e-05,  1.20243440e-05,  1.08483682e-05,  1.00325306e-05,\n",
      "        1.09713346e-05,  1.08513768e-05,  1.04595965e-05,  1.13483902e-05,\n",
      "        1.34764450e-05,  1.84042037e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc2e8>, <tf.Tensor: id=892717, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-1.1489693e-06, -1.6833233e-06,  5.9379965e-07, ...,\n",
      "         1.0746279e-05, -4.5882483e-05,  3.4948596e-05],\n",
      "       [ 2.9546601e-07,  1.4546269e-07,  4.0393999e-09, ...,\n",
      "        -3.2223601e-05,  7.3070223e-06,  3.6851063e-06],\n",
      "       [-3.1058203e-09,  4.7890303e-08,  1.6453328e-08, ...,\n",
      "        -5.1261482e-06, -3.4403272e-06, -2.1731659e-05],\n",
      "       ...,\n",
      "       [ 3.4124471e-07,  1.0139497e-07,  4.0130946e-08, ...,\n",
      "        -2.4248163e-06, -6.8022773e-06,  5.2027281e-06],\n",
      "       [ 8.2291542e-08,  2.7803077e-08, -1.2357490e-08, ...,\n",
      "        -3.8713868e-05,  3.3265733e-05,  2.1135284e-05],\n",
      "       [-2.7898780e-08,  2.2691609e-07,  1.6405274e-07, ...,\n",
      "        -2.4632647e-05, -1.0671930e-06, -3.8037304e-06]], dtype=float32)>, <tf.Tensor: id=892718, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-3.7156542e-06, -6.6833991e-06, -7.9101219e-06, ...,\n",
      "        -2.7370607e-04,  2.7344522e-05, -3.9929448e-05],\n",
      "       [ 7.7220830e-07,  1.5854060e-06,  2.0704015e-06, ...,\n",
      "         4.5892655e-05,  7.4696250e-06,  4.7086625e-05],\n",
      "       [-1.2682417e-06, -5.8245996e-07, -3.9302931e-06, ...,\n",
      "        -9.3059985e-05,  1.7359858e-05,  2.5900947e-05],\n",
      "       ...,\n",
      "       [-2.3165373e-07, -4.6692168e-07, -2.7753160e-07, ...,\n",
      "        -3.0149053e-05,  3.2306459e-06, -4.0779482e-06],\n",
      "       [ 5.6110875e-06,  8.3939958e-06,  2.0959049e-05, ...,\n",
      "        -3.6439445e-04,  4.2921329e-05, -2.1970340e-05],\n",
      "       [ 7.2909888e-06,  1.8676878e-05, -2.2305176e-05, ...,\n",
      "         3.3179891e-05,  9.5948053e-06, -1.5268393e-05]], dtype=float32)>, <tf.Tensor: id=892719, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 1.3958638e-05, -2.0584569e-06,  7.1160393e-06, ...,\n",
      "       -3.0409066e-05, -3.7030122e-04,  3.4861609e-05], dtype=float32)>, <tf.Tensor: id=892622, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.09539064e-06, -4.22999328e-05, -1.17975219e-06, ...,\n",
      "        -1.24721703e-06, -1.67892040e-06, -2.40870577e-06],\n",
      "       [-1.83914756e-06, -1.03259496e-04, -2.07261542e-06, ...,\n",
      "        -1.99643068e-06, -2.97642873e-06, -4.22758058e-06],\n",
      "       [-6.24987081e-07, -1.72292755e-04, -7.31328385e-07, ...,\n",
      "        -8.38258813e-07, -8.69524683e-07, -1.52936457e-06],\n",
      "       ...,\n",
      "       [ 8.28171687e-06,  6.54409698e-04,  7.72707244e-06, ...,\n",
      "         9.15282726e-06,  1.14760960e-05,  1.60215113e-05],\n",
      "       [-8.69765017e-07, -9.78468597e-06, -8.82541940e-07, ...,\n",
      "        -9.74781301e-07, -1.25003135e-06, -1.85834892e-06],\n",
      "       [ 9.13461946e-08, -8.72118108e-06,  2.72480833e-07, ...,\n",
      "         8.38691250e-08,  4.77180379e-07,  4.93556627e-07]], dtype=float32)>, <tf.Tensor: id=892614, shape=(110,), dtype=float32, numpy=\n",
      "array([ 8.46313196e-06,  6.67444838e-04,  7.86936835e-06,  1.10782229e-03,\n",
      "       -4.69588861e-03,  2.68809217e-05,  4.14113951e-04,  1.07771020e-05,\n",
      "        1.01122414e-05,  1.27333133e-05,  1.63085904e-04,  1.87237252e-04,\n",
      "       -3.97606229e-04,  1.77860329e-05,  1.23835844e-05,  9.69460991e-04,\n",
      "       -1.09972537e-03, -1.93823606e-03,  4.13386442e-05,  1.35698152e-04,\n",
      "        4.57585091e-04, -1.95071800e-04, -7.13051122e-05, -1.71302483e-04,\n",
      "       -1.35401002e-04, -3.26702313e-04, -1.62768556e-04,  1.54971276e-04,\n",
      "        7.42678822e-05,  7.66477751e-05,  2.14106447e-04,  1.21991598e-05,\n",
      "        1.27816484e-05,  1.44077130e-05,  6.75762276e-05, -1.78075454e-04,\n",
      "       -7.34421599e-04, -2.39393616e-04, -9.07560170e-05, -5.77274586e-05,\n",
      "       -2.74302671e-04,  2.68152595e-04, -1.64345067e-04,  1.17917173e-03,\n",
      "       -5.92260098e-04,  2.60077039e-04,  1.43196958e-04, -2.94432411e-05,\n",
      "       -1.41452329e-04,  1.45286613e-04, -1.91286264e-04, -1.06927691e-04,\n",
      "       -2.48396304e-04, -3.53483018e-04, -6.01374311e-04,  2.20483285e-04,\n",
      "        1.27216903e-04, -1.10989040e-05,  6.07142501e-05,  4.58601498e-05,\n",
      "        1.50513351e-05,  2.95171958e-05, -1.38829302e-04,  2.17456436e-05,\n",
      "        1.85934332e-05, -5.77040389e-03, -1.62509875e-03, -3.63719650e-03,\n",
      "        5.05500939e-03, -2.01795786e-03, -2.31255312e-03, -2.07033037e-04,\n",
      "       -1.30294298e-03, -4.33481444e-04,  1.80787698e-04, -1.78281160e-03,\n",
      "        4.90102917e-03, -1.90817111e-03,  6.25129323e-03,  4.78313258e-03,\n",
      "       -2.58737104e-03,  7.25702033e-04, -8.95107805e-04, -2.11930182e-03,\n",
      "        2.39930884e-03, -2.14921951e-04,  4.46199020e-03,  2.17589620e-03,\n",
      "       -4.64941171e-04,  2.25511356e-03, -1.61859003e-04,  1.06452653e-05,\n",
      "        8.90147658e-06,  1.08973863e-05,  1.10130195e-05,  8.24499693e-06,\n",
      "        9.59168210e-06,  9.19172544e-06,  8.94950790e-06,  8.69585983e-06,\n",
      "        9.57296925e-06,  9.67847882e-06,  9.17256966e-06,  8.28878456e-06,\n",
      "        8.65337279e-06,  8.78675291e-06,  8.99985935e-06,  9.34399759e-06,\n",
      "        1.16953088e-05,  1.63304867e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdae76a0>, <tf.Tensor: id=893078, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 1.8546887e-06,  8.0681950e-08, -9.5350123e-07, ...,\n",
      "         7.1970586e-05, -3.2405733e-05,  5.0954492e-05],\n",
      "       [-2.2609902e-07, -4.9707932e-08,  2.0914355e-07, ...,\n",
      "        -4.9530230e-05, -2.8860572e-05, -2.9844936e-05],\n",
      "       [-1.6469139e-06, -6.0472553e-07,  3.0293035e-07, ...,\n",
      "         1.9821955e-05,  1.0293670e-05,  3.5647605e-05],\n",
      "       ...,\n",
      "       [-4.3422534e-07, -2.9511625e-07,  1.3223244e-07, ...,\n",
      "         3.0567348e-06, -7.7616396e-06,  2.6062958e-06],\n",
      "       [ 1.1024581e-07,  1.4222876e-07, -3.9449098e-08, ...,\n",
      "        -3.1453706e-05,  3.4183355e-05, -2.1010774e-06],\n",
      "       [-1.8305580e-07,  1.7396806e-08, -1.8690831e-07, ...,\n",
      "         5.9727745e-06,  3.0061072e-05,  3.5210796e-05]], dtype=float32)>, <tf.Tensor: id=893079, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-1.97978238e-07, -3.30807438e-07,  2.16461808e-06, ...,\n",
      "         2.47087428e-06, -2.07327929e-07, -4.59892763e-05],\n",
      "       [-4.22631501e-07, -8.85209943e-07,  5.47250238e-07, ...,\n",
      "         1.84309014e-04, -2.35596017e-05, -5.09715464e-05],\n",
      "       [ 6.00322608e-07,  4.76928790e-07, -2.89783179e-06, ...,\n",
      "        -8.80001986e-04,  1.24612852e-04, -6.94951596e-05],\n",
      "       ...,\n",
      "       [ 4.95151426e-07,  2.95709185e-07, -1.49076584e-06, ...,\n",
      "        -5.91275239e-05,  6.62870025e-06, -9.18562455e-06],\n",
      "       [-2.86713453e-06, -1.07948645e-05, -7.65464677e-07, ...,\n",
      "         3.06052971e-04, -3.87810214e-05,  9.61489422e-06],\n",
      "       [-3.90671721e-06, -1.32670848e-05,  7.88457874e-06, ...,\n",
      "        -2.51648162e-04,  1.04534220e-05,  1.46641096e-05]], dtype=float32)>, <tf.Tensor: id=893080, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 2.4346118e-06,  5.6174514e-07, -6.5136364e-06, ...,\n",
      "       -6.2509484e-05,  3.1123994e-04, -2.6289310e-04], dtype=float32)>, <tf.Tensor: id=892983, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.21241339e-06, -3.20160252e-05, -1.29387786e-06, ...,\n",
      "        -1.36843232e-06, -1.83033910e-06, -2.57714282e-06],\n",
      "       [-1.69352074e-06, -7.81502822e-05, -2.02727460e-06, ...,\n",
      "        -1.83425959e-06, -2.89517220e-06, -4.12639838e-06],\n",
      "       [-6.07684058e-07, -4.22365629e-05, -7.89262629e-07, ...,\n",
      "        -9.00743998e-07, -9.43494229e-07, -1.57114073e-06],\n",
      "       ...,\n",
      "       [ 1.12171547e-05,  5.31118829e-04,  9.92651621e-06, ...,\n",
      "         1.21246330e-05,  1.46225839e-05,  1.96419151e-05],\n",
      "       [-1.14752470e-06, -3.62602914e-05, -1.08716495e-06, ...,\n",
      "        -1.24555368e-06, -1.52779444e-06, -2.17956949e-06],\n",
      "       [-1.16688106e-07,  1.02889171e-05,  1.30707164e-07, ...,\n",
      "        -2.01871885e-07,  2.83364585e-07,  3.13606307e-07]], dtype=float32)>, <tf.Tensor: id=892975, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.20771065e-05,  5.39896428e-04,  1.05274103e-05,  1.46516238e-03,\n",
      "        1.53447967e-02,  3.28962451e-05,  1.14868919e-03,  1.45049698e-05,\n",
      "        1.39028471e-05,  1.65783931e-05, -1.23554689e-03,  4.10543551e-04,\n",
      "        1.52919616e-04,  2.29922189e-05, -1.39518859e-04,  2.22444511e-03,\n",
      "       -1.18543056e-03, -2.77023297e-04,  4.81249372e-05, -5.24608658e-05,\n",
      "       -4.98491630e-04,  1.21808625e-04,  2.24652817e-04, -5.70746022e-04,\n",
      "        1.76124086e-04,  1.29772845e-04, -2.56852509e-05,  2.35885382e-04,\n",
      "       -6.71089161e-04, -1.41560624e-04,  9.68791865e-05,  1.69720861e-05,\n",
      "        1.66901209e-05,  1.88539925e-05, -2.12078885e-04, -3.17596889e-04,\n",
      "        2.93559395e-04,  2.92025827e-04,  1.03740822e-04,  2.94502766e-04,\n",
      "        2.51947669e-04,  8.45866634e-06, -1.09760622e-04,  8.41130211e-04,\n",
      "        6.31476432e-05, -4.92194085e-04,  5.25342912e-05, -3.14436213e-04,\n",
      "        2.14282845e-04,  1.94352906e-04,  1.21954690e-05,  5.81952190e-05,\n",
      "       -5.21636975e-05,  4.03154991e-04, -1.64511308e-04,  7.84880758e-05,\n",
      "        1.44367747e-04,  2.53778446e-04,  7.02368488e-05,  2.44185067e-04,\n",
      "        1.93453579e-05,  2.07781675e-04,  2.01461298e-04,  2.76891096e-05,\n",
      "        2.27874298e-05, -9.87677835e-03, -1.34481408e-03, -6.25012349e-03,\n",
      "        4.54557221e-03, -9.72020626e-03, -2.51661125e-03,  1.37547089e-03,\n",
      "       -1.25977164e-03, -5.26319630e-03,  2.91039381e-04, -8.94479337e-04,\n",
      "       -4.32536611e-03,  1.08060392e-03,  9.87824518e-04,  4.83995676e-03,\n",
      "        2.56506982e-03,  7.36592570e-04, -3.08246980e-03, -2.54222401e-03,\n",
      "        3.62597522e-03,  3.99106788e-03, -1.20605109e-04,  3.85102583e-04,\n",
      "        4.20392113e-04,  1.13549118e-03,  8.81598447e-04,  1.39711237e-05,\n",
      "        1.18063135e-05,  1.54421632e-05,  1.62337783e-05,  1.17322370e-05,\n",
      "        1.35343716e-05,  1.34867969e-05,  1.31707884e-05,  1.25855113e-05,\n",
      "       -1.43084879e-04, -1.42684265e-04,  1.22772744e-05,  1.16061292e-05,\n",
      "        1.27309486e-05,  1.22800284e-05,  1.18033358e-05,  1.29905584e-05,\n",
      "        1.54843674e-05,  2.07656612e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdae6748>, <tf.Tensor: id=893439, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-1.4841555e-06, -1.9000014e-07,  8.7509238e-07, ...,\n",
      "        -4.5226596e-05,  3.8324770e-06,  1.6749886e-05],\n",
      "       [-1.2255746e-08,  1.8434028e-07,  3.5820580e-07, ...,\n",
      "        -1.3846365e-05, -3.9030042e-06, -4.2110632e-06],\n",
      "       [ 8.6359222e-07,  4.7695312e-07, -1.9099022e-07, ...,\n",
      "         1.7499358e-05, -1.8652903e-05,  2.9079592e-05],\n",
      "       ...,\n",
      "       [-3.2879950e-07, -2.3723484e-07,  2.6263046e-07, ...,\n",
      "         2.3249843e-06,  2.5063534e-06,  5.6014219e-06],\n",
      "       [-3.5317374e-08, -3.5383039e-09,  4.4402039e-08, ...,\n",
      "        -2.2921855e-05,  4.4713885e-05,  5.0690373e-06],\n",
      "       [-7.4203984e-07, -4.8551975e-08,  3.1712750e-07, ...,\n",
      "         1.5271964e-05, -4.9501261e-05,  3.2018906e-06]], dtype=float32)>, <tf.Tensor: id=893440, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-6.4882233e-06, -6.7560504e-06, -5.6071649e-06, ...,\n",
      "        -8.0871210e-04,  1.0617055e-04, -1.5727736e-04],\n",
      "       [-8.1465129e-07, -1.4673524e-06, -1.5807619e-06, ...,\n",
      "         2.5933760e-04, -2.5599444e-05,  3.1643485e-06],\n",
      "       [-8.6737913e-07,  7.4426936e-07, -5.8467849e-06, ...,\n",
      "        -4.1617721e-04,  5.9899543e-05, -3.2120235e-05],\n",
      "       ...,\n",
      "       [-1.7524867e-07, -3.3103797e-07, -8.4914728e-07, ...,\n",
      "        -1.9127674e-05,  2.2519052e-06, -5.4942529e-07],\n",
      "       [ 1.2076584e-06, -9.2471428e-06, -2.5636873e-06, ...,\n",
      "        -4.1173811e-05, -1.0241254e-06, -2.8834371e-05],\n",
      "       [-6.5156437e-06, -6.1848032e-06,  2.0562020e-06, ...,\n",
      "         6.9409645e-05, -5.5133183e-05,  2.8687009e-07]], dtype=float32)>, <tf.Tensor: id=893441, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 1.25365195e-05,  1.15426360e-06,  3.48090316e-06, ...,\n",
      "       -2.03318723e-05, -4.39965152e-05,  6.75294432e-05], dtype=float32)>, <tf.Tensor: id=893344, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.1550812e-06,  4.0470968e-06, -1.2389913e-06, ...,\n",
      "        -1.2846439e-06, -1.7444023e-06, -2.5161517e-06],\n",
      "       [-1.7867119e-06,  9.0194633e-05, -2.0583411e-06, ...,\n",
      "        -1.8895640e-06, -2.9400030e-06, -4.2119800e-06],\n",
      "       [-6.8051111e-07, -3.7308221e-04, -7.9497551e-07, ...,\n",
      "        -9.0633853e-07, -9.0120335e-07, -1.6098018e-06],\n",
      "       ...,\n",
      "       [ 1.0444801e-05, -1.1234009e-03,  9.1638221e-06, ...,\n",
      "         1.1056747e-05,  1.3606650e-05,  1.8757750e-05],\n",
      "       [-1.0566484e-06,  1.4126267e-04, -9.9946601e-07, ...,\n",
      "        -1.1362379e-06, -1.4040169e-06, -2.0691255e-06],\n",
      "       [-1.2007834e-07,  4.1033502e-04,  2.2142383e-07, ...,\n",
      "        -8.8927223e-08,  3.6597521e-07,  3.6488689e-07]], dtype=float32)>, <tf.Tensor: id=893336, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.0797048e-05, -1.1910493e-03,  9.4110810e-06,  1.2822198e-03,\n",
      "        7.7834823e-03,  3.1354699e-05,  1.3861497e-03,  1.3179559e-05,\n",
      "        1.2501206e-05,  1.5718595e-05, -1.7307188e-04,  4.4185593e-04,\n",
      "        2.9988494e-04, -2.9047258e-04,  1.5269419e-05, -2.4064144e-04,\n",
      "       -3.1167053e-04, -1.6894555e-03, -1.0914208e-04,  5.7985708e-05,\n",
      "        8.2859304e-04, -4.8624637e-04,  1.3096383e-04, -7.9873913e-05,\n",
      "        1.7955355e-04,  2.8365903e-04, -1.7646897e-04,  6.6231674e-05,\n",
      "        4.4506608e-04,  1.1212433e-06, -8.0800834e-05, -1.4077616e-04,\n",
      "        1.5211884e-05, -1.3878066e-04, -8.3498796e-04, -1.7500759e-04,\n",
      "       -1.4855062e-04, -1.2901799e-04,  2.5790225e-04, -4.5329024e-04,\n",
      "        2.6431758e-04,  5.0094770e-04, -2.3244014e-04,  1.4677747e-03,\n",
      "       -8.1384263e-05, -7.6546694e-06, -8.3619641e-05, -2.8860653e-04,\n",
      "       -9.7043892e-05,  2.1940262e-04, -1.1938374e-04,  5.7190191e-05,\n",
      "        2.9549640e-04,  2.8691770e-04, -8.8649167e-04,  7.7615507e-05,\n",
      "       -5.7458246e-06,  3.1972994e-04,  6.8057212e-05,  2.4889983e-04,\n",
      "        1.8319695e-05,  2.0270383e-04,  1.9126940e-04,  2.5744039e-05,\n",
      "       -2.9107294e-04, -1.4123648e-02,  1.6491770e-04, -7.0299506e-03,\n",
      "       -9.9830492e-04,  1.0149535e-02, -1.2823093e-03, -1.2320053e-03,\n",
      "       -2.1812867e-03,  2.9551988e-03, -2.9889209e-04,  8.6094456e-04,\n",
      "       -1.8655073e-03,  3.2725409e-03,  3.1526701e-04, -6.6654920e-04,\n",
      "       -3.7312647e-03,  3.4961643e-04,  3.8491185e-03, -4.1784542e-03,\n",
      "        2.0267677e-03,  1.5697597e-03,  3.1558725e-03,  1.6862351e-03,\n",
      "        6.8380112e-05, -2.2044012e-03,  2.6734138e-04,  1.2597461e-05,\n",
      "        1.0805033e-05,  1.3900722e-05,  1.4241720e-05,  1.0507671e-05,\n",
      "        1.2138246e-05,  1.1706194e-05,  1.1245295e-05,  1.1038984e-05,\n",
      "        1.1945401e-05,  1.2232454e-05,  1.1127350e-05,  1.0215595e-05,\n",
      "        1.1241798e-05,  1.0926545e-05,  1.0783047e-05,  1.1407325e-05,\n",
      "        1.3992910e-05,  1.9273124e-05], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc4a8>, <tf.Tensor: id=893800, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-8.75284840e-08,  5.46237345e-07,  7.32585249e-07, ...,\n",
      "        -5.16853543e-05, -1.92451043e-05,  1.04550745e-05],\n",
      "       [-2.64116249e-08, -2.09665725e-08,  2.28612222e-07, ...,\n",
      "         2.55431260e-05,  3.24149119e-06,  5.04124000e-06],\n",
      "       [-1.28585029e-07,  6.96675073e-08,  7.19946414e-08, ...,\n",
      "        -3.13476958e-05, -4.30184582e-06,  7.17550961e-07],\n",
      "       ...,\n",
      "       [ 5.45307557e-07,  4.29971543e-08, -1.38846687e-07, ...,\n",
      "        -8.61377339e-06, -1.06996440e-05, -2.88448314e-06],\n",
      "       [ 1.14335791e-07,  3.80101710e-08, -1.59219269e-08, ...,\n",
      "        -5.73781836e-05,  1.72189302e-05, -5.59733905e-07],\n",
      "       [ 3.39385366e-07,  4.99876265e-08, -1.18154169e-07, ...,\n",
      "        -7.61232286e-06,  2.84242615e-05,  2.60863180e-05]], dtype=float32)>, <tf.Tensor: id=893801, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-1.1916499e-06,  2.3491925e-06,  5.5971134e-07, ...,\n",
      "        -4.7688375e-04,  5.1236049e-05, -8.5920277e-05],\n",
      "       [ 4.5150068e-07,  4.9106967e-07,  1.8755268e-06, ...,\n",
      "         1.3022398e-05, -7.0265869e-06,  4.4466310e-06],\n",
      "       [-6.7376099e-08, -3.6275330e-07,  3.0711092e-07, ...,\n",
      "         2.6941142e-04, -3.9521350e-05, -2.2020682e-05],\n",
      "       ...,\n",
      "       [ 3.9436188e-07,  8.7422802e-07,  7.3943693e-07, ...,\n",
      "         1.4046967e-05, -1.4280681e-06,  2.6505895e-06],\n",
      "       [ 6.7250394e-06,  9.7054190e-06,  1.5919777e-05, ...,\n",
      "        -2.5788744e-04,  2.8910348e-05, -4.8798123e-05],\n",
      "       [ 6.6569023e-06,  1.1803479e-05, -7.1792083e-06, ...,\n",
      "        -4.8631607e-04,  6.9082926e-05, -2.8080023e-05]], dtype=float32)>, <tf.Tensor: id=893802, shape=(6144,), dtype=float32, numpy=\n",
      "array([-7.0755632e-06, -5.2867749e-06, -3.2692774e-06, ...,\n",
      "        1.4361518e-05, -2.6384418e-04, -4.9193844e-04], dtype=float32)>, <tf.Tensor: id=893705, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.08243330e-06,  9.39174133e-05, -1.17189484e-06, ...,\n",
      "        -1.17918125e-06, -1.66430095e-06, -2.39284691e-06],\n",
      "       [-1.62103845e-06,  1.09351131e-04, -1.92318521e-06, ...,\n",
      "        -1.69034479e-06, -2.74399122e-06, -3.94104700e-06],\n",
      "       [-5.34313074e-07, -2.47299264e-04, -6.67516588e-07, ...,\n",
      "        -7.56117572e-07, -7.43121632e-07, -1.33380036e-06],\n",
      "       ...,\n",
      "       [ 1.00449379e-05, -1.02575263e-03,  8.77022285e-06, ...,\n",
      "         1.06048810e-05,  1.31711968e-05,  1.80446932e-05],\n",
      "       [-1.04431081e-06,  1.17525255e-04, -9.68649260e-07, ...,\n",
      "        -1.10837061e-06, -1.38706810e-06, -2.02042065e-06],\n",
      "       [ 4.48608688e-08,  2.67217576e-04,  3.37895074e-07, ...,\n",
      "         1.82182518e-08,  5.79328002e-07,  6.15336319e-07]], dtype=float32)>, <tf.Tensor: id=893697, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.03878501e-05, -1.06401020e-03,  9.00817031e-06,  1.05503295e-03,\n",
      "       -7.22395862e-03,  2.86749182e-05,  1.29059685e-04,  1.27881149e-05,\n",
      "        1.20373934e-05,  1.55495072e-05, -1.34815084e-04,  2.95525038e-04,\n",
      "        9.19326267e-05,  2.17929301e-05, -1.41385390e-04, -2.21487763e-03,\n",
      "        2.13461841e-04, -2.48200051e-03, -2.67310621e-04, -2.43230039e-04,\n",
      "       -4.95359767e-04,  3.31417978e-05,  6.27859961e-04, -6.31177027e-05,\n",
      "        5.31511090e-04,  4.53722867e-04, -5.82169334e-04,  7.58548049e-05,\n",
      "       -1.00274076e-04, -3.83990555e-05, -1.17517171e-04,  1.51309414e-05,\n",
      "        1.49297703e-05,  1.68427905e-05, -3.62148770e-04,  2.54737359e-04,\n",
      "        3.24601242e-05, -1.08833265e-04, -2.15990614e-04,  6.49260983e-06,\n",
      "        2.58695218e-04,  1.95476183e-04,  2.51453690e-04, -8.97574180e-04,\n",
      "        8.36159306e-05, -1.53365618e-04,  2.48363882e-04,  2.06492823e-05,\n",
      "        2.15527223e-04,  2.04038777e-04,  3.51999697e-05, -9.96017843e-05,\n",
      "        4.37711657e-04, -7.90167134e-04,  8.56526545e-04,  6.65566768e-05,\n",
      "       -6.42990653e-06, -9.26138426e-04, -8.96584315e-05, -6.08117916e-05,\n",
      "        1.82129916e-05,  1.96447814e-04,  1.74531728e-04,  2.47664666e-05,\n",
      "        2.07371504e-05, -8.00601556e-04,  8.84673849e-04, -1.33967830e-03,\n",
      "       -9.63768456e-04,  1.22996923e-02,  4.48993582e-04,  3.02879536e-03,\n",
      "        3.71485774e-04,  5.55743463e-03, -1.09934399e-03, -1.54600816e-03,\n",
      "       -1.87724899e-03,  3.43377981e-03, -3.03836237e-03, -3.67261260e-03,\n",
      "        1.42336707e-04,  2.90976895e-04,  1.78987090e-03, -1.84215547e-03,\n",
      "       -7.25452602e-03,  5.90026844e-03,  3.17763351e-03, -1.72225526e-03,\n",
      "       -7.90898979e-04,  1.59185205e-04, -1.49340500e-04,  1.20402356e-05,\n",
      "        1.05026702e-05,  1.33315471e-05,  1.41294495e-05,  1.01988126e-05,\n",
      "        1.14891072e-05,  1.16666042e-05,  1.12369189e-05,  1.07347041e-05,\n",
      "        1.17708223e-05,  1.19710230e-05,  1.06603493e-05,  9.73528859e-06,\n",
      "        1.08917147e-05,  1.05281779e-05,  1.04326227e-05,  1.09523535e-05,\n",
      "        1.35465507e-05,  1.85564404e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc0f0>, <tf.Tensor: id=894161, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 2.70604062e-07, -1.08156280e-06, -2.05632705e-06, ...,\n",
      "        -2.07718585e-05,  8.23906976e-06,  5.18948327e-05],\n",
      "       [-1.41061605e-07, -8.69030146e-08,  2.29989453e-07, ...,\n",
      "         3.45065473e-07,  6.13265947e-05,  8.29377495e-06],\n",
      "       [-2.23633734e-07,  1.98961221e-08,  4.38316121e-08, ...,\n",
      "        -2.81022858e-05,  6.76890249e-06, -2.17416914e-06],\n",
      "       ...,\n",
      "       [ 8.89003047e-07,  3.73853453e-07, -1.64448110e-07, ...,\n",
      "        -7.16006207e-06,  1.88693684e-05,  7.19672289e-06],\n",
      "       [ 2.84961263e-08,  3.69779087e-08, -1.35670115e-08, ...,\n",
      "         1.15393386e-05, -4.91841165e-05, -6.03641383e-05],\n",
      "       [-3.34873391e-07,  3.44385086e-07,  2.28160431e-07, ...,\n",
      "        -1.94858876e-05, -1.17360005e-05,  2.56010617e-06]], dtype=float32)>, <tf.Tensor: id=894162, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 6.44162083e-06,  3.65736810e-06,  1.11045165e-05, ...,\n",
      "         3.91672133e-04, -4.96493521e-05,  1.80296847e-05],\n",
      "       [-8.10524853e-07, -1.12705834e-06,  1.34437187e-07, ...,\n",
      "        -1.41920027e-05,  9.77154377e-06, -9.32172043e-06],\n",
      "       [-5.21625566e-07, -6.20180970e-07, -4.06762581e-07, ...,\n",
      "         2.69416370e-04, -5.04186537e-05,  4.89400009e-05],\n",
      "       ...,\n",
      "       [ 4.62371361e-07,  3.03776289e-07, -4.61003857e-08, ...,\n",
      "        -3.09657793e-08,  2.95961513e-07, -6.81130587e-07],\n",
      "       [-2.36006122e-06, -3.42474982e-06,  2.38368602e-06, ...,\n",
      "        -2.80939712e-05,  1.42056047e-06, -2.35550742e-05],\n",
      "       [-2.45211395e-05, -4.38362003e-05,  2.18688401e-05, ...,\n",
      "         1.19580065e-04, -4.45582955e-05,  9.77400778e-05]], dtype=float32)>, <tf.Tensor: id=894163, shape=(6144,), dtype=float32, numpy=\n",
      "array([-1.10039655e-05,  2.58901423e-06,  4.58185468e-06, ...,\n",
      "       -2.75760840e-07, -2.82047768e-05,  1.18284253e-04], dtype=float32)>, <tf.Tensor: id=894066, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.8698717e-07, -1.0515734e-04, -1.1015850e-06, ...,\n",
      "        -1.0701322e-06, -1.5499406e-06, -2.1550643e-06],\n",
      "       [-1.5155241e-06, -1.8687433e-04, -1.8450867e-06, ...,\n",
      "        -1.5975224e-06, -2.6197658e-06, -3.6721835e-06],\n",
      "       [-6.9668954e-07, -1.5758845e-04, -7.7790367e-07, ...,\n",
      "        -9.0181857e-07, -9.3532157e-07, -1.5459384e-06],\n",
      "       ...,\n",
      "       [ 8.9869318e-06,  2.1721292e-03,  8.3153172e-06, ...,\n",
      "         9.5259693e-06,  1.2224571e-05,  1.6211903e-05],\n",
      "       [-9.0429273e-07, -1.9265930e-04, -8.9767912e-07, ...,\n",
      "        -9.5787505e-07, -1.2489925e-06, -1.7554860e-06],\n",
      "       [ 9.7853636e-08, -2.3787111e-04,  3.2095591e-07, ...,\n",
      "         6.9920169e-08,  5.8665063e-07,  6.6943244e-07]], dtype=float32)>, <tf.Tensor: id=894058, shape=(110,), dtype=float32, numpy=\n",
      "array([ 9.25476434e-06,  2.27769604e-03,  8.51652385e-06,  2.61671812e-04,\n",
      "       -8.37051589e-03,  2.81547927e-05,  9.01548832e-04,  1.12463649e-05,\n",
      "        1.09571110e-05,  1.46896909e-05, -1.16039056e-03,  2.50179088e-04,\n",
      "        2.45528994e-04,  2.10122398e-05,  1.33329795e-05,  1.26490544e-04,\n",
      "        1.02089881e-03,  5.32402075e-04,  4.40891818e-05, -4.33800189e-04,\n",
      "       -5.05860313e-04, -4.58490802e-04,  3.38308193e-04,  2.11993523e-04,\n",
      "        3.16523045e-04,  2.14242697e-04, -6.23375177e-04, -1.17246695e-04,\n",
      "        4.93562839e-05,  1.50935011e-04, -1.14670453e-04,  1.33844023e-05,\n",
      "        1.33179992e-05,  1.57202721e-05,  1.35110167e-04,  2.88477808e-04,\n",
      "        4.01429832e-04,  3.20947234e-04,  2.11643826e-04, -5.92185024e-05,\n",
      "       -2.95541977e-04,  3.26362351e-04,  1.06837302e-04, -1.19497208e-03,\n",
      "        6.48936330e-05, -1.95304565e-05,  5.26653253e-04, -6.65644766e-04,\n",
      "       -1.36784583e-04, -5.33947547e-04, -2.02275751e-05,  5.15257161e-05,\n",
      "        2.04937256e-04,  2.02682873e-04,  6.02955057e-04,  3.73759067e-05,\n",
      "       -2.02862975e-05,  1.70162075e-05, -9.61014739e-05,  2.14855609e-04,\n",
      "       -1.39529948e-04,  1.67526785e-04,  1.95705579e-05,  2.16649405e-05,\n",
      "        2.01755192e-05,  8.42259172e-03,  1.65102468e-03, -2.73887883e-03,\n",
      "        4.17705014e-04, -5.27251279e-04,  1.15967519e-03,  1.02238008e-03,\n",
      "        5.22696134e-03,  2.82753026e-04,  7.71309889e-04,  1.14062859e-03,\n",
      "       -2.03986024e-03,  6.73610996e-03, -9.43775196e-03, -6.81015989e-03,\n",
      "        2.78740510e-04,  5.86287293e-04,  9.31108720e-04, -3.58252041e-03,\n",
      "       -2.71518948e-03,  2.68515665e-04, -3.94961389e-04, -4.83476237e-04,\n",
      "        9.88193206e-04,  2.51531228e-03,  5.22502087e-05,  1.16417777e-05,\n",
      "        9.50071171e-06,  1.19408642e-05,  1.21710118e-05,  8.96314668e-06,\n",
      "        1.06612051e-05,  9.96031213e-06,  9.58433066e-06,  9.11471852e-06,\n",
      "        1.04943610e-05,  1.06645775e-05,  9.84989038e-06,  8.67630297e-06,\n",
      "        9.50506728e-06,  9.45709508e-06,  9.42251518e-06,  9.79990273e-06,\n",
      "        1.25285169e-05,  1.66114714e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafca58>, <tf.Tensor: id=894522, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-9.03198838e-07, -2.05422737e-07,  1.69298218e-07, ...,\n",
      "        -5.33432758e-05, -4.24132486e-05,  3.36507146e-05],\n",
      "       [ 7.44011572e-07,  3.09358143e-07, -2.33468512e-07, ...,\n",
      "        -2.55939244e-06,  3.91094436e-05,  3.33563439e-05],\n",
      "       [-7.64211222e-07,  9.54477741e-08,  1.61866126e-07, ...,\n",
      "        -2.16346743e-05, -2.40938807e-05, -2.42945885e-06],\n",
      "       ...,\n",
      "       [-1.01715834e-07, -1.21978934e-08,  1.42621914e-07, ...,\n",
      "         8.44821716e-06,  2.16335047e-06,  6.81876327e-06],\n",
      "       [-4.45036363e-08,  1.39494247e-08,  2.35071216e-08, ...,\n",
      "         5.84437385e-06, -1.89198272e-05, -4.25669823e-05],\n",
      "       [ 2.13086679e-07,  3.70455022e-09, -2.32970763e-07, ...,\n",
      "        -2.68599724e-05,  3.64080006e-05, -1.46190487e-05]], dtype=float32)>, <tf.Tensor: id=894523, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 6.99106920e-07, -1.59804222e-06,  8.54734481e-06, ...,\n",
      "        -2.33296829e-04,  5.24722236e-06, -4.96489083e-05],\n",
      "       [-1.39449821e-06, -2.92071354e-06, -2.48606329e-06, ...,\n",
      "        -1.36217684e-04,  1.46185521e-05, -4.01762591e-05],\n",
      "       [-4.60154467e-07, -4.38755507e-07,  9.74578029e-07, ...,\n",
      "         2.98397848e-04, -3.81729697e-05,  6.59576017e-07],\n",
      "       ...,\n",
      "       [-6.60298497e-07,  1.12351552e-06, -4.15712657e-06, ...,\n",
      "         1.06559804e-04, -7.98238580e-06,  1.46854509e-05],\n",
      "       [-1.21110281e-06, -1.88883303e-06,  1.82647964e-05, ...,\n",
      "        -5.56716521e-04,  6.50274014e-05, -9.60684847e-05],\n",
      "       [-5.95992287e-06, -5.56938767e-06, -3.46022171e-06, ...,\n",
      "        -1.01570680e-04, -4.88452679e-06,  3.95890711e-05]], dtype=float32)>, <tf.Tensor: id=894524, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 2.7706271e-06,  1.8277533e-06,  8.0214369e-07, ...,\n",
      "        1.2382526e-04, -6.0239871e-04, -1.0198164e-04], dtype=float32)>, <tf.Tensor: id=894427, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.25878137e-06, -6.61831873e-05, -1.30664534e-06, ...,\n",
      "        -1.31433785e-06, -1.77443485e-06, -2.50145649e-06],\n",
      "       [-1.92566267e-06, -5.84846857e-05, -2.15601835e-06, ...,\n",
      "        -1.92683569e-06, -3.01436830e-06, -4.19196840e-06],\n",
      "       [-9.88424404e-07, -6.98193035e-05, -1.07839617e-06, ...,\n",
      "        -1.17893796e-06, -1.23355335e-06, -1.82073882e-06],\n",
      "       ...,\n",
      "       [ 1.25704109e-05,  5.91943535e-05,  1.13813840e-05, ...,\n",
      "         1.28242009e-05,  1.57125323e-05,  2.09573009e-05],\n",
      "       [-1.08439315e-06,  1.27268504e-06, -1.04843252e-06, ...,\n",
      "        -1.10119572e-06, -1.42444480e-06, -2.05230549e-06],\n",
      "       [-2.04071043e-07,  1.04385894e-04,  4.68459547e-08, ...,\n",
      "        -2.46478521e-07,  2.78745858e-07,  2.32061154e-07]], dtype=float32)>, <tf.Tensor: id=894419, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.39607882e-05,  6.10293137e-05,  1.24643338e-05,  1.02609221e-03,\n",
      "       -1.05911540e-02, -1.20572462e-04,  4.20900556e-04,  1.61876433e-05,\n",
      "        1.59381998e-05,  2.05643264e-05,  1.26918720e-03,  2.47284974e-04,\n",
      "        1.00945501e-04,  2.84860016e-05, -1.37294890e-04, -6.25862274e-04,\n",
      "       -2.12558516e-04,  1.19652686e-04,  5.32734994e-05, -6.27679925e-04,\n",
      "        3.66027700e-04,  5.14939427e-04,  5.53966849e-04, -5.06964861e-05,\n",
      "        1.73346634e-05,  3.96593037e-04,  7.55069850e-05, -9.09611990e-05,\n",
      "        4.60661657e-04,  6.40230137e-04,  5.35803774e-05,  1.93308260e-05,\n",
      "        1.91112231e-05,  2.23863572e-05,  3.26127512e-04, -4.35568218e-04,\n",
      "        2.72834382e-04,  7.56197187e-05, -1.02188322e-03, -1.24464952e-03,\n",
      "        4.41180018e-05, -2.83911038e-04, -1.75750829e-05, -4.64662386e-04,\n",
      "        2.51264282e-04,  1.74646892e-04, -3.78778437e-04,  1.68141778e-05,\n",
      "       -5.70008473e-04, -1.26521429e-03,  2.84883790e-05,  6.31614384e-05,\n",
      "       -3.91143229e-04,  2.77955929e-04, -4.89821774e-04, -1.02422156e-04,\n",
      "       -1.54685069e-04, -6.77688222e-04, -2.38692301e-04, -7.73252323e-05,\n",
      "        2.30393598e-05,  1.75353707e-04,  1.90868814e-04,  2.73319911e-05,\n",
      "        2.70322726e-05,  8.44298676e-03,  1.71177497e-03, -2.83258758e-03,\n",
      "        1.73005392e-05, -1.60466018e-03, -5.98677318e-04, -6.56108372e-04,\n",
      "       -2.04178831e-03,  1.00543117e-03,  1.48872437e-04,  1.49794051e-03,\n",
      "       -2.13392847e-03,  7.60843570e-04,  4.70649626e-04, -2.68986006e-03,\n",
      "       -1.28287892e-03,  2.32597056e-04, -3.78435990e-03, -3.31802585e-04,\n",
      "        5.18826814e-03,  3.04607977e-03,  3.15012387e-03,  2.64169183e-04,\n",
      "       -1.66880229e-04,  3.64122353e-03,  3.08809831e-06,  1.67005564e-05,\n",
      "        1.44438354e-05,  1.68589540e-05,  1.80617844e-05,  1.38541300e-05,\n",
      "        1.58758321e-05,  1.43503321e-05,  1.43367424e-05,  1.35011915e-05,\n",
      "        1.47387200e-05,  1.54775262e-05,  1.43421366e-05,  1.39648409e-05,\n",
      "        1.44304395e-05,  1.38197047e-05,  1.37622892e-05,  1.41895453e-05,\n",
      "        1.71374413e-05,  2.28499412e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdae6198>, <tf.Tensor: id=894883, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-1.5472713e-06, -1.3220176e-06,  1.5076379e-08, ...,\n",
      "         3.4856137e-05, -7.0585156e-05,  2.9614890e-05],\n",
      "       [ 1.6713628e-07, -9.2189268e-08, -9.5597564e-08, ...,\n",
      "        -5.2034273e-05,  2.7719318e-05, -8.5635702e-06],\n",
      "       [-3.6172236e-07,  4.7657664e-08,  5.4705353e-08, ...,\n",
      "         2.7739263e-05, -3.4886057e-06,  1.7549999e-05],\n",
      "       ...,\n",
      "       [ 4.0468706e-07,  6.6817435e-08,  3.7814836e-08, ...,\n",
      "        -1.0699372e-06, -1.1641533e-05, -1.4658450e-06],\n",
      "       [ 1.1045230e-07,  6.6481157e-08, -4.2358209e-08, ...,\n",
      "        -3.2710817e-05,  1.3689195e-05,  4.6987989e-06],\n",
      "       [-7.4788824e-09,  2.7187701e-07,  9.5777736e-08, ...,\n",
      "        -7.1997038e-06, -2.2239950e-05,  2.3171609e-05]], dtype=float32)>, <tf.Tensor: id=894884, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 3.16200294e-06, -2.30747219e-07,  5.74498745e-06, ...,\n",
      "         1.58942756e-04,  1.67783528e-05,  6.73126851e-06],\n",
      "       [-2.30818927e-07, -3.14940323e-08,  1.27067199e-06, ...,\n",
      "         1.14935465e-04, -1.16099163e-05, -3.57294812e-05],\n",
      "       [ 5.37817186e-07,  4.77576521e-07,  1.12535872e-06, ...,\n",
      "        -4.23027755e-04,  4.26383885e-05, -2.08746787e-05],\n",
      "       ...,\n",
      "       [-2.09034667e-07, -7.12591714e-07,  3.43942730e-07, ...,\n",
      "         2.06247842e-05, -2.53822782e-06,  8.79630988e-07],\n",
      "       [-1.42778981e-05, -3.62449900e-05, -3.99891542e-05, ...,\n",
      "         4.80051618e-04, -5.61735251e-05,  9.44045823e-05],\n",
      "       [-1.62331025e-06,  1.46643561e-06, -1.39855429e-05, ...,\n",
      "        -2.01737901e-04,  2.41642156e-05,  4.94997221e-05]], dtype=float32)>, <tf.Tensor: id=894885, shape=(6144,), dtype=float32, numpy=\n",
      "array([-8.5943739e-06, -2.1478371e-08, -2.4883411e-06, ...,\n",
      "        2.0529358e-05,  4.8336928e-04, -2.0256396e-04], dtype=float32)>, <tf.Tensor: id=894788, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.6231864e-07, -1.4140313e-04, -9.8479279e-07, ...,\n",
      "        -9.4510636e-07, -1.3834224e-06, -1.9201893e-06],\n",
      "       [-1.6190170e-06, -2.3581422e-04, -1.9109652e-06, ...,\n",
      "        -1.7083178e-06, -2.7342662e-06, -3.7217121e-06],\n",
      "       [-6.3876138e-07, -2.1441749e-04, -7.1621895e-07, ...,\n",
      "        -7.8881374e-07, -8.6912246e-07, -1.4958224e-06],\n",
      "       ...,\n",
      "       [ 6.9724947e-06,  2.3239353e-03,  6.8814034e-06, ...,\n",
      "         7.4552504e-06,  1.0138263e-05,  1.3495817e-05],\n",
      "       [-6.8471104e-07, -2.1045361e-04, -7.3166859e-07, ...,\n",
      "        -7.3449689e-07, -1.0218581e-06, -1.4296700e-06],\n",
      "       [ 3.9173767e-07, -3.7285482e-04,  5.5189707e-07, ...,\n",
      "         4.0251402e-07,  8.9959821e-07,  1.0091017e-06]], dtype=float32)>, <tf.Tensor: id=894780, shape=(110,), dtype=float32, numpy=\n",
      "array([ 7.13526970e-06,  2.39583082e-03,  7.01473527e-06, -1.05081906e-03,\n",
      "        8.49806797e-03, -1.31062086e-04, -2.64878356e-04,  9.10947165e-06,\n",
      "        8.41629299e-06,  1.30068038e-05,  1.47225845e-04,  1.66962724e-04,\n",
      "       -9.10433155e-05,  1.89839484e-05,  1.15197345e-05,  2.69803975e-04,\n",
      "        8.97602877e-04,  2.67883670e-03,  4.14378446e-05,  4.13077127e-04,\n",
      "       -4.04289240e-05, -4.42515448e-05,  8.68578645e-05, -2.09524485e-04,\n",
      "       -6.66191481e-05,  1.48580468e-04, -1.10644414e-05,  2.27808865e-04,\n",
      "        2.80241045e-04,  1.96970810e-04,  2.21419497e-04, -3.01742461e-04,\n",
      "        1.10631054e-05, -2.99698615e-04,  3.38034297e-05, -1.14549402e-05,\n",
      "       -1.73749955e-04,  4.36584640e-04,  2.80281638e-05, -2.39510075e-04,\n",
      "        1.21065932e-04,  2.95431324e-04,  3.27249028e-04, -2.93871388e-04,\n",
      "        5.36939478e-05, -2.07813355e-05,  3.31983465e-05, -2.04909622e-04,\n",
      "       -1.58058188e-04,  4.02365185e-05, -4.78695656e-05,  4.70490195e-05,\n",
      "        1.42780904e-04,  1.60302588e-04,  1.34210149e-03,  5.97725739e-06,\n",
      "        1.23439080e-04,  2.95131875e-04,  5.37272863e-05, -2.84039357e-04,\n",
      "       -1.41151089e-04, -1.76642730e-04,  4.29537158e-06,  1.70081912e-05,\n",
      "        1.78359260e-05, -7.31321517e-04, -6.25238579e-04,  2.92960135e-03,\n",
      "       -2.17399996e-04, -4.39801533e-03,  1.69458240e-03,  1.39826778e-04,\n",
      "        1.74436637e-03, -2.01484858e-04, -2.50107696e-05,  1.87376246e-03,\n",
      "       -4.06841701e-03, -5.73966710e-04,  2.71346234e-03, -6.15706854e-03,\n",
      "       -1.66579278e-03,  4.79398528e-04, -2.32065096e-03,  5.01783704e-03,\n",
      "       -7.97170866e-03, -1.70042436e-03, -2.03014538e-03,  4.84382559e-04,\n",
      "       -1.61911157e-06, -1.18293555e-03,  5.61682915e-04,  9.60326634e-06,\n",
      "        8.04547653e-06,  9.18896694e-06,  9.12012729e-06,  7.11774601e-06,\n",
      "        8.51777622e-06,  7.21692413e-06,  7.34489095e-06,  7.15462056e-06,\n",
      "        8.85886402e-06,  8.38566211e-06,  7.89915248e-06,  6.73201612e-06,\n",
      "        7.45227180e-06,  7.30740521e-06,  7.87860063e-06,  7.62312311e-06,\n",
      "        1.03396897e-05,  1.37649122e-05], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdae7908>, <tf.Tensor: id=895244, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 1.36475524e-06, -4.54191735e-07, -5.49037736e-07, ...,\n",
      "        -9.17180841e-07, -1.81072628e-05,  4.13859489e-05],\n",
      "       [ 2.58387587e-07,  2.73797241e-07,  7.53864029e-08, ...,\n",
      "         2.94987501e-07,  3.01829059e-05, -1.64935846e-05],\n",
      "       [-2.57675708e-07,  7.30786667e-08, -1.54188928e-08, ...,\n",
      "        -8.50966444e-07,  2.29584111e-05,  2.11771094e-05],\n",
      "       ...,\n",
      "       [ 5.64291270e-07,  3.03136630e-07,  1.73132108e-07, ...,\n",
      "        -2.26537637e-07,  1.50320602e-06, -7.53027365e-08],\n",
      "       [ 6.86958330e-08,  3.39574502e-09,  1.68407706e-08, ...,\n",
      "         1.36885628e-05,  1.31898223e-05, -2.12809173e-05],\n",
      "       [-4.67853454e-07,  1.22398859e-07,  1.62880312e-07, ...,\n",
      "        -1.44338355e-05, -1.28392503e-05,  3.56654273e-05]], dtype=float32)>, <tf.Tensor: id=895245, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 1.4976076e-06, -7.6703452e-07,  4.1884095e-06, ...,\n",
      "         1.7312936e-04, -1.6426409e-05,  1.0967479e-05],\n",
      "       [-4.1210666e-07, -1.8529349e-06, -2.3466998e-07, ...,\n",
      "        -1.4964143e-04,  2.5891506e-05,  5.8072260e-07],\n",
      "       [ 7.1054853e-07,  1.2547051e-06, -1.0366348e-07, ...,\n",
      "        -2.4396034e-04,  1.4922862e-05, -8.8325505e-05],\n",
      "       ...,\n",
      "       [ 3.6821896e-07,  4.8145580e-07,  9.1090237e-07, ...,\n",
      "         8.9397590e-06, -1.0693745e-06,  1.8955723e-06],\n",
      "       [-3.9099941e-06, -6.0047228e-06,  2.0459302e-06, ...,\n",
      "         5.7512875e-07, -5.3177178e-07, -2.3498835e-06],\n",
      "       [-1.2519490e-05, -2.6761585e-05,  1.1864983e-05, ...,\n",
      "        -8.5637817e-05, -2.4336687e-05,  7.7632540e-05]], dtype=float32)>, <tf.Tensor: id=895246, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 6.2374943e-06,  1.6063231e-07, -1.6656884e-06, ...,\n",
      "        9.2945793e-06,  1.1694974e-06, -8.5261985e-05], dtype=float32)>, <tf.Tensor: id=895149, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.8371343e-07, -9.4148039e-05, -9.7422821e-07, ...,\n",
      "        -9.6021040e-07, -1.3996204e-06, -1.9649347e-06],\n",
      "       [-1.6239927e-06, -1.7751136e-04, -1.8509095e-06, ...,\n",
      "        -1.6750064e-06, -2.6848829e-06, -3.7628195e-06],\n",
      "       [-6.2900608e-07, -1.3399569e-04, -7.0183455e-07, ...,\n",
      "        -8.1171038e-07, -8.3398783e-07, -1.4121542e-06],\n",
      "       ...,\n",
      "       [ 8.8839133e-06,  1.3792594e-03,  7.8204421e-06, ...,\n",
      "         9.2312694e-06,  1.1829969e-05,  1.5870530e-05],\n",
      "       [-8.6619292e-07, -1.4199935e-04, -8.1869132e-07, ...,\n",
      "        -9.1896311e-07, -1.1810305e-06, -1.6926770e-06],\n",
      "       [ 1.5226792e-07,  1.7175873e-05,  4.0829866e-07, ...,\n",
      "         1.4948176e-07,  6.9646592e-07,  7.5689263e-07]], dtype=float32)>, <tf.Tensor: id=895141, shape=(110,), dtype=float32, numpy=\n",
      "array([ 9.1518541e-06,  1.4440147e-03,  8.0125446e-06, -4.3859909e-04,\n",
      "        6.4655393e-03,  2.8259656e-05, -8.6593610e-04,  1.0888873e-05,\n",
      "        1.0518006e-05,  1.4451529e-05,  1.1895683e-05, -4.9535150e-04,\n",
      "       -5.2195467e-04,  2.2529435e-05,  1.4159305e-05,  2.6823226e-03,\n",
      "        1.3376096e-03, -3.2550868e-04,  4.6784782e-05,  1.0720795e-04,\n",
      "       -1.7179988e-04, -5.5712888e-05,  3.1580514e-04, -5.0659670e-05,\n",
      "       -3.1125161e-04, -4.7108531e-04,  6.5487629e-04,  4.1959254e-05,\n",
      "       -7.1718969e-04,  3.4062445e-04, -2.5223815e-04,  1.3689257e-05,\n",
      "        1.3492676e-05,  1.6044729e-05,  2.0563460e-04, -1.3268157e-04,\n",
      "        7.3068324e-05,  4.7925624e-04,  5.4543449e-05,  6.8121153e-06,\n",
      "       -6.1547762e-04,  4.4092690e-04, -1.7068484e-04, -8.2750269e-04,\n",
      "        7.1379327e-05,  1.7037967e-04, -7.2241161e-04, -5.8916846e-04,\n",
      "       -1.0697431e-04,  1.4755336e-04, -4.6040851e-04,  5.2856834e-05,\n",
      "       -5.8707938e-04, -9.8193786e-04,  1.0736783e-03,  1.7416215e-04,\n",
      "       -1.5302441e-05,  2.2038136e-04,  6.3068284e-05, -2.3856903e-04,\n",
      "       -1.3859903e-04, -9.1177926e-06,  1.7120925e-04,  2.1363267e-05,\n",
      "        2.0326355e-05,  5.6283157e-03,  1.3074729e-03,  2.6311474e-03,\n",
      "        1.5725354e-03, -3.7355528e-03,  8.7646593e-05,  7.4647425e-05,\n",
      "        2.0951487e-03,  1.8137116e-03, -5.1238848e-04,  1.3978787e-03,\n",
      "       -6.1624330e-03,  6.6015881e-04, -4.7470448e-03, -2.1418314e-03,\n",
      "       -2.9018784e-03,  5.5660232e-04, -7.8951626e-04, -1.1455867e-03,\n",
      "        3.3276104e-03, -1.5967307e-03,  7.7798741e-04, -3.7109600e-03,\n",
      "       -1.1559599e-03, -3.1222467e-04,  7.4812488e-07,  1.1385691e-05,\n",
      "        9.4684983e-06,  1.1742471e-05,  1.1966991e-05,  8.8547713e-06,\n",
      "        1.0461715e-05,  9.8758710e-06,  9.4155375e-06,  9.0493750e-06,\n",
      "        1.0486630e-05,  1.0694330e-05,  9.4045854e-06,  8.3354535e-06,\n",
      "        9.4946517e-06,  9.2600048e-06,  9.4598163e-06,  9.4985862e-06,\n",
      "        1.2131259e-05,  1.6274969e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafce10>, <tf.Tensor: id=895605, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 1.7828012e-06,  1.1681409e-06, -1.7240179e-06, ...,\n",
      "         5.8358951e-06, -3.5071046e-06,  2.5757481e-05],\n",
      "       [ 4.2245344e-07, -4.7038075e-08,  1.4946446e-07, ...,\n",
      "        -1.8108629e-05,  3.2608597e-05,  2.3552795e-06],\n",
      "       [-5.2758093e-08,  2.9097387e-07,  2.3019604e-08, ...,\n",
      "        -1.6919445e-05,  1.7166485e-05,  4.9534610e-06],\n",
      "       ...,\n",
      "       [ 1.3222837e-07, -8.0839158e-08,  2.8470376e-08, ...,\n",
      "         1.1199288e-06, -4.9302889e-06,  3.8836047e-06],\n",
      "       [ 5.5060276e-08,  2.2465769e-09, -8.7580068e-09, ...,\n",
      "        -5.4441749e-05,  1.7922212e-05, -4.7154626e-05],\n",
      "       [ 2.7571329e-07,  2.0609050e-07,  1.6650108e-07, ...,\n",
      "         4.0035949e-05, -2.1093185e-05,  2.8310289e-05]], dtype=float32)>, <tf.Tensor: id=895606, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 1.0046770e-05,  4.8874299e-06,  4.9169666e-06, ...,\n",
      "         6.7576411e-04, -6.5200751e-05,  4.9779534e-05],\n",
      "       [-3.1904921e-07, -4.7864432e-07,  6.8524793e-07, ...,\n",
      "         9.5788855e-06,  1.2716302e-05,  3.7070358e-06],\n",
      "       [ 2.2948382e-07,  3.1113660e-07,  1.4711027e-06, ...,\n",
      "         2.6239152e-04, -2.1124240e-05,  2.7826385e-05],\n",
      "       ...,\n",
      "       [ 6.1496047e-07,  7.3180934e-07,  2.2670527e-07, ...,\n",
      "        -4.8383367e-06,  9.3875786e-07, -2.7008441e-06],\n",
      "       [-8.5685474e-07, -5.3196259e-06,  2.6661351e-07, ...,\n",
      "         2.3434738e-04, -2.7406497e-05,  2.8507506e-05],\n",
      "       [ 6.1851370e-07,  9.2541313e-08,  1.9111203e-06, ...,\n",
      "         2.0160500e-05,  1.4964287e-06,  9.8184813e-05]], dtype=float32)>, <tf.Tensor: id=895607, shape=(6144,), dtype=float32, numpy=\n",
      "array([-6.6181469e-06, -5.9789312e-07, -5.7241755e-07, ...,\n",
      "       -4.8303846e-06,  2.3796971e-04,  1.9241637e-05], dtype=float32)>, <tf.Tensor: id=895510, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-7.2896694e-07, -1.1258351e-04, -8.4575925e-07, ...,\n",
      "        -8.0476900e-07, -1.1929853e-06, -1.6335966e-06],\n",
      "       [-1.4426098e-06, -2.2418000e-04, -1.7056128e-06, ...,\n",
      "        -1.5254200e-06, -2.4540636e-06, -3.2936807e-06],\n",
      "       [-4.2526125e-07, -1.8200124e-04, -5.0140687e-07, ...,\n",
      "        -5.7977780e-07, -5.6983106e-07, -1.0458799e-06],\n",
      "       ...,\n",
      "       [ 5.9268082e-06,  1.5732468e-03,  5.8675296e-06, ...,\n",
      "         6.3661155e-06,  8.7240078e-06,  1.1440244e-05],\n",
      "       [-5.6859665e-07, -1.4172659e-04, -6.1278627e-07, ...,\n",
      "        -6.2077203e-07, -8.5915036e-07, -1.2041469e-06],\n",
      "       [ 4.7147790e-07, -3.2729359e-04,  6.0484012e-07, ...,\n",
      "         4.7196733e-07,  9.8060582e-07,  1.1298976e-06]], dtype=float32)>, <tf.Tensor: id=895502, shape=(110,), dtype=float32, numpy=\n",
      "array([ 6.04545767e-06,  1.61434046e-03,  5.96270365e-06, -5.97238250e-04,\n",
      "        2.24055699e-03,  2.20615002e-05, -4.61452291e-04,  7.78931098e-06,\n",
      "        7.23614994e-06,  1.16144784e-05,  1.14717265e-03, -5.36689877e-05,\n",
      "       -2.98731407e-04,  1.73245571e-05,  1.04389101e-05, -1.30353670e-03,\n",
      "       -3.94086412e-04,  9.81432502e-04,  3.75839227e-05,  3.17513244e-04,\n",
      "        4.27179475e-04,  7.41056283e-05, -3.74973984e-04, -4.86777542e-04,\n",
      "       -5.12446168e-05, -1.69724823e-04,  3.81436286e-04, -8.49391508e-05,\n",
      "        5.14590472e-04,  2.40312278e-04, -4.44267411e-04,  9.59886074e-06,\n",
      "        9.67745291e-06,  1.16865594e-05,  2.84110647e-05,  1.74943852e-04,\n",
      "       -5.37425978e-04,  2.35400541e-04,  8.22173024e-06, -7.54702851e-05,\n",
      "       -2.13082647e-04, -8.65335460e-05, -3.23663640e-04,  1.62299388e-04,\n",
      "        3.70838497e-05,  2.80751730e-04,  3.04354617e-04,  3.03555978e-04,\n",
      "       -4.64089635e-06,  3.14718563e-05, -4.81486277e-05, -1.13847185e-04,\n",
      "       -1.88117192e-04,  4.54417052e-04,  6.85078732e-04,  1.41009295e-04,\n",
      "        1.17912881e-04,  2.50254379e-05, -1.06167245e-04,  1.85179262e-04,\n",
      "        1.44533205e-05,  1.17676791e-04,  1.37337745e-04,  1.44791411e-05,\n",
      "        1.58251951e-05,  6.88659912e-03,  8.25462455e-04,  3.33163771e-03,\n",
      "       -1.11735775e-03, -4.94826213e-03, -4.42184974e-04, -1.39118230e-03,\n",
      "       -4.36400805e-05, -4.25956910e-03, -2.19121372e-04,  1.57789048e-03,\n",
      "        1.99788390e-03, -4.08789981e-03,  5.81519213e-03, -3.48048308e-03,\n",
      "       -3.02236830e-03,  1.66884085e-04, -4.34133661e-04, -3.37189808e-03,\n",
      "        7.24419719e-03, -6.51908515e-04, -2.93857767e-03, -1.55975504e-04,\n",
      "        3.13798082e-04, -2.69662030e-03, -1.87819998e-04,  8.24420113e-06,\n",
      "        6.99000748e-06,  8.01734132e-06,  7.67889651e-06,  6.18676086e-06,\n",
      "        7.14617181e-06,  6.34116259e-06,  6.25756093e-06,  6.02969521e-06,\n",
      "        7.41230178e-06,  7.21434390e-06,  6.81434403e-06,  5.69000986e-06,\n",
      "        6.34909566e-06,  6.40090184e-06,  6.87453485e-06,  6.48962350e-06,\n",
      "        8.87329952e-06,  1.16335796e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc8d0>, <tf.Tensor: id=895966, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 1.6726023e-06, -2.0782839e-07, -1.8369326e-06, ...,\n",
      "        -4.1115884e-05, -1.5939506e-05,  6.6102302e-07],\n",
      "       [ 3.2941853e-07,  1.2646011e-07, -7.9070524e-08, ...,\n",
      "        -4.5433564e-05, -1.5623515e-05,  1.6560176e-05],\n",
      "       [-1.5541026e-07, -3.4307627e-07,  9.4025188e-09, ...,\n",
      "         2.4892175e-05,  1.2596059e-05, -1.3871008e-05],\n",
      "       ...,\n",
      "       [ 9.0839609e-07,  2.5731455e-07,  1.4871965e-09, ...,\n",
      "         1.1014015e-05, -1.2103412e-05,  4.2234640e-07],\n",
      "       [-9.7671808e-08, -1.7056331e-08, -1.0796054e-09, ...,\n",
      "        -2.3231751e-05,  3.6456317e-05, -2.7640408e-05],\n",
      "       [ 4.3752581e-07,  1.2639946e-07,  1.9800285e-07, ...,\n",
      "         3.3072131e-06,  7.2221678e-07,  2.2344631e-05]], dtype=float32)>, <tf.Tensor: id=895967, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 5.60949184e-06, -3.79894277e-06, -1.10644271e-06, ...,\n",
      "         7.73614098e-04, -8.34840976e-05,  7.94664666e-05],\n",
      "       [-3.20534213e-07, -6.69055112e-07, -2.58309024e-06, ...,\n",
      "         9.66067637e-06,  5.01082650e-06, -5.74215665e-05],\n",
      "       [-8.12023643e-07, -1.08563324e-06,  1.34787228e-06, ...,\n",
      "         1.13286733e-04, -2.09689438e-06,  1.76287303e-05],\n",
      "       ...,\n",
      "       [ 1.04277365e-06,  2.06867867e-06, -2.28843737e-06, ...,\n",
      "        -7.56453373e-05,  7.45606712e-06, -1.77905840e-05],\n",
      "       [ 3.09902293e-06, -1.87632259e-06, -1.11565691e-06, ...,\n",
      "         1.83373137e-04, -2.61364239e-05,  1.90473547e-05],\n",
      "       [ 4.91261244e-06,  1.28856445e-05, -1.10223173e-05, ...,\n",
      "         3.18901759e-04, -6.01903484e-05,  4.97570909e-05]], dtype=float32)>, <tf.Tensor: id=895968, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 7.0740880e-06,  3.0769129e-06,  3.6464476e-06, ...,\n",
      "       -7.8641933e-05,  1.8937222e-04,  3.2088294e-04], dtype=float32)>, <tf.Tensor: id=895871, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-7.0977273e-07, -1.1449730e-04, -8.2432177e-07, ...,\n",
      "        -7.7605841e-07, -1.1549275e-06, -1.5689079e-06],\n",
      "       [-1.3962530e-06, -2.5064457e-04, -1.6435205e-06, ...,\n",
      "        -1.4552511e-06, -2.3433963e-06, -3.1440488e-06],\n",
      "       [-4.0104544e-07, -3.4111709e-04, -4.8767367e-07, ...,\n",
      "        -5.8418226e-07, -5.5380099e-07, -9.6808049e-07],\n",
      "       ...,\n",
      "       [ 6.7828487e-06,  2.4596432e-03,  6.2767704e-06, ...,\n",
      "         7.1210143e-06,  9.3428062e-06,  1.2167100e-05],\n",
      "       [-6.3271727e-07, -2.2595288e-04, -6.4080905e-07, ...,\n",
      "        -6.7706196e-07, -8.9209732e-07, -1.2563819e-06],\n",
      "       [ 4.2756074e-07, -2.6664353e-04,  5.5772978e-07, ...,\n",
      "         4.0930018e-07,  9.1343622e-07,  1.0695860e-06]], dtype=float32)>, <tf.Tensor: id=895863, shape=(110,), dtype=float32, numpy=\n",
      "array([ 6.9845382e-06,  2.5907485e-03,  6.4208116e-06, -7.6044188e-04,\n",
      "       -1.8906222e-03, -1.3305701e-04, -1.0950223e-03,  8.5868132e-06,\n",
      "        8.1104426e-06,  1.2268309e-05,  2.2171056e-04,  2.3827086e-04,\n",
      "        1.8599009e-04,  1.8738303e-05,  1.1773623e-05, -5.3725846e-05,\n",
      "        2.9920012e-04, -4.6844446e-04,  3.8945651e-05,  5.6627207e-04,\n",
      "       -2.3046722e-04,  4.1036247e-04, -1.6439933e-04, -2.7945350e-04,\n",
      "        2.4815425e-05, -2.8069739e-04,  1.7312684e-04, -4.6999048e-05,\n",
      "        1.6594194e-04,  1.9941530e-04,  1.8907513e-04,  1.0954447e-05,\n",
      "        1.0608612e-05,  1.3109993e-05,  1.9398369e-04,  1.1995819e-04,\n",
      "       -5.0703634e-04, -2.4349766e-04,  1.7412165e-04, -1.9660340e-04,\n",
      "        1.2835466e-04,  2.0909584e-04,  1.9878609e-04,  3.9436467e-04,\n",
      "        4.1677980e-05, -2.6882171e-05,  4.6192121e-04, -1.1102694e-04,\n",
      "        2.1350148e-05, -2.4507366e-04, -1.3442288e-05,  4.4343800e-05,\n",
      "        3.0066748e-04,  4.9316505e-04,  1.5255858e-04, -1.7687575e-05,\n",
      "        1.2307092e-04,  5.9682620e-04, -1.0297209e-04,  2.0533457e-04,\n",
      "        1.6048407e-05, -1.9506837e-04, -1.7864286e-04,  1.6032473e-05,\n",
      "        1.6723070e-05, -1.4602088e-03, -1.8933524e-03,  1.3409358e-03,\n",
      "        1.6757578e-03, -8.5045286e-03, -3.4846822e-03,  1.3582181e-03,\n",
      "        1.2444432e-03, -3.0931334e-03,  2.4193448e-04,  1.3812284e-03,\n",
      "        4.7733560e-03, -5.2127084e-03,  1.3615339e-03,  5.6364015e-03,\n",
      "       -1.3477101e-03, -5.1151850e-04,  3.4948036e-03, -5.2489992e-04,\n",
      "        5.6351204e-03, -2.3864058e-03, -9.9549722e-04,  1.8702449e-03,\n",
      "        1.7081776e-04, -2.6566936e-03,  2.7985996e-04,  9.0636840e-06,\n",
      "        7.6849483e-06,  9.1198690e-06,  8.9352716e-06,  6.9632615e-06,\n",
      "        7.9510946e-06,  7.5413200e-06,  7.2405660e-06, -3.0562707e-04,\n",
      "        8.1374410e-06,  8.1505514e-06, -3.0493602e-04,  6.4315300e-06,\n",
      "        7.2873913e-06,  7.1584886e-06,  7.6732267e-06,  7.3237102e-06,\n",
      "        9.5708328e-06,  1.2465192e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc940>, <tf.Tensor: id=896327, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-5.0031326e-07, -9.1350580e-07, -8.5431361e-07, ...,\n",
      "        -8.1121754e-05,  1.1419252e-05,  2.4909770e-05],\n",
      "       [ 1.2569565e-07,  8.3704847e-08,  6.0458468e-08, ...,\n",
      "         1.1749271e-05,  9.3658537e-06,  1.7424187e-05],\n",
      "       [ 5.1977725e-07,  3.3173009e-07, -1.6308482e-07, ...,\n",
      "        -1.5813850e-05,  2.0160447e-05, -8.5877382e-06],\n",
      "       ...,\n",
      "       [-5.5108035e-07, -1.2508461e-07,  8.3673442e-08, ...,\n",
      "         2.2766683e-06, -7.3108390e-06,  2.7229548e-06],\n",
      "       [-3.5989519e-08,  3.6741603e-09,  6.5102931e-08, ...,\n",
      "         3.8203994e-05,  3.5246740e-05,  4.3006598e-06],\n",
      "       [ 4.8104363e-07,  5.6009361e-07, -1.4285789e-07, ...,\n",
      "        -7.2332463e-05,  3.8491753e-06, -6.1179030e-06]], dtype=float32)>, <tf.Tensor: id=896328, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 3.0026356e-06,  5.9518929e-06,  4.6809223e-06, ...,\n",
      "        -4.7842154e-04,  4.0194172e-05, -8.4234780e-05],\n",
      "       [ 3.4242838e-08, -1.7788966e-06, -4.9090670e-07, ...,\n",
      "        -2.9577548e-04,  4.1189211e-05,  5.1546587e-05],\n",
      "       [ 6.8945775e-07,  1.5644918e-06, -1.3682484e-07, ...,\n",
      "         2.7007025e-04, -4.6437810e-05, -8.6872162e-05],\n",
      "       ...,\n",
      "       [ 6.1718492e-07,  5.9618884e-07,  1.5767776e-07, ...,\n",
      "        -1.7268108e-05,  1.6957462e-06, -3.5825615e-06],\n",
      "       [ 4.6344358e-06,  5.6353383e-06,  1.3928480e-05, ...,\n",
      "        -3.4578281e-04,  4.9206683e-05, -5.9581238e-05],\n",
      "       [-6.2584684e-07,  4.7209014e-06, -8.3346104e-06, ...,\n",
      "         2.7639561e-04, -3.7228987e-05, -6.8423600e-05]], dtype=float32)>, <tf.Tensor: id=896329, shape=(6144,), dtype=float32, numpy=\n",
      "array([-5.9947697e-06,  5.5879434e-07, -4.5593852e-06, ...,\n",
      "       -1.7656292e-05, -3.4712348e-04,  2.8174013e-04], dtype=float32)>, <tf.Tensor: id=896232, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.21594426e-07, -5.50654840e-06, -9.01683961e-07, ...,\n",
      "        -8.97048608e-07, -1.27403382e-06, -1.75471041e-06],\n",
      "       [-1.23417794e-06, -7.08434818e-05, -1.48577612e-06, ...,\n",
      "        -1.29128398e-06, -2.12978034e-06, -2.95880318e-06],\n",
      "       [-5.26519216e-07, -1.44698803e-04, -5.74534283e-07, ...,\n",
      "        -6.87774445e-07, -6.65051061e-07, -1.06313564e-06],\n",
      "       ...,\n",
      "       [ 8.21880349e-06,  3.99470446e-06,  7.16503519e-06, ...,\n",
      "         8.58653038e-06,  1.07271208e-05,  1.38938885e-05],\n",
      "       [-8.10771667e-07,  1.03395432e-05, -7.44262593e-07, ...,\n",
      "        -8.62668628e-07, -1.05846266e-06, -1.47868559e-06],\n",
      "       [-1.00175725e-08, -2.42718219e-04,  2.83688735e-07, ...,\n",
      "        -1.39820449e-08,  4.80393055e-07,  5.83635597e-07]], dtype=float32)>, <tf.Tensor: id=896224, shape=(110,), dtype=float32, numpy=\n",
      "array([ 8.46626426e-06, -9.21238643e-06,  7.34388959e-06, -5.38421809e-05,\n",
      "       -8.13080091e-03,  2.57622723e-05, -2.36349821e-04,  9.88287138e-06,\n",
      "        9.74237264e-06,  1.30854605e-05,  7.17582763e-04, -4.02465026e-04,\n",
      "       -8.64097034e-04,  2.11355145e-05, -1.42714882e-04, -5.29250829e-05,\n",
      "        8.37071624e-04, -1.87125138e-03,  4.21485456e-05,  1.06549320e-04,\n",
      "       -1.62626465e-03,  8.96532147e-04, -3.28782538e-04, -1.50573265e-04,\n",
      "        5.12031431e-04, -3.17970291e-04,  1.19557815e-04, -7.33589346e-04,\n",
      "       -1.28336003e-04, -4.55406640e-04,  3.62200008e-05,  1.33877811e-05,\n",
      "        1.22327056e-05,  1.51970335e-05, -1.14604351e-04,  1.59771822e-04,\n",
      "        4.20625001e-05,  2.35649364e-04, -1.23030797e-04,  1.93716041e-04,\n",
      "        8.95415724e-07, -1.22034486e-04,  4.34499321e-04,  2.07652687e-04,\n",
      "       -2.65412120e-04, -1.31556908e-05,  1.56643408e-04, -6.67702174e-04,\n",
      "       -9.50004833e-05,  2.96535523e-04,  3.46404064e-04,  4.80384697e-05,\n",
      "       -1.30802146e-04,  5.52305195e-04,  1.38077317e-04,  1.38166288e-04,\n",
      "        1.29987719e-04, -4.57955357e-05,  5.94012636e-05, -8.20215064e-05,\n",
      "        1.80400548e-05, -3.29595132e-05, -1.76917165e-05,  1.94987952e-05,\n",
      "        1.84445853e-05, -8.09514895e-03,  1.31986017e-04,  5.66270715e-03,\n",
      "        1.91320875e-03,  8.66783725e-04,  2.68805103e-04, -1.36529631e-03,\n",
      "       -8.16953252e-04,  2.34362879e-03, -5.42119960e-04, -5.41262561e-04,\n",
      "        2.06827419e-03, -3.11878393e-03, -8.62790854e-04,  9.58165433e-03,\n",
      "        1.63966173e-03, -5.61450899e-04,  2.47213664e-03,  5.42243710e-04,\n",
      "        2.76859500e-03, -1.48537220e-03, -2.67655170e-03,  5.19864378e-04,\n",
      "        7.03657512e-04, -5.78411971e-04, -4.06545063e-04,  1.06391544e-05,\n",
      "        8.65315542e-06,  1.10002356e-05,  1.14233781e-05,  8.38737651e-06,\n",
      "        9.42026509e-06,  9.46818000e-06,  8.81050255e-06,  8.43851376e-06,\n",
      "        9.58697001e-06,  9.98435735e-06,  8.73517456e-06,  7.63401385e-06,\n",
      "        8.91110631e-06,  8.56270435e-06,  8.81382039e-06,  8.83332996e-06,\n",
      "        1.10014244e-05,  1.42445006e-05], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdae7908>, <tf.Tensor: id=896688, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-2.6618577e-06, -1.4033328e-07,  7.3306302e-07, ...,\n",
      "        -4.0544655e-05, -3.4646077e-05, -2.9443236e-05],\n",
      "       [ 2.6974516e-07,  2.0295353e-07,  4.6090307e-08, ...,\n",
      "        -4.7707486e-05,  2.9198311e-05, -6.8353679e-06],\n",
      "       [ 2.1070534e-07,  2.7015903e-07, -1.6213720e-07, ...,\n",
      "        -2.6446074e-05,  2.4137804e-05,  8.7696508e-06],\n",
      "       ...,\n",
      "       [-2.0985125e-07, -1.0028765e-07,  5.0027268e-07, ...,\n",
      "         3.8401345e-06, -6.8673780e-06, -1.0023811e-05],\n",
      "       [-5.8329427e-08,  3.4447358e-09,  7.5222950e-08, ...,\n",
      "        -5.2653053e-05,  4.6579924e-05,  1.3836817e-05],\n",
      "       [-1.3626418e-07, -1.7497362e-07, -6.3079042e-08, ...,\n",
      "         2.7722845e-05, -2.3591361e-05, -1.5116768e-05]], dtype=float32)>, <tf.Tensor: id=896689, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-5.05242406e-06, -2.32799903e-06, -3.32311265e-06, ...,\n",
      "        -7.52156600e-04,  7.70719853e-05, -1.23425634e-04],\n",
      "       [ 8.91664570e-07,  1.60257787e-06,  8.44482031e-07, ...,\n",
      "        -5.74383648e-06,  8.38047890e-06, -2.71602621e-05],\n",
      "       [-7.41082317e-07, -8.40487417e-07, -7.88647981e-07, ...,\n",
      "        -9.49520982e-05,  3.18753191e-05,  4.51626584e-05],\n",
      "       ...,\n",
      "       [ 1.01321163e-07, -3.89872866e-07,  5.12332861e-07, ...,\n",
      "         7.46848718e-06,  3.67082237e-07, -1.41376961e-06],\n",
      "       [-3.35104119e-07, -4.31703666e-06,  7.98897190e-07, ...,\n",
      "        -2.29700221e-04,  3.18727289e-05, -7.44639547e-05],\n",
      "       [ 6.65924745e-06,  2.89367790e-06,  6.15791669e-06, ...,\n",
      "         1.02956175e-04, -4.81243387e-06,  4.24414284e-05]], dtype=float32)>, <tf.Tensor: id=896690, shape=(6144,), dtype=float32, numpy=\n",
      "array([-7.8802259e-08, -5.3429176e-06,  2.6059740e-06, ...,\n",
      "        7.7409932e-06, -2.3173609e-04,  1.0297503e-04], dtype=float32)>, <tf.Tensor: id=896593, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-7.9266022e-07, -3.2906191e-05, -9.0151980e-07, ...,\n",
      "        -8.7161374e-07, -1.2463909e-06, -1.7097200e-06],\n",
      "       [-1.3698980e-06, -1.7196458e-04, -1.6134808e-06, ...,\n",
      "        -1.4206973e-06, -2.2832528e-06, -3.0695555e-06],\n",
      "       [-4.4008237e-07,  5.5439246e-05, -5.1044435e-07, ...,\n",
      "        -6.0015452e-07, -5.9465788e-07, -1.0130292e-06],\n",
      "       ...,\n",
      "       [ 6.5582713e-06,  8.5031986e-04,  6.1467181e-06, ...,\n",
      "         6.8506588e-06,  9.0209833e-06,  1.1831378e-05],\n",
      "       [-6.5512114e-07, -6.0755698e-05, -6.4903452e-07, ...,\n",
      "        -6.9479290e-07, -8.9782804e-07, -1.2669732e-06],\n",
      "       [ 3.1989268e-07, -7.0153917e-05,  5.0139562e-07, ...,\n",
      "         3.1280427e-07,  7.7426677e-07,  8.7978060e-07]], dtype=float32)>, <tf.Tensor: id=896585, shape=(110,), dtype=float32, numpy=\n",
      "array([ 6.72200485e-06,  8.73009383e-04,  6.26931387e-06,  1.58429684e-04,\n",
      "       -3.11135408e-03,  2.39755682e-05, -1.96471438e-03,  8.14186842e-06,\n",
      "        7.81448671e-06,  1.17812615e-05,  6.36264740e-04, -2.36687061e-04,\n",
      "       -8.19430716e-05, -1.37615280e-04,  1.16085530e-05, -2.65727658e-03,\n",
      "        2.71790195e-04,  2.14049872e-03, -1.17837713e-04,  4.71540028e-04,\n",
      "        3.00436004e-05,  3.24228255e-04, -3.59543832e-04,  4.05608953e-05,\n",
      "       -3.18207778e-04, -3.00368876e-04,  2.92536599e-04, -6.78475946e-04,\n",
      "        2.72160483e-04,  3.19344108e-04, -2.61648005e-04,  1.07400465e-05,\n",
      "        1.01562819e-05,  1.29807268e-05, -2.56982865e-04, -5.17150147e-05,\n",
      "        1.13731236e-04, -1.31430555e-04, -1.63521137e-04, -9.14821794e-06,\n",
      "        1.08628519e-04, -1.42000150e-04, -1.86130128e-04,  3.30698182e-04,\n",
      "        3.20382824e-05, -4.60599040e-05,  1.08595392e-04,  9.41450708e-05,\n",
      "       -2.84923415e-04, -1.08406814e-04,  1.46011851e-04,  4.35963302e-05,\n",
      "        2.91703094e-04,  6.46214467e-05, -1.42589837e-04,  1.19726814e-04,\n",
      "        1.15693023e-04,  4.70105297e-05,  5.23840317e-05,  1.85411307e-04,\n",
      "        1.64667181e-05, -5.17731125e-04, -4.98065085e-04,  1.54901118e-05,\n",
      "        1.63417135e-05, -1.37812290e-02,  4.08810505e-04,  3.34814959e-03,\n",
      "        1.86704053e-03,  7.13491766e-03,  2.02484080e-04, -1.67373335e-03,\n",
      "       -3.94020742e-03,  4.15104767e-03,  6.27252157e-04,  2.80014263e-03,\n",
      "        8.28472059e-03, -4.33385256e-04,  4.55481187e-03,  5.50565030e-03,\n",
      "        9.85522987e-04, -8.86800583e-04, -3.41137685e-03,  7.04844890e-04,\n",
      "       -8.55042413e-03, -2.17451644e-03, -4.34498303e-04, -6.32267096e-04,\n",
      "       -8.91344171e-05,  9.31357936e-05,  1.13422502e-04,  8.97610244e-06,\n",
      "        7.26497456e-06,  8.83288521e-06,  8.40665689e-06,  6.65915240e-06,\n",
      "        7.60624425e-06,  7.03705246e-06,  6.83966618e-06,  6.70719419e-06,\n",
      "        7.79024685e-06,  7.81185281e-06,  7.48000593e-06,  6.17095702e-06,\n",
      "        6.96968709e-06,  6.80398716e-06,  7.44493536e-06,  7.01447334e-06,\n",
      "        9.20878392e-06,  1.20746799e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc5f8>, <tf.Tensor: id=897049, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-1.3108181e-06,  1.5437449e-06, -1.1255134e-06, ...,\n",
      "        -4.2368461e-05, -1.9668985e-05,  2.3387827e-06],\n",
      "       [ 3.8806286e-07,  2.9576168e-07, -1.3481561e-07, ...,\n",
      "         1.4594799e-05, -3.4663983e-05,  8.1828130e-06],\n",
      "       [-8.7661647e-07, -2.2399350e-07,  3.0862614e-07, ...,\n",
      "        -1.9675139e-05,  1.7835737e-05,  1.9067633e-05],\n",
      "       ...,\n",
      "       [-8.3392116e-08,  6.8416924e-08,  1.3625909e-07, ...,\n",
      "         7.8978155e-06, -1.0116406e-05, -4.8768616e-06],\n",
      "       [ 5.0267769e-08,  2.2426745e-08, -2.3698089e-08, ...,\n",
      "         5.3636322e-05, -3.6257195e-05, -2.7074371e-05],\n",
      "       [-2.5646864e-07, -1.3816855e-07,  1.6449340e-07, ...,\n",
      "        -3.2084208e-05, -4.4033095e-06,  3.1701809e-05]], dtype=float32)>, <tf.Tensor: id=897050, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 4.30020964e-06,  1.04479141e-05,  9.08727725e-06, ...,\n",
      "        -4.04169987e-04,  5.09080783e-05, -7.31663386e-05],\n",
      "       [-8.48702484e-07, -2.11637507e-06, -1.17440413e-06, ...,\n",
      "         1.33888694e-04, -3.30786243e-05,  1.00513804e-04],\n",
      "       [-1.34132154e-06, -2.01731382e-06, -2.47206299e-06, ...,\n",
      "         1.48490566e-04, -3.77401557e-05,  4.50109619e-05],\n",
      "       ...,\n",
      "       [ 1.99046166e-08, -4.97146232e-07,  5.46408501e-07, ...,\n",
      "         5.05784847e-05, -5.10077507e-06,  6.95253948e-06],\n",
      "       [-2.14260649e-06, -1.99558190e-06,  8.49678509e-06, ...,\n",
      "        -1.53363770e-04,  2.32050406e-05,  1.68188235e-05],\n",
      "       [ 2.00707982e-05,  3.16175137e-05, -2.58344207e-05, ...,\n",
      "        -3.62080405e-04,  3.08467643e-05,  1.95568828e-05]], dtype=float32)>, <tf.Tensor: id=897051, shape=(6144,), dtype=float32, numpy=\n",
      "array([-2.6064679e-05,  3.5121427e-06,  9.7699303e-06, ...,\n",
      "        5.2369596e-05, -1.5978084e-04, -3.6327378e-04], dtype=float32)>, <tf.Tensor: id=896954, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.9530562e-07, -4.7176954e-06, -1.0122690e-06, ...,\n",
      "        -9.5685755e-07, -1.3845671e-06,  3.5372726e-05],\n",
      "       [-1.4990611e-06, -3.8958180e-05, -1.7521090e-06, ...,\n",
      "        -1.5084537e-06, -2.4605131e-06,  9.1156908e-05],\n",
      "       [-5.5726900e-07, -1.7945153e-04, -6.2117476e-07, ...,\n",
      "        -7.0039198e-07, -7.2808473e-07, -6.2298364e-05],\n",
      "       ...,\n",
      "       [ 8.0194441e-06,  3.3782839e-04,  7.5466473e-06, ...,\n",
      "         8.3291434e-06,  1.0935270e-05, -2.8503191e-04],\n",
      "       [-7.8629938e-07, -2.4674562e-05, -7.8403968e-07, ...,\n",
      "        -8.3450732e-07, -1.0763703e-06,  4.4622917e-05],\n",
      "       [ 2.3784781e-07, -2.1469209e-04,  4.6404887e-07, ...,\n",
      "         2.3174982e-07,  7.0835347e-07, -6.2701285e-05]], dtype=float32)>, <tf.Tensor: id=896946, shape=(110,), dtype=float32, numpy=\n",
      "array([ 8.31582747e-06,  3.77473392e-04,  7.77969035e-06, -1.15488539e-03,\n",
      "       -3.13438708e-03, -2.84347509e-04,  2.42997601e-04,  1.00949874e-05,\n",
      "        9.73234728e-06,  1.43610068e-05, -1.78543298e-04, -1.65504316e-04,\n",
      "       -7.38914241e-06,  2.26453358e-05,  1.43431253e-05,  1.32492045e-04,\n",
      "       -1.25964428e-03,  2.43996899e-03,  4.38939678e-05, -4.05220693e-04,\n",
      "        1.00009842e-04,  2.96023500e-05, -4.20440250e-04, -2.51353486e-05,\n",
      "        2.46045121e-04,  2.55358085e-04,  4.11034620e-04, -2.81619024e-04,\n",
      "        2.55759107e-04, -7.50750187e-04, -6.63966421e-05,  1.32662699e-05,\n",
      "        1.21210360e-05,  1.58310559e-05,  4.03926737e-04, -3.54617281e-04,\n",
      "        5.47925069e-04, -2.34031584e-04,  1.89936603e-04,  8.59066204e-05,\n",
      "        1.36549315e-05, -4.26430925e-04,  1.93776519e-04,  3.44099419e-04,\n",
      "       -9.56578660e-05,  2.99362902e-04, -2.97063525e-06,  2.13430598e-04,\n",
      "       -5.92117867e-05, -3.51237162e-04, -2.47605174e-04,  5.08446610e-05,\n",
      "       -2.67923169e-04, -1.27210631e-03, -6.00100553e-04, -1.83679062e-04,\n",
      "       -2.53882354e-05,  2.36095475e-05,  6.20079009e-05,  2.12369399e-04,\n",
      "       -1.35896145e-04, -2.89783311e-05, -1.42208137e-05,  1.86672405e-05,\n",
      "        1.97915797e-05, -1.93182286e-03, -2.62031402e-03, -3.95639538e-04,\n",
      "       -3.32061853e-03,  1.21001806e-02,  1.29920733e-03, -1.79679308e-04,\n",
      "        1.70472800e-03, -4.06001834e-03, -9.81682169e-05,  1.50335676e-04,\n",
      "        2.18449906e-03, -3.09006847e-03,  1.28641119e-03,  4.63367440e-04,\n",
      "        6.63597928e-03, -7.58229799e-05, -3.35971965e-03,  5.42392582e-03,\n",
      "       -9.51320492e-03,  2.15886394e-03, -1.46374339e-03,  4.05011838e-03,\n",
      "        6.66710723e-04, -2.91962083e-03,  7.51464569e-04,  1.13815477e-05,\n",
      "        8.80123116e-06, -3.01619759e-04,  1.06104708e-05,  8.41476776e-06,\n",
      "       -3.03108274e-04,  8.72587498e-06,  8.54640530e-06,  8.47441333e-06,\n",
      "        9.42559473e-06,  9.79715242e-06,  9.32811236e-06,  7.81544350e-06,\n",
      "        8.79659819e-06,  8.40224948e-06,  8.84945621e-06,  8.62721481e-06,\n",
      "        1.12775888e-05, -2.98015802e-04], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc2b0>, <tf.Tensor: id=897410, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-2.9555526e-06, -7.1759922e-07,  3.3597217e-07, ...,\n",
      "        -9.4774405e-05, -3.6326834e-05, -1.2771932e-05],\n",
      "       [ 6.3234563e-07, -2.1072150e-08, -3.9523986e-08, ...,\n",
      "        -9.1919901e-05,  2.3137638e-05, -1.0975243e-05],\n",
      "       [-2.8268249e-07,  1.3152928e-08, -1.0551635e-07, ...,\n",
      "        -5.0375002e-06, -2.1761820e-05, -7.9389707e-07],\n",
      "       ...,\n",
      "       [-1.8856397e-07, -3.5050695e-07,  4.0973705e-07, ...,\n",
      "        -5.3829858e-06, -1.0073626e-05, -5.6011381e-06],\n",
      "       [ 2.5323281e-08,  3.8183945e-08,  5.3672874e-08, ...,\n",
      "         3.6920549e-06,  5.8792040e-05,  4.9313257e-05],\n",
      "       [-8.7767233e-07, -1.3694462e-07,  1.2062853e-07, ...,\n",
      "         1.2710005e-04, -8.3458348e-05, -5.7365395e-07]], dtype=float32)>, <tf.Tensor: id=897411, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-3.7071502e-06, -4.4813009e-06, -2.8869586e-06, ...,\n",
      "        -8.1180391e-05, -3.9155912e-06, -5.8074056e-05],\n",
      "       [-4.9630592e-07, -4.5386628e-07,  1.0349588e-07, ...,\n",
      "         2.2976061e-04, -2.7699632e-05,  4.5709792e-05],\n",
      "       [-4.6922406e-07,  3.6082824e-08, -2.3817367e-06, ...,\n",
      "        -4.0554465e-04,  6.0133520e-05, -1.8483239e-05],\n",
      "       ...,\n",
      "       [-2.7918364e-07, -2.7505891e-07,  5.7080189e-07, ...,\n",
      "         3.3696575e-05, -2.4037479e-06,  6.9470025e-06],\n",
      "       [-7.3149913e-06, -2.1218797e-05, -1.3961886e-05, ...,\n",
      "         3.2349789e-04, -4.3493375e-05,  6.9569302e-05],\n",
      "       [-5.9843824e-06, -1.2231627e-05,  1.0338525e-05, ...,\n",
      "         9.1648355e-05, -2.8752469e-05,  1.4658780e-04]], dtype=float32)>, <tf.Tensor: id=897412, shape=(6144,), dtype=float32, numpy=\n",
      "array([1.8925080e-05, 1.3299515e-06, 2.2406432e-06, ..., 3.4287437e-05,\n",
      "       3.2635679e-04, 9.1704831e-05], dtype=float32)>, <tf.Tensor: id=897315, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.6818553e-07,  1.1126786e-04, -1.0577951e-06, ...,\n",
      "        -1.0252068e-06, -1.4647345e-06, -2.0863727e-06],\n",
      "       [-1.6259996e-06, -1.0794829e-06, -1.8601395e-06, ...,\n",
      "        -1.6310244e-06, -2.6395326e-06, -3.6658826e-06],\n",
      "       [-5.3992710e-07,  2.7904278e-04, -6.3963978e-07, ...,\n",
      "        -7.2623141e-07, -7.6283698e-07, -1.2053129e-06],\n",
      "       ...,\n",
      "       [ 8.3887990e-06, -1.2484756e-03,  7.6405868e-06, ...,\n",
      "         8.5040965e-06,  1.1258600e-05,  1.5089513e-05],\n",
      "       [-8.3827410e-07,  1.6147262e-04, -8.0103666e-07, ...,\n",
      "        -8.6540604e-07, -1.1019588e-06, -1.6233400e-06],\n",
      "       [ 2.8460630e-07,  2.3325000e-04,  5.4984832e-07, ...,\n",
      "         2.8705534e-07,  8.3044489e-07,  9.8103806e-07]], dtype=float32)>, <tf.Tensor: id=897307, shape=(110,), dtype=float32, numpy=\n",
      "array([ 8.6036580e-06, -1.2931323e-03,  7.8046223e-06,  1.7859066e-04,\n",
      "        1.0198978e-02,  3.0587726e-05, -1.4492818e-03,  1.0187521e-05,\n",
      "        9.8310929e-06,  1.4791957e-05,  5.9956161e-04, -2.6963570e-04,\n",
      "        2.3428134e-04,  2.4121782e-05,  1.5198695e-05,  9.2062471e-04,\n",
      "       -9.9266844e-04,  1.3203773e-03, -1.0936090e-04, -2.9412917e-05,\n",
      "        1.2754404e-05,  1.3529949e-04, -4.5993216e-05,  2.0092366e-04,\n",
      "        3.0247384e-04, -1.8813300e-04,  1.4273138e-04, -2.2213216e-04,\n",
      "        3.2233234e-05,  2.3859060e-04, -6.3605828e-04,  1.3870480e-05,\n",
      "        1.2409162e-05,  1.6577833e-05, -1.8720594e-04,  2.7082543e-04,\n",
      "        1.5671180e-04, -5.1454103e-05,  5.0592404e-05,  2.9761362e-04,\n",
      "        5.2252086e-04,  3.7283529e-04,  2.3566268e-04, -3.5549325e-04,\n",
      "        2.3575459e-04, -1.5498104e-04,  1.9331806e-05, -1.4628033e-05,\n",
      "        4.7195208e-04, -4.4923619e-04,  1.2806752e-04, -2.5873596e-04,\n",
      "        2.4620662e-04,  5.2708719e-04, -6.2824751e-05, -2.2741302e-05,\n",
      "       -2.0487063e-05, -3.6139740e-05,  6.4870088e-05,  2.2923687e-04,\n",
      "        2.1856511e-05,  1.4146310e-04,  1.5115377e-04,  1.9536306e-05,\n",
      "        2.0215184e-05, -2.1511042e-03, -7.5250986e-04, -2.7734376e-03,\n",
      "       -1.8227299e-03, -4.5599934e-04,  2.2621304e-03, -1.1992489e-03,\n",
      "       -1.1600677e-03,  6.6923504e-03, -9.1737602e-05, -5.9845287e-04,\n",
      "        1.7491743e-03,  1.1830098e-03, -1.2179809e-03, -2.3288364e-03,\n",
      "        1.2804435e-03,  4.1379945e-04,  1.8602766e-03,  2.9332100e-03,\n",
      "       -1.6034277e-02, -1.7950267e-03,  6.5362238e-04,  2.1068791e-03,\n",
      "        1.0764062e-03, -1.2453184e-03, -7.8735384e-04,  1.1238531e-05,\n",
      "        9.0364683e-06,  1.1392985e-05,  1.0954912e-05,  8.5446536e-06,\n",
      "        9.5849773e-06,  9.1103338e-06,  8.6680329e-06,  8.9316927e-06,\n",
      "        1.0020746e-05,  1.0228859e-05,  9.7393422e-06,  7.8491958e-06,\n",
      "        8.9740015e-06,  8.4838939e-06,  9.2053178e-06,  8.7161870e-06,\n",
      "        1.1509855e-05,  1.5423524e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc400>, <tf.Tensor: id=897771, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 1.33024537e-06, -5.49968888e-07, -9.03686669e-07, ...,\n",
      "        -3.33174867e-05, -2.50646935e-05,  3.67221692e-05],\n",
      "       [ 3.69090685e-07,  2.17967056e-07,  1.69586318e-08, ...,\n",
      "        -3.44930268e-05,  1.51382756e-05,  1.07460401e-05],\n",
      "       [-2.76149990e-07, -6.54737349e-08,  1.98203082e-07, ...,\n",
      "        -4.25882399e-06,  9.04758781e-06, -2.82929595e-05],\n",
      "       ...,\n",
      "       [ 1.15944715e-07,  4.19621244e-08,  1.12743571e-07, ...,\n",
      "        -1.04148626e-06,  3.68934889e-06, -1.29472915e-06],\n",
      "       [ 2.15455143e-09,  1.97418988e-08,  2.12235864e-08, ...,\n",
      "         1.07628664e-04, -5.61197303e-05,  7.11968778e-06],\n",
      "       [ 1.84471418e-07,  1.21929730e-07,  1.18166227e-07, ...,\n",
      "        -1.13966071e-05, -4.02345177e-06, -6.87422926e-07]], dtype=float32)>, <tf.Tensor: id=897772, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 3.9309202e-06,  4.6460368e-06,  6.4405958e-06, ...,\n",
      "         1.2516248e-04, -3.1788859e-05, -3.6595109e-05],\n",
      "       [ 1.4892233e-07,  2.2588634e-07,  5.6880646e-07, ...,\n",
      "         1.4722761e-04, -8.1514872e-06,  1.5793727e-05],\n",
      "       [-6.9119335e-07, -6.8564265e-07, -3.4480450e-07, ...,\n",
      "         2.0999259e-04, -1.8510782e-05,  7.0276095e-05],\n",
      "       ...,\n",
      "       [ 2.0191341e-07,  4.6751765e-08,  2.4008946e-07, ...,\n",
      "         3.5285455e-05, -2.9692587e-06,  6.0865700e-06],\n",
      "       [ 1.2706022e-06, -6.0676016e-06,  3.6191113e-07, ...,\n",
      "         1.3948978e-04, -1.8210427e-05,  6.6726221e-05],\n",
      "       [-8.5039192e-06, -1.1581893e-05,  5.4960224e-06, ...,\n",
      "        -6.2898733e-05,  1.8668035e-05,  7.8866076e-05]], dtype=float32)>, <tf.Tensor: id=897773, shape=(6144,), dtype=float32, numpy=\n",
      "array([-3.2772007e-06, -1.0278363e-06,  4.5191809e-06, ...,\n",
      "        3.5761805e-05,  1.4065318e-04, -6.5051572e-05], dtype=float32)>, <tf.Tensor: id=897676, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.4061932e-07, -8.3888328e-05, -1.0808724e-06, ...,\n",
      "        -1.0208255e-06, -1.4500238e-06, -2.1507458e-06],\n",
      "       [-1.6110296e-06, -1.2441858e-04, -1.9309327e-06, ...,\n",
      "        -1.6553900e-06, -2.6523676e-06, -3.8194075e-06],\n",
      "       [-4.9485141e-07, -1.2719340e-04, -6.0636853e-07, ...,\n",
      "        -6.6362475e-07, -7.0870306e-07, -1.2017113e-06],\n",
      "       ...,\n",
      "       [ 7.4528944e-06,  8.4340770e-04,  7.3135475e-06, ...,\n",
      "         7.8031117e-06,  1.0385250e-05,  1.4518209e-05],\n",
      "       [-8.2775017e-07, -7.7148157e-05, -8.3064612e-07, ...,\n",
      "        -8.6484653e-07, -1.1093873e-06, -1.6849635e-06],\n",
      "       [ 5.2185931e-07, -1.7986605e-04,  7.0787860e-07, ...,\n",
      "         5.1741563e-07,  1.0750324e-06,  1.4065622e-06]], dtype=float32)>, <tf.Tensor: id=897668, shape=(110,), dtype=float32, numpy=\n",
      "array([ 7.64390370e-06,  8.61997309e-04,  7.46757041e-06,  5.67661409e-05,\n",
      "       -5.66436443e-03, -1.26558385e-04, -8.18327768e-04,  9.44272415e-06,\n",
      "        8.95576522e-06,  1.45980130e-05, -1.33606768e-03,  1.12549351e-04,\n",
      "        2.07364559e-04,  2.32336079e-05,  1.43641937e-05,  1.65504101e-03,\n",
      "       -1.63101809e-04, -5.12899598e-04, -1.11177585e-04, -3.58216115e-04,\n",
      "        4.39768424e-04,  2.86709343e-04,  2.74840073e-04,  7.53080676e-05,\n",
      "       -1.65137884e-04,  1.64415585e-04, -8.38028282e-05, -1.41041513e-04,\n",
      "        3.21915402e-04,  6.82621612e-04,  1.63972494e-04,  1.28830325e-05,\n",
      "        1.17013233e-05,  1.52863431e-05,  2.80762877e-04,  5.54943836e-06,\n",
      "        4.39031894e-04, -3.67136672e-04,  5.08994344e-05,  2.59325694e-04,\n",
      "       -1.32137095e-04,  3.73764604e-04, -1.32098910e-04, -1.34545728e-03,\n",
      "        7.68586615e-05,  1.45460246e-04, -3.14097706e-04, -1.70643223e-04,\n",
      "       -6.40043891e-06, -1.69359002e-04, -3.46941670e-04,  5.40890978e-05,\n",
      "        8.16125612e-05,  3.11484153e-04,  9.10070521e-05, -1.82885735e-04,\n",
      "       -1.84519216e-04, -1.11185516e-04, -9.54945135e-05, -1.00043282e-04,\n",
      "        2.21922965e-05,  1.37596959e-04,  1.43640602e-04,  1.77721722e-05,\n",
      "        1.92843727e-05,  8.31562188e-03,  1.24015729e-03,  1.05144933e-03,\n",
      "       -3.31590418e-03,  9.83488280e-04, -2.83545669e-04, -2.06705392e-03,\n",
      "        3.77714494e-03,  1.08366739e-03, -3.19200306e-04, -2.23627011e-03,\n",
      "       -2.91612092e-03,  3.94490315e-03, -5.75240282e-03, -7.11867213e-03,\n",
      "        2.45971605e-03, -6.60830992e-04,  4.79160296e-03,  3.84951336e-03,\n",
      "       -3.87838157e-03,  1.77166180e-03, -7.49165425e-04, -6.02324842e-04,\n",
      "        2.98284926e-04,  8.99445731e-04,  4.60234209e-04,  1.06997823e-05,\n",
      "        8.54485279e-06,  1.04797127e-05,  9.80964887e-06,  7.67469919e-06,\n",
      "        8.98519647e-06,  7.86286364e-06,  7.76425986e-06,  8.14085797e-06,\n",
      "        9.40056361e-06,  9.23815787e-06,  9.27529800e-06,  7.30147303e-06,\n",
      "        8.09691846e-06,  7.72109888e-06,  8.46837884e-06,  7.99120608e-06,\n",
      "        1.06093694e-05,  1.48288418e-05], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc828>, <tf.Tensor: id=898132, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 2.2263330e-06, -3.2739024e-07, -1.3569587e-06, ...,\n",
      "        -2.0080030e-05,  1.6974804e-05,  2.8670180e-05],\n",
      "       [-1.8426172e-07,  1.5220088e-07,  1.9001835e-07, ...,\n",
      "        -2.4696275e-05, -8.1146936e-06,  3.9929123e-06],\n",
      "       [ 2.4341958e-07,  4.3786406e-07, -2.4769253e-07, ...,\n",
      "        -1.6796913e-06,  3.0566145e-05,  1.3322496e-05],\n",
      "       ...,\n",
      "       [ 1.1311903e-06,  5.5226803e-07, -2.9014902e-07, ...,\n",
      "         1.7838012e-06,  1.4236198e-05,  1.4426812e-05],\n",
      "       [ 1.3413858e-08, -7.9511064e-09,  2.3781471e-08, ...,\n",
      "        -1.5501188e-05, -2.5561685e-05, -2.9154628e-05],\n",
      "       [-2.1542931e-07,  1.7959380e-07,  2.6600921e-07, ...,\n",
      "         3.8017854e-05, -6.4588239e-05,  3.1650929e-05]], dtype=float32)>, <tf.Tensor: id=898133, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 6.6433417e-06,  3.8104176e-06,  1.1129166e-05, ...,\n",
      "         6.6537008e-04, -9.9522651e-05,  1.0329481e-04],\n",
      "       [-9.0684341e-07, -1.3540720e-06, -8.1537212e-07, ...,\n",
      "         6.8197856e-05, -2.6525899e-05,  1.7166576e-06],\n",
      "       [ 1.0319920e-07,  6.5978935e-07, -1.3513775e-06, ...,\n",
      "        -1.2950288e-04,  2.3286435e-05, -4.9287348e-05],\n",
      "       ...,\n",
      "       [ 4.3127940e-07,  5.0787651e-07, -3.1253034e-08, ...,\n",
      "        -2.5078931e-05,  2.0632272e-06, -6.8817685e-06],\n",
      "       [ 2.4727174e-06, -1.1432120e-05, -4.6151885e-07, ...,\n",
      "         1.6461554e-04, -3.1961870e-05,  2.4555882e-05],\n",
      "       [ 3.3709741e-06,  3.7821405e-06,  1.1651907e-06, ...,\n",
      "        -2.9445055e-04,  2.3895931e-05,  2.5886875e-06]], dtype=float32)>, <tf.Tensor: id=898134, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 1.9144459e-06,  7.2996972e-06,  1.6495145e-06, ...,\n",
      "       -2.5677393e-05,  1.6630226e-04, -2.9694976e-04], dtype=float32)>, <tf.Tensor: id=898037, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.1906713e-07, -3.2275402e-05, -1.0439453e-06, ...,\n",
      "        -9.7653287e-07, -1.4044390e-06, -2.0684918e-06],\n",
      "       [-1.5126452e-06, -1.2937683e-04, -1.8387594e-06, ...,\n",
      "        -1.5140320e-06, -2.5203528e-06, -3.7066723e-06],\n",
      "       [-5.9458262e-07,  1.0140295e-04, -6.6523711e-07, ...,\n",
      "        -7.4645038e-07, -8.2507012e-07, -1.2146287e-06],\n",
      "       ...,\n",
      "       [ 9.0461526e-06,  4.1413773e-04,  8.2988900e-06, ...,\n",
      "         9.2276887e-06,  1.1958563e-05,  1.6129405e-05],\n",
      "       [-9.3093564e-07, -4.8232232e-05, -9.0248142e-07, ...,\n",
      "        -9.7665338e-07, -1.1991383e-06, -1.7944101e-06],\n",
      "       [ 2.5876165e-07,  1.1059899e-05,  5.2416141e-07, ...,\n",
      "         2.3902928e-07,  7.6929837e-07,  1.1363759e-06]], dtype=float32)>, <tf.Tensor: id=898029, shape=(110,), dtype=float32, numpy=\n",
      "array([ 9.36330798e-06,  4.23858117e-04,  8.53661822e-06,  8.28472548e-05,\n",
      "        4.37811809e-03,  3.30288713e-05,  4.93535190e-04,  1.10603560e-05,\n",
      "        1.08797540e-05,  1.59856700e-05,  7.29277963e-05, -1.76912596e-04,\n",
      "        9.95713417e-05,  2.64378614e-05,  1.70121621e-05,  1.23706844e-03,\n",
      "       -8.60010681e-04,  2.87720090e-04,  4.85557102e-05,  1.50627195e-04,\n",
      "       -5.27013384e-04,  6.07747468e-04, -1.67613442e-04,  4.13929694e-04,\n",
      "        5.03083516e-04, -9.11045878e-04,  1.63321587e-04, -9.88116153e-05,\n",
      "       -3.01299238e-04, -8.91629606e-05, -9.85850565e-05,  1.52441071e-05,\n",
      "        1.36754088e-05,  1.81111045e-05,  1.37710915e-04,  7.98042398e-04,\n",
      "        2.63123016e-04,  2.74108315e-04,  5.11193066e-05, -4.99425863e-04,\n",
      "        3.36995523e-04,  4.27471350e-05, -1.22255631e-04,  3.66062828e-04,\n",
      "        2.33657658e-04,  3.05825350e-04,  3.11097130e-04,  6.62678271e-04,\n",
      "        3.29109404e-04,  3.59818485e-04,  4.76122310e-04,  5.81302411e-05,\n",
      "       -2.12039551e-04,  2.90524331e-04, -7.01993122e-04,  1.35671784e-04,\n",
      "       -1.89168550e-05,  2.28665915e-04,  6.71705784e-05,  6.05701825e-05,\n",
      "        2.39313667e-05,  1.41852070e-04,  1.54311972e-04,  2.11151146e-05,\n",
      "       -1.34852962e-04,  8.29817727e-03, -1.74685137e-03, -3.82229732e-03,\n",
      "       -1.10367590e-04, -7.15948455e-03,  8.87991395e-04,  1.55992270e-03,\n",
      "        3.66328540e-03, -1.18641714e-02,  8.60645887e-05, -7.05127197e-04,\n",
      "       -5.36746345e-03,  2.56977510e-03, -2.14884523e-03, -4.97788773e-04,\n",
      "        1.74994988e-03, -1.72852087e-04, -3.16900248e-03, -3.48716020e-03,\n",
      "        4.20275098e-03,  2.61738501e-03,  1.19538279e-03, -8.99408886e-04,\n",
      "        9.76845273e-04,  2.84920633e-03, -6.44932734e-05,  1.25086144e-05,\n",
      "        9.83482641e-06,  1.27825915e-05,  1.22919009e-05,  9.45423199e-06,\n",
      "        1.09155226e-05,  1.01196883e-05,  9.51195580e-06,  9.86931263e-06,\n",
      "        1.07425385e-05,  1.11335949e-05,  1.09003786e-05,  8.70023814e-06,\n",
      "        9.92364130e-06,  9.19619379e-06,  9.77343552e-06,  9.53988638e-06,\n",
      "        1.23124910e-05,  1.65942947e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc320>, <tf.Tensor: id=898493, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 2.61217815e-06,  1.08969644e-07, -1.91304343e-06, ...,\n",
      "        -8.01300412e-05,  6.17904880e-05,  3.84110826e-05],\n",
      "       [ 4.24452423e-07,  1.50505088e-07,  7.01868004e-08, ...,\n",
      "        -6.24085897e-06,  4.92978907e-05,  6.22442030e-06],\n",
      "       [-1.87565249e-07, -1.76401016e-07,  9.53604840e-08, ...,\n",
      "        -6.15010595e-06,  5.00342467e-05,  1.13732667e-05],\n",
      "       ...,\n",
      "       [ 1.31136335e-06,  3.88100602e-07, -3.37746116e-07, ...,\n",
      "         2.15326554e-06,  1.02027780e-05,  5.67641609e-06],\n",
      "       [ 7.58659979e-08,  4.09039558e-08, -6.55535404e-09, ...,\n",
      "         1.09283792e-05, -5.64398615e-05, -4.74828521e-05],\n",
      "       [-9.88944961e-08,  8.22564701e-08,  8.85762006e-08, ...,\n",
      "         4.03785598e-05, -2.29936522e-05,  2.49190343e-05]], dtype=float32)>, <tf.Tensor: id=898494, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 1.10553628e-05,  3.53334917e-06,  2.84300131e-06, ...,\n",
      "         5.48324198e-04, -1.03772822e-04,  5.06036740e-05],\n",
      "       [-7.01767647e-07, -2.43384852e-06, -1.50674691e-06, ...,\n",
      "        -3.56423116e-04,  5.23629533e-05,  1.80810403e-05],\n",
      "       [-5.49786478e-07, -6.62123853e-07, -1.73194394e-06, ...,\n",
      "         1.10214256e-04, -3.88558828e-06, -1.56743881e-05],\n",
      "       ...,\n",
      "       [ 5.02218214e-08, -5.92228275e-08, -4.20662673e-07, ...,\n",
      "        -2.74994291e-05,  3.63709023e-06, -2.61523064e-06],\n",
      "       [ 2.06866218e-07, -8.72469445e-06,  1.41246892e-06, ...,\n",
      "         2.31765880e-05, -9.43931536e-06,  5.30592879e-05],\n",
      "       [-6.94565017e-07,  1.12734620e-06,  6.93292395e-06, ...,\n",
      "         6.16707621e-05, -8.85747431e-06,  5.42668886e-06]], dtype=float32)>, <tf.Tensor: id=898495, shape=(6144,), dtype=float32, numpy=\n",
      "array([-9.5954183e-06,  2.4837925e-06,  4.1098901e-06, ...,\n",
      "       -2.8121372e-05,  2.3945708e-05,  5.9849317e-05], dtype=float32)>, <tf.Tensor: id=898398, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.4510805e-07,  1.2317364e-04, -1.0057313e-06, ...,\n",
      "        -9.3831778e-07, -1.3411966e-06, -2.0252662e-06],\n",
      "       [-1.5831927e-06,  1.3859553e-05, -1.8903165e-06, ...,\n",
      "        -1.6183776e-06, -2.5836225e-06, -3.8280459e-06],\n",
      "       [-4.4040524e-07,  3.0006777e-04, -5.7077170e-07, ...,\n",
      "        -6.3391690e-07, -7.0735507e-07, -1.0440967e-06],\n",
      "       ...,\n",
      "       [ 6.6410821e-06, -9.7133961e-05,  6.6740663e-06, ...,\n",
      "         7.1115028e-06,  9.5144387e-06,  1.3528557e-05],\n",
      "       [-7.5504022e-07,  6.1028022e-05, -7.8575306e-07, ...,\n",
      "        -8.2402005e-07, -1.0484707e-06, -1.6181236e-06],\n",
      "       [ 6.1164195e-07,  5.6851699e-05,  7.8813770e-07, ...,\n",
      "         6.0741758e-07,  1.1097327e-06,  1.6081872e-06]], dtype=float32)>, <tf.Tensor: id=898390, shape=(110,), dtype=float32, numpy=\n",
      "array([ 6.80914945e-06, -8.83738976e-05,  6.81258371e-06,  6.08556322e-04,\n",
      "        3.83904041e-03,  2.96174057e-05, -6.60011719e-04,  8.44591978e-06,\n",
      "        8.20283549e-06,  1.34498850e-05, -3.89884645e-03,  4.97226574e-06,\n",
      "        7.80573464e-05,  2.23649276e-05,  1.39667063e-05, -5.99428837e-04,\n",
      "        8.64693371e-04, -1.00657449e-03,  4.43673052e-05,  2.74947932e-04,\n",
      "        9.67251486e-04, -3.57843455e-05,  1.40050775e-04,  9.92467249e-05,\n",
      "       -4.90589824e-04,  2.34007421e-05,  4.93120460e-05,  7.86865203e-05,\n",
      "       -1.46533712e-04,  2.04335956e-04, -2.27868732e-04,  1.19182469e-05,\n",
      "        1.08894819e-05,  1.46300608e-05, -1.77428330e-04,  2.18307614e-04,\n",
      "       -2.73987593e-04, -7.63499775e-05,  2.22969793e-05,  5.70402306e-04,\n",
      "        1.24932762e-04,  3.20523686e-04,  1.46988154e-04, -4.13653179e-04,\n",
      "        5.54755388e-05,  1.21692960e-04,  1.12388414e-04,  9.95308073e-06,\n",
      "        4.51640692e-04,  1.77000940e-04,  2.97503517e-04,  5.28449891e-05,\n",
      "       -9.16937934e-05,  6.53663301e-05, -7.42498669e-04, -3.30910807e-05,\n",
      "       -1.90936407e-04,  1.10002438e-05,  5.74790247e-05,  1.87538884e-04,\n",
      "        2.15222644e-05,  1.29217879e-04,  1.39081254e-04,  1.66640657e-05,\n",
      "        1.78161972e-05,  5.50095830e-03,  1.12951500e-03,  5.16530243e-04,\n",
      "        1.82575756e-03, -6.21433649e-03,  2.66950764e-03,  1.55094755e-03,\n",
      "       -2.80561484e-03, -9.65083251e-04,  6.11980679e-04, -1.57513260e-03,\n",
      "       -3.23031703e-03,  3.51128401e-03, -1.28271151e-03, -9.03779233e-04,\n",
      "        2.33721198e-03, -1.67241800e-04, -1.40295736e-03, -1.08958585e-02,\n",
      "        8.32830835e-03, -7.24496902e-04, -9.58880875e-04,  1.56587863e-03,\n",
      "        3.80223442e-04, -4.24228900e-04,  1.78596427e-04,  9.75492731e-06,\n",
      "        7.76848174e-06,  9.86174382e-06,  8.81792585e-06,  7.05565071e-06,\n",
      "        8.21434514e-06,  7.26556209e-06,  6.98010263e-06,  7.50146773e-06,\n",
      "        8.47762749e-06,  8.43518501e-06,  8.73429417e-06,  6.52713834e-06,\n",
      "       -3.05143738e-04,  7.09827827e-06,  7.74816726e-06,  7.28609393e-06,\n",
      "        9.72504586e-06,  1.38159703e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc668>, <tf.Tensor: id=898854, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-1.82045233e-06, -1.28720671e-06,  1.66459938e-06, ...,\n",
      "        -3.45409558e-06,  5.74055730e-06,  4.98481095e-05],\n",
      "       [ 9.34498530e-07,  3.97028344e-07, -1.10735414e-07, ...,\n",
      "         2.88280285e-06,  3.38979771e-05, -3.43950683e-06],\n",
      "       [-9.32558706e-08,  2.61745129e-07, -1.81405454e-07, ...,\n",
      "        -2.50478161e-06,  3.49140646e-05,  1.86539055e-05],\n",
      "       ...,\n",
      "       [-1.02294528e-07, -2.35692340e-07, -1.16704832e-07, ...,\n",
      "        -5.79615460e-07,  1.18151356e-05,  1.55504167e-05],\n",
      "       [-1.40261553e-08,  7.88536028e-08,  1.15588534e-08, ...,\n",
      "        -6.88598739e-05, -6.86012136e-05, -6.77562566e-05],\n",
      "       [ 4.57835938e-07,  2.34493839e-07, -1.60053688e-07, ...,\n",
      "         3.20766485e-05, -3.23113018e-05,  2.63574020e-05]], dtype=float32)>, <tf.Tensor: id=898855, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-1.07404458e-05, -1.81224023e-05, -9.59292083e-06, ...,\n",
      "        -1.80492876e-04,  3.45362096e-05, -1.75456580e-06],\n",
      "       [-6.14618784e-07, -1.84221506e-06,  1.56306434e-08, ...,\n",
      "        -3.02507542e-04,  4.37427734e-05, -5.37339292e-05],\n",
      "       [ 3.96753933e-07,  9.50382457e-07,  1.27515619e-07, ...,\n",
      "         2.34926665e-05, -2.74395079e-05,  1.65887450e-05],\n",
      "       ...,\n",
      "       [-8.84595153e-08,  2.16782595e-07, -1.83083159e-06, ...,\n",
      "        -1.75727509e-05,  1.51612778e-06, -6.86550675e-06],\n",
      "       [ 5.40454630e-06,  8.01344322e-06,  8.37430525e-06, ...,\n",
      "        -2.24357776e-04,  2.55445502e-05, -3.34845899e-05],\n",
      "       [-1.20257873e-05, -3.19004030e-05,  1.56034876e-05, ...,\n",
      "         1.07596774e-04, -3.29379509e-05,  1.61557953e-04]], dtype=float32)>, <tf.Tensor: id=898856, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 4.5794925e-05,  1.1323302e-06, -1.0949698e-06, ...,\n",
      "       -1.7622897e-05, -2.2839569e-04,  1.1185200e-04], dtype=float32)>, <tf.Tensor: id=898759, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.0529167e-06,  5.7377649e-05, -1.1581178e-06, ...,\n",
      "        -1.1435216e-06, -1.5184663e-06, -2.2792829e-06],\n",
      "       [-1.3292374e-06, -3.0474388e-05, -1.6894395e-06, ...,\n",
      "        -1.3348031e-06, -2.2774509e-06, -3.5268295e-06],\n",
      "       [-8.4036157e-07,  4.1060193e-05, -8.7191933e-07, ...,\n",
      "        -1.0603455e-06, -1.0859118e-06, -1.5737767e-06],\n",
      "       ...,\n",
      "       [ 1.0213181e-05, -7.5262156e-04,  9.3728058e-06, ...,\n",
      "         1.0987305e-05,  1.3041999e-05,  1.8027367e-05],\n",
      "       [-1.0353249e-06,  1.5349902e-04, -9.9451313e-07, ...,\n",
      "        -1.1271399e-06, -1.3206237e-06, -1.9690510e-06],\n",
      "       [ 6.4031553e-08, -1.5230378e-04,  3.7426992e-07, ...,\n",
      "        -2.6088856e-09,  5.5804219e-07,  1.0046230e-06]], dtype=float32)>, <tf.Tensor: id=898751, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.09353696e-05, -7.83662719e-04,  9.92950572e-06,  1.42271328e-03,\n",
      "       -3.79747408e-03,  3.67939138e-05,  9.99660813e-04,  1.25225197e-05,\n",
      "        1.27084386e-05,  1.70096046e-05,  8.25702853e-04,  3.26287554e-04,\n",
      "        4.30665183e-04,  2.88872307e-05,  1.91254476e-05, -1.72147073e-03,\n",
      "       -7.74215732e-05, -1.75291824e-03, -4.16525378e-04, -8.93593882e-04,\n",
      "        9.12055402e-05, -1.53434084e-04, -1.59244315e-04, -2.23414841e-04,\n",
      "        4.65506659e-04, -2.82418565e-04, -2.16485525e-04,  2.89868040e-04,\n",
      "        2.16059401e-04, -1.49006359e-04,  2.61826237e-04,  1.78476912e-05,\n",
      "        1.58925432e-05,  2.02750562e-05,  2.88669311e-04, -8.43245725e-05,\n",
      "       -2.80163746e-04, -6.68307985e-05, -4.37266659e-04, -5.01858245e-04,\n",
      "       -5.05292497e-04, -6.36122888e-04, -1.36208037e-04,  4.15191636e-04,\n",
      "        5.54956787e-05, -3.63194995e-05, -3.60079255e-04,  6.50143134e-04,\n",
      "       -3.09202529e-04,  5.45920280e-04, -1.77135094e-04, -4.09662840e-04,\n",
      "       -1.16822054e-03, -4.09825443e-04, -8.18186090e-04, -1.81098687e-04,\n",
      "       -2.09920345e-05, -9.84800427e-05,  6.78114957e-05, -1.09587883e-04,\n",
      "        2.57501106e-05, -2.30322985e-05, -3.77681499e-06,  2.43716622e-05,\n",
      "        2.28273420e-05, -1.12803525e-03,  3.15932138e-03,  3.47426464e-03,\n",
      "       -3.74096632e-03, -9.73690185e-04,  5.94912388e-04,  7.57082831e-04,\n",
      "        8.62721354e-04, -3.86026991e-03,  3.05444955e-05, -2.57823849e-04,\n",
      "       -5.91997057e-03, -6.61281636e-04, -7.54919776e-04,  1.83652190e-03,\n",
      "        2.40362226e-03, -4.74921282e-04, -3.33448336e-03, -2.56347843e-03,\n",
      "        1.86021421e-02,  1.20271754e-03,  2.91730888e-04,  1.46523162e-04,\n",
      "       -1.08524261e-03,  6.72579976e-04,  2.48249300e-04,  1.45438644e-05,\n",
      "        1.14242675e-05,  1.51637769e-05,  1.51153508e-05,  1.10642823e-05,\n",
      "        1.33832773e-05,  1.26497762e-05,  1.14661652e-05,  1.20004024e-05,\n",
      "        1.27674984e-05,  1.32530913e-05,  1.27313806e-05,  1.06686630e-05,\n",
      "        1.23211867e-05,  1.08842132e-05,  1.13294582e-05,  1.17430318e-05,\n",
      "        1.37936577e-05,  1.89836228e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc748>, <tf.Tensor: id=899215, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-5.6659937e-06, -8.7667570e-07,  2.8406002e-06, ...,\n",
      "        -6.3489351e-05,  4.4683558e-05,  3.0462375e-05],\n",
      "       [ 7.0728021e-08,  3.0789712e-07,  2.5956953e-07, ...,\n",
      "         7.7770337e-06,  3.5435753e-06,  4.4200960e-06],\n",
      "       [-2.5551029e-07,  7.8110439e-08,  1.0178158e-07, ...,\n",
      "         1.4777704e-06, -4.8548100e-06,  8.3502027e-06],\n",
      "       ...,\n",
      "       [ 1.3252892e-06,  4.4753068e-07, -4.8128578e-07, ...,\n",
      "        -8.7086155e-06, -2.8473428e-06,  1.6930517e-07],\n",
      "       [-3.1697876e-08,  2.6492033e-09,  5.7010276e-08, ...,\n",
      "        -1.8204462e-05, -1.1980524e-05,  1.1402110e-05],\n",
      "       [-4.0536611e-07,  1.2394329e-07,  3.3055863e-07, ...,\n",
      "        -3.6922884e-06, -2.0956180e-05,  1.0165936e-05]], dtype=float32)>, <tf.Tensor: id=899216, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-8.4899830e-06, -5.9769332e-06, -3.4164855e-06, ...,\n",
      "        -9.0655073e-04,  1.2584728e-04, -1.5089885e-04],\n",
      "       [ 4.3059913e-07,  4.8646825e-07,  2.0389614e-08, ...,\n",
      "         8.6015090e-05,  6.7625879e-06, -2.1725231e-05],\n",
      "       [ 4.6170052e-07,  3.9756671e-07,  2.1058204e-06, ...,\n",
      "         3.4857570e-05, -1.2913864e-05, -6.4957443e-05],\n",
      "       ...,\n",
      "       [-2.0767247e-07,  8.8049364e-07, -8.9963675e-07, ...,\n",
      "        -8.4152098e-06,  1.4106454e-06, -5.3129870e-06],\n",
      "       [ 7.6960960e-06,  1.6728409e-05,  1.6340120e-05, ...,\n",
      "        -1.0804670e-04,  1.6763741e-05,  4.7002681e-07],\n",
      "       [ 9.6266640e-06,  9.9576064e-06, -6.0019320e-06, ...,\n",
      "        -3.6275806e-04,  2.1379339e-05,  2.9582348e-05]], dtype=float32)>, <tf.Tensor: id=899217, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 2.5267322e-05, -3.5154419e-06, -7.1089858e-06, ...,\n",
      "       -8.3686200e-06, -1.1031458e-04, -3.7304254e-04], dtype=float32)>, <tf.Tensor: id=899120, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.3210656e-07,  2.5799783e-04, -1.0268619e-06, ...,\n",
      "        -1.0035989e-06, -1.3852699e-06, -2.1142810e-06],\n",
      "       [-1.5241276e-06,  1.6070597e-04, -1.7392158e-06, ...,\n",
      "        -1.4825888e-06, -2.4460639e-06, -3.6646752e-06],\n",
      "       [-3.8219298e-07, -1.1253231e-04, -5.2301658e-07, ...,\n",
      "        -6.2504103e-07, -5.8746525e-07, -8.8605719e-07],\n",
      "       ...,\n",
      "       [ 9.6378699e-06, -3.9050984e-03,  8.1885191e-06, ...,\n",
      "         9.9278941e-06,  1.2259436e-05,  1.6975799e-05],\n",
      "       [-1.0506050e-06,  4.0052098e-04, -9.3599783e-07, ...,\n",
      "        -1.0953120e-06, -1.3028507e-06, -1.9303357e-06],\n",
      "       [ 2.2743059e-07,  9.2318095e-04,  5.5707181e-07, ...,\n",
      "         1.8960949e-07,  7.7264338e-07,  1.1643309e-06]], dtype=float32)>, <tf.Tensor: id=899112, shape=(110,), dtype=float32, numpy=\n",
      "array([ 9.98056566e-06, -4.03346866e-03,  8.42976988e-06,  1.99766015e-03,\n",
      "       -5.51131612e-04,  3.34632678e-05,  1.72895088e-04,  1.17190775e-05,\n",
      "        1.13719079e-05,  1.59751762e-05, -3.35780234e-04,  3.42854852e-04,\n",
      "        3.97807191e-04,  2.76434184e-05,  1.89979801e-05,  3.30989528e-03,\n",
      "       -1.16068311e-03, -4.96955961e-03,  5.03077354e-05,  1.55541184e-05,\n",
      "       -2.72613252e-04,  7.81011942e-04,  4.73892607e-04,  5.02764073e-04,\n",
      "        2.15298176e-04, -4.68034879e-04,  1.95896777e-04,  4.90931707e-05,\n",
      "        3.18469771e-04, -8.46801595e-06,  3.08226736e-04,  1.66471164e-05,\n",
      "        1.47356077e-05,  1.87456444e-05, -2.31736522e-05,  3.69889895e-05,\n",
      "       -3.49164111e-06,  2.83716072e-04, -2.51939986e-04,  2.29005556e-04,\n",
      "        1.77940106e-04,  3.02955974e-04,  3.06777889e-04,  1.02242595e-03,\n",
      "       -5.61773079e-04,  1.37666706e-04,  1.32543864e-04, -2.35408443e-05,\n",
      "        5.29693556e-04,  4.62873395e-05, -7.66214507e-05,  6.22535517e-05,\n",
      "       -4.59302682e-04,  1.82511954e-04,  3.49939801e-05,  1.39314594e-04,\n",
      "        1.46693987e-04,  1.14957838e-05,  7.04942504e-05,  7.30861357e-05,\n",
      "        2.58198143e-05, -3.23137356e-04, -3.30201961e-04,  2.25081512e-05,\n",
      "        2.11146471e-05, -7.81312399e-03,  4.10272245e-04, -1.78403128e-03,\n",
      "        2.81094946e-03,  7.28789577e-03,  9.02661821e-04,  6.81456993e-04,\n",
      "       -4.28597815e-03,  4.46271105e-03, -1.87480706e-04, -1.49753178e-03,\n",
      "       -4.38348856e-03,  1.63016561e-03,  6.27606036e-03, -1.83544669e-03,\n",
      "       -1.25102769e-03, -3.15128156e-04, -9.17567522e-04, -3.58258653e-03,\n",
      "        5.38367266e-03,  2.13506771e-03,  6.00383733e-04, -5.10927988e-03,\n",
      "        7.33022898e-06,  3.54895776e-04,  3.35241348e-04,  1.22260599e-05,\n",
      "        1.04852170e-05,  1.43389998e-05,  1.34882966e-05,  1.01959313e-05,\n",
      "        1.14060595e-05,  1.11575146e-05,  1.05449935e-05,  1.11362033e-05,\n",
      "        1.15856556e-05,  1.19944116e-05,  1.15704570e-05,  9.25587301e-06,\n",
      "        1.10296696e-05,  9.94090078e-06,  1.04627770e-05,  1.02701570e-05,\n",
      "        1.26393870e-05,  1.74743473e-05], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc6a0>, <tf.Tensor: id=899576, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-8.2381342e-07,  9.4139875e-07,  7.3331535e-07, ...,\n",
      "        -5.7580746e-05, -2.4101577e-05,  8.4441872e-06],\n",
      "       [ 5.8174265e-07,  8.1297607e-08,  1.0682929e-07, ...,\n",
      "        -5.3092816e-05, -2.4264150e-06, -7.7975074e-06],\n",
      "       [-5.3187495e-08,  1.3825731e-07,  7.6740385e-08, ...,\n",
      "        -1.5469463e-05, -1.7459937e-05, -2.8405417e-05],\n",
      "       ...,\n",
      "       [ 7.1809689e-07,  2.7688759e-07, -1.3812887e-07, ...,\n",
      "        -6.5442664e-06,  1.0377055e-06, -3.5111357e-06],\n",
      "       [ 8.7202082e-08,  8.0422900e-08, -2.4928465e-08, ...,\n",
      "        -6.0162136e-05,  1.0814681e-05,  1.8923492e-05],\n",
      "       [ 6.3591460e-07,  5.1452265e-08,  3.6551117e-07, ...,\n",
      "         1.4768156e-05, -1.8852365e-05, -5.4938071e-05]], dtype=float32)>, <tf.Tensor: id=899577, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-3.91440921e-07, -6.06702997e-06, -8.23350945e-07, ...,\n",
      "        -5.40753936e-05,  1.42273593e-05,  2.39052242e-05],\n",
      "       [-7.71949033e-07, -3.96404687e-09, -1.28660690e-06, ...,\n",
      "         3.31440569e-05,  1.57412778e-05, -1.25204009e-04],\n",
      "       [-5.20925028e-07, -4.12875636e-07, -4.65943231e-07, ...,\n",
      "        -8.24439412e-05,  1.25771257e-05,  6.42562736e-05],\n",
      "       ...,\n",
      "       [ 1.94843693e-07,  2.43461216e-07, -8.66322239e-07, ...,\n",
      "        -1.42218805e-05,  2.61625951e-06, -5.69035046e-06],\n",
      "       [ 5.95357778e-06,  4.58081286e-06,  1.07754986e-05, ...,\n",
      "        -9.46452110e-06, -2.17554043e-06, -4.50649422e-05],\n",
      "       [-2.13568928e-05, -4.24841382e-05,  2.73644346e-05, ...,\n",
      "         2.89985212e-04, -6.20255742e-05,  1.22236568e-04]], dtype=float32)>, <tf.Tensor: id=899578, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 3.6813144e-06,  2.5365271e-06, -5.5220653e-08, ...,\n",
      "       -1.4439592e-05, -6.7343644e-06,  2.9256078e-04], dtype=float32)>, <tf.Tensor: id=899481, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.5561396e-07,  1.2510053e-04, -9.7581096e-07, ...,\n",
      "        -9.2250059e-07, -1.3093959e-06, -1.9415982e-06],\n",
      "       [-1.5201024e-06,  5.5876932e-05, -1.7426877e-06, ...,\n",
      "        -1.5011111e-06, -2.4072669e-06, -3.5039563e-06],\n",
      "       [-3.6958463e-07, -1.0528530e-04, -4.8884561e-07, ...,\n",
      "        -5.6981202e-07, -5.9972865e-07, -8.7709475e-07],\n",
      "       ...,\n",
      "       [ 6.9404368e-06, -9.6551079e-04,  6.5124314e-06, ...,\n",
      "         7.2292651e-06,  9.5875566e-06,  1.3260651e-05],\n",
      "       [-8.0157338e-07,  1.2691262e-04, -7.7498032e-07, ...,\n",
      "        -8.4798626e-07, -1.0546420e-06, -1.5811972e-06],\n",
      "       [ 5.7676533e-07, -1.2831992e-04,  7.7984737e-07, ...,\n",
      "         5.2625194e-07,  1.0726568e-06,  1.5237268e-06]], dtype=float32)>, <tf.Tensor: id=899473, shape=(110,), dtype=float32, numpy=\n",
      "array([ 7.13672125e-06, -9.64258215e-04,  6.66375354e-06, -2.19027657e-04,\n",
      "        5.23965980e-04,  2.99198637e-05,  1.23089994e-03,  8.70759050e-06,\n",
      "        8.44222450e-06,  1.33439098e-05, -9.77093587e-04,  1.60193202e-04,\n",
      "       -1.29598280e-04,  2.24873165e-05,  1.48215104e-05,  7.10854947e-04,\n",
      "       -8.84914363e-04, -1.69982063e-03, -1.11824091e-04,  3.75272466e-05,\n",
      "        8.76831182e-04,  1.21775163e-04,  2.83909554e-04,  5.74896039e-05,\n",
      "        3.65872023e-04, -4.52658307e-04, -1.76825284e-04,  2.33805229e-04,\n",
      "        5.66786039e-04, -4.94316104e-04,  6.45327324e-04,  1.23823284e-05,\n",
      "        1.11302143e-05,  1.50893966e-05, -2.79831384e-05, -4.76733549e-04,\n",
      "        1.64027617e-04,  3.62743711e-04,  1.21190678e-05,  2.72941572e-04,\n",
      "        2.45980656e-04,  1.13862145e-04,  7.74972250e-06,  4.94680658e-04,\n",
      "        3.46987617e-05,  2.47770484e-04, -5.53977385e-04,  4.54638153e-04,\n",
      "        1.07407206e-04,  2.00211114e-04,  4.58650466e-04,  5.54220787e-05,\n",
      "       -2.16167027e-04,  1.07725791e-05,  3.67864777e-05,  1.16352407e-04,\n",
      "        1.24849961e-04,  2.31956146e-04,  5.87544419e-05,  2.01490657e-05,\n",
      "       -1.34121379e-04,  1.26323808e-04,  1.29230582e-04,  1.63611985e-05,\n",
      "        1.75202713e-05, -2.15529371e-03, -2.97230389e-03, -3.80198495e-03,\n",
      "        1.94393433e-04, -1.66201242e-03,  6.34984521e-04,  2.04087445e-03,\n",
      "       -2.90496089e-03, -8.65068752e-04, -3.12773220e-04, -5.67423413e-04,\n",
      "       -1.71747117e-03,  3.48023896e-04, -5.40186709e-04,  1.99509040e-03,\n",
      "       -4.48408723e-03,  1.95871224e-04,  3.34274070e-03,  1.84217771e-03,\n",
      "        5.29514067e-03, -1.33359432e-03,  2.69531761e-03, -1.15385861e-03,\n",
      "       -1.63454199e-04,  3.33907595e-03,  1.35848008e-04,  9.56169697e-06,\n",
      "        7.87079171e-06,  1.04823794e-05,  9.19569447e-06,  7.51207699e-06,\n",
      "        8.42545887e-06,  7.69535109e-06,  7.28488021e-06,  7.88277066e-06,\n",
      "        8.57883970e-06,  8.64031972e-06,  8.95703579e-06, -1.49484680e-04,\n",
      "        8.09071389e-06,  7.34432797e-06,  8.05441687e-06,  7.43028932e-06,\n",
      "        9.82741312e-06,  1.35748733e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc278>, <tf.Tensor: id=899937, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-5.2558607e-07,  9.3121258e-08,  1.9084371e-06, ...,\n",
      "        -2.6254891e-05, -4.3829164e-05, -1.8034381e-05],\n",
      "       [ 7.2852821e-08,  7.4620061e-08,  1.2984124e-07, ...,\n",
      "        -3.7804337e-05,  1.7621729e-05,  2.1267067e-05],\n",
      "       [-1.6524353e-07, -2.1629231e-07,  3.5710819e-08, ...,\n",
      "        -1.5171647e-05, -6.2764761e-05, -5.9670791e-05],\n",
      "       ...,\n",
      "       [ 3.3926887e-07,  4.2604816e-07, -2.9411942e-07, ...,\n",
      "         7.8790163e-06, -2.1338381e-05, -1.8264022e-07],\n",
      "       [-4.5264102e-08, -2.7071399e-08,  4.4180613e-08, ...,\n",
      "        -7.9516351e-05,  5.3202926e-05, -2.3389877e-05],\n",
      "       [ 2.3796109e-07,  3.2622456e-08,  1.2916362e-07, ...,\n",
      "        -4.4959786e-05, -8.7002154e-06,  5.9373083e-06]], dtype=float32)>, <tf.Tensor: id=899938, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-7.5692037e-06, -1.2553894e-05, -4.4479637e-07, ...,\n",
      "        -3.3410385e-04,  3.4574245e-05, -4.4950957e-06],\n",
      "       [-7.8405878e-07,  2.7654957e-07, -1.9271491e-07, ...,\n",
      "         1.9321508e-04, -3.0345058e-05,  1.4334771e-05],\n",
      "       [ 2.7362870e-07,  3.5332664e-07,  4.4361613e-07, ...,\n",
      "         9.3400944e-05, -9.9779245e-06, -2.1695772e-05],\n",
      "       ...,\n",
      "       [ 3.4351658e-07,  7.8770256e-07,  5.2463287e-07, ...,\n",
      "         9.5476189e-06, -1.3932613e-07, -1.7009037e-07],\n",
      "       [ 9.7725197e-06,  1.6131555e-05,  3.1641619e-05, ...,\n",
      "        -2.2577628e-04,  1.6919050e-05, -9.7219541e-05],\n",
      "       [ 1.4011900e-05,  3.0530766e-05, -2.7414657e-05, ...,\n",
      "        -4.4933063e-04,  6.4319538e-05,  4.7980033e-05]], dtype=float32)>, <tf.Tensor: id=899939, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 2.7983415e-05,  1.6898132e-06,  1.0056788e-06, ...,\n",
      "        1.0121731e-05, -2.2691979e-04, -4.5798993e-04], dtype=float32)>, <tf.Tensor: id=899842, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.8066548e-07,  1.1405334e-04, -9.8978308e-07, ...,\n",
      "        -9.4255336e-07, -1.3389169e-06, -1.9386448e-06],\n",
      "       [-1.3680810e-06, -9.4159412e-05, -1.6270149e-06, ...,\n",
      "        -1.3520935e-06, -2.2517816e-06, -3.2713106e-06],\n",
      "       [-4.6158186e-07,  9.7913013e-05, -5.6377382e-07, ...,\n",
      "        -6.6041804e-07, -6.8411168e-07, -9.9748308e-07],\n",
      "       ...,\n",
      "       [ 8.0259470e-06, -9.3876442e-04,  7.2365033e-06, ...,\n",
      "         8.3056002e-06,  1.0779041e-05,  1.4205601e-05],\n",
      "       [-8.3832384e-07,  1.2410898e-04, -8.0442305e-07, ...,\n",
      "        -8.8329148e-07, -1.1080797e-06, -1.5884956e-06],\n",
      "       [ 3.8545033e-07,  7.9225749e-05,  6.3511015e-07, ...,\n",
      "         3.5133775e-07,  9.0046888e-07,  1.3086928e-06]], dtype=float32)>, <tf.Tensor: id=899834, shape=(110,), dtype=float32, numpy=\n",
      "array([ 8.29082546e-06, -1.04593707e-03,  7.43083956e-06, -2.92437762e-05,\n",
      "       -9.05807968e-03,  3.17510057e-05,  5.81366068e-04,  9.76123283e-06,\n",
      "        9.62129980e-06,  1.44071746e-05,  7.55339104e-04,  1.60393378e-04,\n",
      "        3.38772603e-04,  2.40483278e-05,  1.61936805e-05,  1.39054470e-03,\n",
      "        2.44286872e-04,  8.41300003e-04, -1.09130167e-04, -2.99446911e-05,\n",
      "       -2.52355938e-04,  5.63615991e-04, -5.36267180e-06, -2.28853853e-04,\n",
      "       -8.23089504e-05,  1.38920775e-04,  3.90060595e-05, -1.71758802e-06,\n",
      "        1.60854994e-04, -3.15716206e-05,  1.63523233e-04,  1.37712441e-05,\n",
      "        1.22178399e-05,  1.61934040e-05,  1.28354295e-04, -2.21378898e-04,\n",
      "       -1.76541347e-04, -4.34981077e-04, -1.41510944e-04,  4.27624298e-04,\n",
      "       -3.95508105e-04,  1.07586486e-04,  1.53738205e-04, -3.40761180e-05,\n",
      "       -1.21006109e-04,  2.40273279e-04,  8.31715734e-05,  1.05000421e-04,\n",
      "        2.45212897e-04,  3.46011919e-04,  2.95633246e-04, -9.81954799e-05,\n",
      "       -3.15158832e-05,  4.57929622e-04,  5.28044417e-04,  1.14601651e-04,\n",
      "        1.28646847e-04, -2.65623763e-04,  6.21984218e-05, -1.39268450e-04,\n",
      "        2.37332533e-05,  1.31304900e-04,  1.34324699e-04, -6.06638205e-04,\n",
      "        1.90468872e-05,  3.69485049e-03,  1.43896928e-03,  5.54362836e-04,\n",
      "        2.73615052e-03, -3.43261636e-03, -2.73271953e-03, -6.30330585e-04,\n",
      "       -4.59420960e-03,  1.49745285e-03,  6.54313015e-04, -3.27377464e-03,\n",
      "       -4.49793879e-06,  3.26768321e-04,  1.80867454e-03,  3.03280633e-03,\n",
      "        1.99375500e-05, -8.60517321e-04, -1.65970670e-03,  9.34928004e-03,\n",
      "       -3.90898157e-03, -9.56901233e-04,  9.63607628e-04, -3.18391365e-03,\n",
      "        3.17677885e-04,  2.39594327e-03,  5.28741977e-04,  1.11157451e-05,\n",
      "        8.58022850e-06,  1.20101004e-05,  1.08003751e-05,  8.49734079e-06,\n",
      "        9.72611269e-06,  9.17534726e-06,  8.49326716e-06,  8.99615316e-06,\n",
      "        9.58444980e-06,  9.67650703e-06,  1.00123398e-05,  7.78156300e-06,\n",
      "        9.42039878e-06,  8.34295042e-06,  8.86401631e-06,  8.57411669e-06,\n",
      "        1.10864885e-05,  1.45904951e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc978>, <tf.Tensor: id=900298, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-7.93384004e-07, -9.63122943e-07,  8.24891941e-08, ...,\n",
      "         1.72635191e-05, -4.33771929e-05, -1.34707516e-05],\n",
      "       [ 9.17134368e-09,  2.33147261e-07,  1.81998416e-07, ...,\n",
      "        -1.77980764e-05, -3.68631554e-05,  5.44280056e-06],\n",
      "       [-7.30726697e-07, -9.89504727e-08,  5.27367021e-08, ...,\n",
      "         2.47003427e-05, -8.27238546e-05, -1.47980736e-05],\n",
      "       ...,\n",
      "       [ 6.95009930e-07,  7.53424558e-08, -9.07287188e-08, ...,\n",
      "         1.25426723e-05, -2.63521324e-05, -1.26165587e-05],\n",
      "       [-6.86757895e-09,  5.67886493e-09,  4.76750834e-08, ...,\n",
      "        -4.15858085e-05,  1.01724196e-04,  2.30498154e-05],\n",
      "       [ 1.46380799e-07,  2.86443420e-08, -1.62470428e-08, ...,\n",
      "        -4.72921820e-05,  2.08150905e-05,  6.93561741e-09]], dtype=float32)>, <tf.Tensor: id=900299, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 5.2999676e-06,  1.4885404e-06,  2.2110801e-06, ...,\n",
      "         3.7381600e-04, -5.1896550e-05,  3.5667512e-05],\n",
      "       [ 1.2014460e-06,  1.3908607e-06,  1.5620990e-06, ...,\n",
      "         1.7105612e-04, -2.4775560e-05,  1.4845112e-05],\n",
      "       [ 1.0211232e-06,  8.3663804e-07,  2.5672505e-06, ...,\n",
      "         2.9890947e-05,  1.8520932e-05, -6.4586675e-05],\n",
      "       ...,\n",
      "       [-3.1953761e-07, -4.5505735e-07, -1.9207380e-07, ...,\n",
      "         3.4446959e-06,  2.6411954e-08, -8.7977975e-07],\n",
      "       [ 6.8837408e-06,  9.6443937e-06,  1.9422931e-05, ...,\n",
      "         1.3040932e-04, -2.0204192e-05,  2.3486362e-05],\n",
      "       [ 1.2245111e-05,  3.0283880e-05, -1.9521518e-05, ...,\n",
      "         1.3742680e-04, -4.3680353e-05,  4.9111477e-05]], dtype=float32)>, <tf.Tensor: id=900300, shape=(6144,), dtype=float32, numpy=\n",
      "array([-5.1860588e-06, -3.7822606e-06, -1.0005417e-05, ...,\n",
      "        3.4806205e-06,  1.3189061e-04,  1.4166796e-04], dtype=float32)>, <tf.Tensor: id=900203, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.0017579e-07,  7.0536364e-05, -9.9141312e-07, ...,\n",
      "        -9.3945312e-07, -1.3419177e-06,  8.3286955e-05],\n",
      "       [-1.5915878e-06, -2.1397329e-05, -1.7792343e-06, ...,\n",
      "        -1.5417864e-06, -2.4848641e-06,  1.4044190e-04],\n",
      "       [-4.7366331e-07, -1.9965082e-06, -5.6551437e-07, ...,\n",
      "        -6.5693985e-07, -7.1833193e-07,  9.6880904e-05],\n",
      "       ...,\n",
      "       [ 8.5176616e-06, -1.1542240e-03,  7.4905993e-06, ...,\n",
      "         8.5731326e-06,  1.1119084e-05, -4.4833042e-04],\n",
      "       [-8.5060191e-07,  1.5146716e-04, -8.0194144e-07, ...,\n",
      "        -8.7232780e-07, -1.0811291e-06,  5.9663082e-05],\n",
      "       [ 5.2593077e-07,  2.2145061e-05,  7.4910844e-07, ...,\n",
      "         4.7926568e-07,  1.0559982e-06, -2.1268828e-05]], dtype=float32)>, <tf.Tensor: id=900195, shape=(110,), dtype=float32, numpy=\n",
      "array([ 8.7645467e-06, -1.2217917e-03,  7.6710885e-06,  4.3920838e-05,\n",
      "        3.7682357e-03, -1.2285948e-04,  1.6324318e-03,  9.9989729e-06,\n",
      "        9.8420751e-06,  1.4728750e-05, -1.5011031e-04,  4.6812778e-04,\n",
      "        2.3426430e-05,  2.5295669e-05,  1.7117854e-05, -2.6691193e-04,\n",
      "       -2.0723157e-03,  1.9997964e-03, -2.6171134e-04,  2.1131297e-04,\n",
      "        4.8430692e-04,  2.6335471e-04, -2.2398363e-04,  3.3122930e-04,\n",
      "        3.2875524e-04, -6.3536725e-05,  2.7043917e-04, -6.1875144e-05,\n",
      "       -3.4410725e-04, -4.7840879e-04,  3.3213856e-04,  1.4060262e-05,\n",
      "        1.3008829e-05, -1.3909621e-04, -1.9635379e-05, -3.1114067e-04,\n",
      "        3.3405796e-04, -1.1519019e-04,  1.8778708e-04,  2.7731163e-04,\n",
      "       -2.1935937e-04, -3.7031158e-04,  3.2857643e-04,  1.1192381e-03,\n",
      "       -2.6832975e-04, -7.8044090e-05,  9.7513032e-05,  2.5385755e-04,\n",
      "        2.4631363e-04, -2.7177006e-04, -7.9112244e-04,  6.1377650e-05,\n",
      "        1.7894202e-04, -1.0957687e-05,  7.9239183e-04, -4.1059648e-05,\n",
      "       -2.4171895e-05,  4.4484547e-04,  6.4015054e-05, -2.9051508e-04,\n",
      "        2.5266498e-05, -4.8717513e-04, -3.3275087e-04,  1.9113304e-05,\n",
      "        1.9777672e-05, -3.0954272e-04,  5.3863286e-04, -4.2192666e-03,\n",
      "        2.1792634e-03, -7.1352297e-03, -4.3540909e-03, -2.5123670e-03,\n",
      "        4.5558983e-03,  1.8741712e-03,  3.3408171e-04,  5.8870309e-04,\n",
      "        6.3674576e-03, -2.9778637e-03,  6.7060732e-04,  1.4324733e-03,\n",
      "       -2.3328918e-03, -3.4301070e-04, -9.2976494e-04,  1.0009345e-02,\n",
      "       -5.9379176e-03, -2.8248653e-03,  6.7193154e-04,  1.3197931e-03,\n",
      "       -6.3613587e-04, -1.2963731e-03,  1.6621995e-04,  1.1367705e-05,\n",
      "        8.8553134e-06, -1.4358890e-04,  1.0887233e-05,  8.7167264e-06,\n",
      "       -1.4607358e-04,  9.2701212e-06,  8.7342069e-06,  9.3534154e-06,\n",
      "        9.8858591e-06,  1.0109459e-05,  1.0431491e-05,  8.0729696e-06,\n",
      "        9.6857939e-06,  8.6020664e-06,  9.3788385e-06,  8.8177203e-06,\n",
      "        1.1399377e-05, -4.5327147e-04], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafcb70>, <tf.Tensor: id=900659, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 4.28999147e-06, -8.54638301e-07, -2.22516610e-06, ...,\n",
      "        -2.24187170e-05, -4.29017346e-05,  3.33462481e-06],\n",
      "       [-5.23245127e-08,  2.79856152e-07,  1.75224031e-07, ...,\n",
      "        -3.12029224e-05, -7.77192838e-07, -5.59739874e-06],\n",
      "       [-1.16104502e-07,  1.50714612e-07, -2.65587033e-07, ...,\n",
      "         3.08204726e-05,  1.66667978e-05,  1.51967606e-05],\n",
      "       ...,\n",
      "       [ 1.00702823e-06,  4.43028256e-07, -4.12548530e-07, ...,\n",
      "         2.70372152e-06,  1.88304091e-06, -5.57474959e-06],\n",
      "       [ 4.96730408e-08,  3.84180865e-08, -1.40296965e-08, ...,\n",
      "        -7.08531006e-05,  2.21175887e-05,  1.13821170e-05],\n",
      "       [-2.42239452e-07,  5.34170283e-07,  3.85587640e-07, ...,\n",
      "         1.36379076e-05, -2.14242082e-05,  1.03227503e-05]], dtype=float32)>, <tf.Tensor: id=900660, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 7.2623502e-06,  6.9923508e-06,  9.9543022e-06, ...,\n",
      "         4.1469838e-04, -6.6683024e-05,  2.9487315e-05],\n",
      "       [ 4.9306429e-08,  7.1554030e-07, -4.4563620e-07, ...,\n",
      "         2.5244354e-04, -3.8914150e-05,  2.1329304e-05],\n",
      "       [ 2.6634316e-07, -4.2270818e-07,  6.9468689e-07, ...,\n",
      "        -8.3646722e-05,  1.0155743e-05,  1.6405698e-05],\n",
      "       ...,\n",
      "       [ 4.6412720e-07,  1.1743771e-06, -1.9110641e-07, ...,\n",
      "        -1.6769371e-05,  6.8614020e-07, -2.9142245e-06],\n",
      "       [ 8.3666191e-06,  9.7064149e-06,  1.1589855e-05, ...,\n",
      "         6.4203778e-05, -1.5177071e-05, -3.6770707e-06],\n",
      "       [-5.3144654e-06, -6.7097017e-06, -5.8163369e-06, ...,\n",
      "         2.4189439e-04, -1.6385669e-05,  5.0591269e-05]], dtype=float32)>, <tf.Tensor: id=900661, shape=(6144,), dtype=float32, numpy=\n",
      "array([-9.9755889e-06, -1.3315253e-06, -2.2625086e-06, ...,\n",
      "       -1.7041257e-05,  6.5386543e-05,  2.4517815e-04], dtype=float32)>, <tf.Tensor: id=900564, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.3297092e-07, -6.2647319e-05, -1.0151191e-06, ...,\n",
      "        -9.7810903e-07, -1.3804595e-06, -2.0731031e-06],\n",
      "       [-1.5939061e-06, -9.6047705e-05, -1.7823429e-06, ...,\n",
      "        -1.5021144e-06, -2.5133381e-06, -3.7465122e-06],\n",
      "       [-4.7266568e-07, -1.8545240e-04, -5.6416928e-07, ...,\n",
      "        -6.8170067e-07, -7.0985737e-07, -1.0583497e-06],\n",
      "       ...,\n",
      "       [ 8.9779387e-06,  1.2396637e-03,  7.7778222e-06, ...,\n",
      "         8.9540808e-06,  1.1632657e-05,  1.6044136e-05],\n",
      "       [-1.0115604e-06, -1.0711240e-04, -9.0085064e-07, ...,\n",
      "        -1.0237657e-06, -1.2396207e-06, -1.8632538e-06],\n",
      "       [ 6.0354523e-07, -3.9105653e-04,  8.2831690e-07, ...,\n",
      "         5.2554560e-07,  1.1877996e-06,  1.7331736e-06]], dtype=float32)>, <tf.Tensor: id=900556, shape=(110,), dtype=float32, numpy=\n",
      "array([ 9.28599275e-06,  1.29819778e-03,  7.99617283e-06, -5.83049899e-04,\n",
      "        3.84799251e-03,  3.55130505e-05,  1.55435840e-03,  1.03748898e-05,\n",
      "        1.03377542e-05,  1.54219942e-05,  5.36911655e-04,  4.66203521e-04,\n",
      "        1.69654668e-04,  2.64012742e-05,  1.78929222e-05, -3.29781394e-03,\n",
      "       -6.09750452e-04, -1.14363385e-03,  5.37246378e-05, -4.23185062e-04,\n",
      "        1.62394819e-04,  3.60687147e-04, -4.42425808e-04, -3.51516093e-04,\n",
      "       -3.72126437e-04,  1.16971896e-04,  5.34058199e-04,  3.79934791e-04,\n",
      "       -2.08901358e-04,  7.29805288e-06,  1.59713207e-04,  1.52688444e-05,\n",
      "       -2.98733445e-04,  1.82841868e-05,  2.92473240e-04,  5.44964605e-05,\n",
      "        1.98326175e-04, -8.89972434e-05,  4.80851158e-05,  4.44297475e-04,\n",
      "       -2.85975075e-05, -5.24772149e-05, -3.08553717e-04,  3.74196738e-04,\n",
      "        2.15884022e-04,  7.66365702e-05,  2.60334840e-04,  3.95443116e-04,\n",
      "        3.85135267e-04, -1.05790619e-04, -7.46326812e-04,  6.59691141e-05,\n",
      "       -5.69161493e-04, -1.10280316e-03,  1.58971641e-03, -3.61199272e-05,\n",
      "       -3.31053976e-04,  1.29162814e-04,  6.50883885e-05,  1.92584150e-04,\n",
      "        2.68479907e-05,  1.51042506e-04,  1.41466735e-04,  2.03290801e-05,\n",
      "        2.05219567e-05, -1.51033001e-03,  2.61848699e-03, -5.06956270e-03,\n",
      "        3.35691468e-04, -4.05698607e-04,  6.09300681e-04, -1.90989376e-04,\n",
      "        1.02425914e-03,  2.08416348e-03, -3.05153553e-05,  1.32092298e-03,\n",
      "        8.22179765e-03, -2.00584205e-03, -4.49430523e-03,  3.32071236e-03,\n",
      "       -3.25270463e-03,  6.53982162e-04,  1.82325792e-04,  4.76938148e-05,\n",
      "       -1.13756303e-02,  2.29412969e-03,  6.33845804e-04,  1.98637266e-04,\n",
      "       -3.09130228e-05,  5.61520690e-04,  1.93627624e-04,  1.19537972e-05,\n",
      "        9.19332342e-06,  1.34814336e-05,  1.17540621e-05,  9.25435779e-06,\n",
      "        1.07712131e-05,  1.01561045e-05,  9.29834187e-06,  9.84576036e-06,\n",
      "        1.04301535e-05,  1.06830103e-05,  1.09173052e-05,  8.56902534e-06,\n",
      "        1.05777408e-05,  9.03616728e-06,  9.84663529e-06,  9.25557106e-06,\n",
      "        1.19739343e-05,  1.64950270e-05], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafce80>, <tf.Tensor: id=901020, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-8.69135533e-07,  5.91458388e-07, -2.31767808e-06, ...,\n",
      "        -1.35950504e-05, -9.64664650e-06, -4.28175917e-05],\n",
      "       [-1.12509518e-07,  1.33809905e-07,  6.02908230e-08, ...,\n",
      "        -2.38856173e-05, -3.74885603e-05,  2.80835575e-05],\n",
      "       [-2.85592904e-07,  1.21996591e-08, -1.18411187e-07, ...,\n",
      "        -4.52951735e-06,  2.34367981e-05,  1.59062274e-05],\n",
      "       ...,\n",
      "       [ 5.10482096e-07,  2.23561614e-07, -3.56532261e-08, ...,\n",
      "         3.68515862e-06, -9.54087500e-06, -5.65714981e-06],\n",
      "       [-2.58647841e-08,  8.01800297e-08,  2.42353693e-08, ...,\n",
      "        -3.66983149e-05,  6.43909225e-05,  3.43250504e-05],\n",
      "       [-1.02528205e-07,  4.62947696e-07,  2.84310431e-07, ...,\n",
      "        -1.25897714e-05,  4.44505349e-06,  1.16029451e-05]], dtype=float32)>, <tf.Tensor: id=901021, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 1.3504500e-05,  1.7544589e-05,  1.1858861e-05, ...,\n",
      "         7.6464145e-05,  5.4377815e-06,  3.0142705e-06],\n",
      "       [-5.2678189e-07, -6.9167618e-07, -9.1069722e-07, ...,\n",
      "        -3.6133086e-05,  3.5014004e-06, -6.7921312e-05],\n",
      "       [ 1.1737874e-06,  6.3646871e-07,  3.9132142e-06, ...,\n",
      "         5.2795402e-04, -8.6045009e-05,  3.2133319e-06],\n",
      "       ...,\n",
      "       [ 4.5923039e-07,  1.0192817e-06,  1.8576316e-06, ...,\n",
      "         3.5956262e-05, -4.9242340e-06,  7.7869317e-06],\n",
      "       [ 5.1890775e-06,  7.4701702e-06,  1.3444281e-05, ...,\n",
      "        -2.1231876e-04,  2.2132432e-05, -3.6647052e-05],\n",
      "       [ 7.8333369e-07, -2.2850179e-06,  1.1347591e-05, ...,\n",
      "         3.5192503e-04, -3.6819729e-05,  1.1692630e-05]], dtype=float32)>, <tf.Tensor: id=901022, shape=(6144,), dtype=float32, numpy=\n",
      "array([-3.9987310e-05, -3.1698852e-07, -7.3967249e-06, ...,\n",
      "        3.6238453e-05, -2.1411688e-04,  3.5682871e-04], dtype=float32)>, <tf.Tensor: id=900925, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.0304991e-07,  1.1963812e-05, -9.9743431e-07, ...,\n",
      "        -9.2235524e-07, -1.3205836e-06, -2.0838988e-06],\n",
      "       [-1.5076420e-06, -2.7839942e-05, -1.7827956e-06, ...,\n",
      "        -1.4309055e-06, -2.4194342e-06, -3.7710115e-06],\n",
      "       [-3.7891056e-07, -1.1020787e-04, -4.8834704e-07, ...,\n",
      "        -5.4670215e-07, -5.8755546e-07, -9.4648249e-07],\n",
      "       ...,\n",
      "       [ 8.5632710e-06,  3.3015996e-05,  7.6046936e-06, ...,\n",
      "         8.4521726e-06,  1.1040132e-05,  1.5681428e-05],\n",
      "       [-9.9347926e-07,  1.6670969e-05, -8.9277512e-07, ...,\n",
      "        -9.9146109e-07, -1.2146340e-06, -1.8497547e-06],\n",
      "       [ 6.7853830e-07, -5.0846662e-04,  9.0571064e-07, ...,\n",
      "         6.0606521e-07,  1.2466068e-06,  1.9683846e-06]], dtype=float32)>, <tf.Tensor: id=900917, shape=(110,), dtype=float32, numpy=\n",
      "array([ 8.91221498e-06,  4.66545534e-05,  7.84878739e-06,  2.94639933e-04,\n",
      "       -4.31982940e-03,  3.43534048e-05,  1.53172971e-03,  1.02062313e-05,\n",
      "        9.88762804e-06,  1.53492219e-05, -2.11953811e-04, -3.49475245e-04,\n",
      "       -3.66458873e-04,  2.54328388e-05, -1.39221433e-04, -3.87011073e-03,\n",
      "       -8.57225503e-04,  2.87818303e-03, -8.85247486e-04, -3.77654942e-04,\n",
      "        2.13610590e-04, -1.36364019e-04,  2.56505795e-04, -1.22909871e-04,\n",
      "       -1.48495921e-04,  3.55773489e-04, -5.44904906e-05,  4.21143486e-04,\n",
      "        5.87466020e-05,  4.75833076e-06,  4.34967631e-04,  1.48511353e-05,\n",
      "        1.33975436e-05,  1.71733791e-05, -4.11383880e-05, -2.24840507e-04,\n",
      "        2.91117874e-04, -5.65123628e-04, -4.25555190e-04,  5.44245297e-04,\n",
      "        2.76350038e-04,  9.68747481e-05, -5.14724874e-04,  1.33310386e-04,\n",
      "        5.50236364e-05,  2.15151711e-04, -7.76276720e-05, -1.61265605e-04,\n",
      "        1.78309172e-04,  3.69499321e-06,  3.53737414e-04, -9.09829178e-05,\n",
      "        3.75826028e-04,  2.72243575e-04,  5.57163614e-04,  1.16520336e-04,\n",
      "       -1.76318630e-04,  1.99795730e-04,  6.13679586e-05,  1.76634552e-04,\n",
      "       -1.29330409e-04,  1.48402847e-04,  1.34042872e-04,  1.91041963e-05,\n",
      "        1.97949503e-05, -2.67153722e-03, -1.53295416e-03, -1.01507478e-03,\n",
      "        3.76088731e-03,  1.38353612e-02,  1.53027149e-03, -8.23072507e-04,\n",
      "        5.44887315e-03, -1.79697503e-03, -1.32801797e-04,  2.45290040e-03,\n",
      "        3.44527187e-03, -4.29037353e-03,  4.47727181e-03, -5.68038691e-03,\n",
      "       -1.61945634e-03,  5.45839954e-04,  7.16171833e-03, -7.53653282e-03,\n",
      "       -1.10855643e-02, -4.70578903e-03,  2.37781787e-03,  2.65741465e-03,\n",
      "       -1.98773603e-04, -9.51783266e-04, -1.72762506e-04,  1.15885632e-05,\n",
      "        8.96688016e-06,  1.30129329e-05,  1.15189878e-05,  8.94293407e-06,\n",
      "        1.05811387e-05,  9.35217031e-06,  8.98238886e-06,  9.42089810e-06,\n",
      "        1.00067764e-05,  1.00923080e-05,  1.04540513e-05,  8.43924954e-06,\n",
      "       -3.02113098e-04,  8.69091946e-06,  9.39840083e-06,  8.78784795e-06,\n",
      "        1.14142813e-05,  1.61790740e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc588>, <tf.Tensor: id=901381, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 2.1983474e-06,  1.4913724e-06, -1.9225251e-06, ...,\n",
      "         1.1114144e-05,  9.0568128e-06, -3.2469077e-05],\n",
      "       [ 1.1997120e-07, -1.9937531e-07,  1.8240189e-07, ...,\n",
      "        -3.0447833e-05, -2.5550473e-05, -3.5377011e-06],\n",
      "       [-3.2889511e-07, -7.4686362e-08,  8.4819185e-08, ...,\n",
      "         1.9123374e-05,  3.8435443e-05,  1.9869614e-05],\n",
      "       ...,\n",
      "       [ 6.2960771e-07, -3.0607981e-07, -2.3569520e-07, ...,\n",
      "        -8.8795068e-06, -2.4626370e-06, -8.7878188e-06],\n",
      "       [ 2.9356714e-08, -1.1347056e-08,  1.2352864e-08, ...,\n",
      "         2.0349029e-05,  4.6952014e-06,  1.7056769e-05],\n",
      "       [-6.8417143e-07,  1.0688035e-07,  3.7114344e-07, ...,\n",
      "        -5.8605874e-05,  7.5253294e-05,  3.4323948e-05]], dtype=float32)>, <tf.Tensor: id=901382, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 1.2900278e-05,  1.3949267e-05,  1.2362319e-05, ...,\n",
      "         3.5939028e-04, -6.7997978e-05,  2.8538516e-06],\n",
      "       [-4.7463820e-07, -4.4619469e-07,  3.8688219e-07, ...,\n",
      "         2.2798368e-04, -3.8072154e-05, -7.3832935e-06],\n",
      "       [-5.9288141e-07,  3.4310335e-07, -2.6478501e-06, ...,\n",
      "         5.4979055e-06,  8.4513904e-06, -3.1487230e-05],\n",
      "       ...,\n",
      "       [ 3.8677152e-07,  3.3493160e-07,  6.6403226e-07, ...,\n",
      "         3.2095024e-05, -3.7100147e-06,  5.9556110e-06],\n",
      "       [-3.3323442e-06, -1.7715201e-05, -1.3949689e-05, ...,\n",
      "         2.1799572e-04, -3.0764448e-05,  7.5786054e-05],\n",
      "       [ 1.0847379e-05,  9.8995024e-06, -9.0371959e-06, ...,\n",
      "         3.7974260e-05,  1.6192382e-05, -2.5650792e-05]], dtype=float32)>, <tf.Tensor: id=901383, shape=(6144,), dtype=float32, numpy=\n",
      "array([-2.8270948e-05,  1.4627332e-06,  6.4480412e-07, ...,\n",
      "        3.2583699e-05,  2.2020865e-04,  3.8687678e-05], dtype=float32)>, <tf.Tensor: id=901286, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.0051257e-07, -3.5105426e-05, -9.4426809e-07, ...,\n",
      "        -8.4385681e-07,  5.1700667e-05, -2.0206444e-06],\n",
      "       [-1.5954283e-06, -1.0820526e-04, -1.8592089e-06, ...,\n",
      "        -1.5485309e-06,  1.5594548e-04, -3.9359047e-06],\n",
      "       [-3.3335016e-07,  4.4678443e-05, -4.3560590e-07, ...,\n",
      "        -4.4274950e-07,  2.0909197e-04, -8.9725631e-07],\n",
      "       ...,\n",
      "       [ 6.3153711e-06,  3.7655974e-04,  6.1863311e-06, ...,\n",
      "         6.3268408e-06, -3.0038066e-04,  1.3091005e-05],\n",
      "       [-6.3500067e-07, -4.5870642e-05, -6.7448809e-07, ...,\n",
      "        -6.5117104e-07,  4.9790935e-05, -1.4416037e-06],\n",
      "       [ 8.9480068e-07,  4.1792173e-05,  1.0646349e-06, ...,\n",
      "         8.7601973e-07,  5.6971163e-05,  2.2955387e-06]], dtype=float32)>, <tf.Tensor: id=901278, shape=(110,), dtype=float32, numpy=\n",
      "array([ 6.4339279e-06,  3.8012033e-04,  6.2865620e-06,  9.3664508e-05,\n",
      "        8.2389973e-03, -1.2577280e-04, -3.8056131e-04,  7.9334914e-06,\n",
      "        7.4356549e-06,  1.3274810e-05,  3.0956648e-03,  2.5294218e-04,\n",
      "        2.2212084e-04,  2.1521999e-05,  1.4056432e-05, -2.4347766e-03,\n",
      "       -1.2628955e-03,  2.8893675e-03,  5.0182891e-05,  1.3785011e-05,\n",
      "        2.4710540e-04, -1.6449971e-04,  1.7138623e-04,  1.1996699e-04,\n",
      "       -2.0292739e-04,  3.6876978e-04, -7.0710958e-06,  3.0908096e-04,\n",
      "       -2.0002396e-04, -5.7620031e-04,  4.3218263e-04,  1.1284653e-05,\n",
      "        1.0605579e-05,  1.3211725e-05,  2.7420509e-04, -1.9337138e-04,\n",
      "        5.2648986e-04,  2.9658870e-05,  1.8408139e-04,  2.8648091e-04,\n",
      "        8.4426341e-05,  2.3544609e-04,  5.6628633e-05, -6.5313914e-04,\n",
      "       -2.7244119e-04,  2.8509092e-05,  2.0288031e-04,  5.1475951e-04,\n",
      "        1.2304747e-04,  8.1425591e-05,  1.7846802e-05,  6.0619775e-05,\n",
      "        3.5753637e-04,  1.9740227e-04, -1.2334005e-04, -5.4264066e-05,\n",
      "        1.2459520e-04, -2.6909803e-04,  5.0618317e-05,  1.4557033e-04,\n",
      "       -1.3160756e-04,  1.3960745e-04,  1.2286239e-04,  1.4896111e-05,\n",
      "        1.6647604e-05,  1.3489835e-03, -1.4071511e-03,  5.3152209e-03,\n",
      "        7.0907717e-04,  5.0107073e-03, -6.8695561e-05, -4.3516807e-04,\n",
      "       -1.5266938e-05, -2.9970808e-03,  4.9297302e-04,  1.3407007e-03,\n",
      "       -8.9004240e-04,  2.4759546e-03, -9.9392189e-04, -8.4624430e-03,\n",
      "       -1.3036723e-03, -6.4187669e-05,  2.6446132e-03, -5.7594329e-03,\n",
      "       -6.4427899e-03,  1.8417818e-04, -8.2698202e-04, -2.9762759e-04,\n",
      "       -3.9886370e-05, -2.5214532e-03, -4.9802917e-04,  9.2458713e-06,\n",
      "        7.0918582e-06,  9.6481299e-06,  8.1384187e-06,  6.5089343e-06,\n",
      "        7.8633675e-06,  6.2829167e-06,  6.4658489e-06,  7.0486085e-06,\n",
      "        7.9632600e-06,  7.4118134e-06,  8.4321337e-06,  6.1737151e-06,\n",
      "       -1.4852486e-04,  6.3324460e-06,  7.4376576e-06,  6.4402448e-06,\n",
      "       -3.0346698e-04,  1.3301887e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc550>, <tf.Tensor: id=901742, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-4.30246416e-07, -7.41337487e-07, -9.60288773e-08, ...,\n",
      "        -6.49710637e-05,  1.52921930e-05,  3.56827222e-05],\n",
      "       [ 2.25849121e-07, -3.18655502e-08,  1.00396349e-07, ...,\n",
      "         7.87037334e-06, -1.45466647e-05,  1.77063375e-05],\n",
      "       [-1.01999706e-06, -7.18464491e-07,  3.51443617e-07, ...,\n",
      "        -3.46255038e-05,  1.69429804e-05,  1.66554582e-05],\n",
      "       ...,\n",
      "       [-5.72256226e-07, -1.10378679e-07,  3.49952927e-07, ...,\n",
      "        -5.51086396e-06,  4.91613855e-06,  9.38469566e-06],\n",
      "       [ 3.80288618e-08,  6.29141752e-08, -8.60478178e-09, ...,\n",
      "         5.32697186e-06,  9.44870044e-06, -1.73056560e-06],\n",
      "       [ 8.68132446e-08,  5.27924556e-07, -3.17669340e-08, ...,\n",
      "        -1.12803973e-05, -1.21380735e-05,  8.99196129e-06]], dtype=float32)>, <tf.Tensor: id=901743, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 3.24714711e-06,  6.07804623e-06,  1.10415176e-05, ...,\n",
      "        -1.88176291e-05, -1.49187854e-05, -3.37482743e-06],\n",
      "       [-1.03447114e-07, -8.76074751e-07, -3.64557280e-07, ...,\n",
      "         9.10041053e-05, -1.98964881e-05, -6.69637484e-06],\n",
      "       [-6.80887808e-08, -2.51039978e-07,  2.29615148e-06, ...,\n",
      "         2.20677524e-04, -3.65868691e-05, -6.02593354e-06],\n",
      "       ...,\n",
      "       [-1.04894760e-07, -5.08797882e-07, -1.73683486e-06, ...,\n",
      "        -3.11920703e-05,  2.84763382e-06, -5.90519448e-06],\n",
      "       [ 1.09351356e-06,  4.31481794e-06,  1.51974782e-05, ...,\n",
      "         4.81164607e-05, -1.11144600e-05, -6.84761972e-07],\n",
      "       [-1.26885507e-05, -2.55704508e-05,  1.76922331e-05, ...,\n",
      "         3.42718413e-04, -4.54581532e-05,  2.87806106e-05]], dtype=float32)>, <tf.Tensor: id=901744, shape=(6144,), dtype=float32, numpy=\n",
      "array([-9.6032409e-06,  1.2190889e-06, -5.2970845e-06, ...,\n",
      "       -3.2231696e-05,  5.1390805e-05,  3.4799843e-04], dtype=float32)>, <tf.Tensor: id=901647, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.3776121e-07,  4.0470026e-05, -9.6296753e-07, ...,\n",
      "        -8.7135766e-07, -1.3078657e-06, -2.0489581e-06],\n",
      "       [-1.2177079e-06, -8.9815556e-05, -1.5718907e-06, ...,\n",
      "        -1.1520092e-06, -2.1910878e-06, -3.4969462e-06],\n",
      "       [-5.0900877e-07,  8.7752960e-05, -6.0566651e-07, ...,\n",
      "        -6.6852806e-07, -8.1547546e-07, -1.1699493e-06],\n",
      "       ...,\n",
      "       [ 8.9451169e-06, -1.6034697e-05,  7.8857120e-06, ...,\n",
      "         8.9313326e-06,  1.1741195e-05,  1.6254671e-05],\n",
      "       [-8.0784247e-07,  1.3878147e-05, -7.5573791e-07, ...,\n",
      "        -8.1766058e-07, -1.0117344e-06, -1.6096080e-06],\n",
      "       [ 3.4832644e-07, -2.3914808e-04,  6.4780909e-07, ...,\n",
      "         2.4764239e-07,  9.0136689e-07,  1.5713429e-06]], dtype=float32)>, <tf.Tensor: id=901639, shape=(110,), dtype=float32, numpy=\n",
      "array([ 9.4241195e-06, -2.0369962e-05,  8.2341894e-06, -8.4609562e-04,\n",
      "       -2.5352512e-03,  3.4496374e-05, -9.1891084e-04,  1.0618312e-05,\n",
      "        1.0649872e-05,  1.5168816e-05,  4.1386898e-05,  6.5617956e-04,\n",
      "        3.9643043e-04,  2.5618874e-05,  1.8224966e-05, -9.2764565e-04,\n",
      "        4.9019465e-04,  2.7379778e-03,  5.8294023e-05,  6.8935449e-05,\n",
      "       -7.6363882e-04,  1.9125757e-05, -1.4021900e-04, -4.1629377e-05,\n",
      "        2.3234001e-04,  3.5983653e-04, -2.4563094e-05,  4.8350828e-04,\n",
      "        1.5135822e-04, -2.2264004e-04,  5.2472856e-04,  1.5399695e-05,\n",
      "        1.4018976e-05,  1.7566774e-05, -2.0846564e-04, -6.5839558e-04,\n",
      "       -1.2907603e-04,  5.6306169e-05,  4.4211680e-05,  1.3877930e-04,\n",
      "       -4.6533372e-05,  7.1059869e-05,  1.1325980e-04, -5.5765774e-04,\n",
      "        5.1861622e-05,  1.8546928e-04,  5.9070389e-05,  4.9405753e-05,\n",
      "        2.8804422e-04, -1.4356947e-04,  5.9937352e-05,  6.4737542e-05,\n",
      "        5.6500366e-04, -6.6878267e-05,  7.2459446e-04, -2.0458979e-04,\n",
      "       -3.2786233e-04,  2.8579010e-04,  5.8744132e-05,  1.6391104e-04,\n",
      "        2.7224156e-05, -1.6685209e-04, -1.7947904e-04,  2.0825224e-05,\n",
      "        1.9966295e-05, -5.0873733e-03,  2.5175728e-03,  4.7218651e-03,\n",
      "       -2.2792462e-03,  2.4769923e-03,  1.0089660e-03,  7.5951620e-04,\n",
      "       -2.7223143e-03,  4.6384026e-04,  1.0320626e-04,  1.2152964e-03,\n",
      "       -9.6760906e-04, -2.2534572e-05, -4.0739113e-03, -7.2735577e-04,\n",
      "       -1.0944812e-03,  2.5075019e-04,  2.1603932e-03, -5.5747395e-03,\n",
      "        6.4559267e-03,  8.5267477e-04,  9.5363555e-04,  5.6520046e-04,\n",
      "        9.6858566e-05, -2.9629236e-03,  3.8134889e-04,  1.2399270e-05,\n",
      "        9.3555363e-06,  1.4408052e-05,  1.2147092e-05,  9.4657562e-06,\n",
      "        1.1628630e-05,  1.0591324e-05,  9.5849846e-06,  1.0452810e-05,\n",
      "        1.0601373e-05,  1.0570848e-05,  1.0864456e-05,  9.0732274e-06,\n",
      "        1.1709350e-05,  9.4828911e-06,  9.8946402e-06,  9.4251445e-06,\n",
      "        1.2278073e-05,  1.6912310e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc908>, <tf.Tensor: id=902103, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-6.7537900e-07, -8.3403495e-07,  7.8923627e-08, ...,\n",
      "        -4.6711226e-05,  1.9120154e-05,  1.0291407e-04],\n",
      "       [ 8.4506112e-08,  4.8217805e-07,  2.8365193e-07, ...,\n",
      "         6.1190520e-05,  1.7658878e-05,  2.5551528e-05],\n",
      "       [ 1.7287432e-08, -9.4996210e-08,  1.3821577e-07, ...,\n",
      "        -9.0732639e-05,  7.2870309e-05, -8.5922056e-06],\n",
      "       ...,\n",
      "       [-4.7324016e-07, -3.9840394e-08,  2.9906997e-07, ...,\n",
      "        -1.2848396e-05,  1.2958546e-05,  8.7569151e-06],\n",
      "       [-1.9254967e-09,  2.4534341e-08, -2.0282164e-08, ...,\n",
      "         1.4132191e-06, -2.6851185e-05, -3.3377521e-05],\n",
      "       [-3.8044627e-07,  4.5224485e-07, -9.3283219e-08, ...,\n",
      "         1.5254371e-06, -8.7581415e-05, -2.9653051e-05]], dtype=float32)>, <tf.Tensor: id=902104, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 1.68082306e-06,  5.28774763e-06,  6.95982362e-06, ...,\n",
      "         1.13095484e-04,  2.42400984e-06, -1.76636458e-05],\n",
      "       [ 5.03580168e-07,  2.37113923e-07,  1.62959657e-06, ...,\n",
      "        -2.11367616e-04,  3.04889400e-05,  2.50309404e-05],\n",
      "       [-1.13140936e-06, -1.34791526e-06, -2.48431661e-07, ...,\n",
      "         3.54788644e-04, -4.53479697e-05,  1.10319306e-04],\n",
      "       ...,\n",
      "       [ 4.10847150e-07,  1.56955150e-06,  5.49428556e-08, ...,\n",
      "        -3.34667056e-05,  2.96463827e-06, -6.54141650e-06],\n",
      "       [-4.60811918e-07, -7.36232050e-06,  6.20683579e-07, ...,\n",
      "        -1.28263375e-04,  1.19060423e-05, -5.52883248e-05],\n",
      "       [-3.89186698e-06, -6.97335463e-06,  2.07007815e-08, ...,\n",
      "         6.73605711e-04, -1.00070414e-04,  5.36172192e-05]], dtype=float32)>, <tf.Tensor: id=902105, shape=(6144,), dtype=float32, numpy=\n",
      "array([-7.9620386e-06, -5.2960381e-06,  4.2742886e-06, ...,\n",
      "       -3.4036926e-05, -1.2903412e-04,  6.7856372e-04], dtype=float32)>, <tf.Tensor: id=902008, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.0822616e-07,  2.5681184e-05, -9.3216215e-07, ...,\n",
      "        -8.4708529e-07, -1.3037555e-06,  4.7053014e-05],\n",
      "       [-1.4001539e-06,  4.0125869e-05, -1.6728779e-06, ...,\n",
      "        -1.3486756e-06, -2.3945963e-06,  4.1297837e-05],\n",
      "       [-5.3438254e-07, -1.7601515e-04, -6.4600232e-07, ...,\n",
      "        -6.8947168e-07, -8.8711505e-07,  4.0864270e-05],\n",
      "       ...,\n",
      "       [ 7.1023114e-06, -4.2603619e-04,  6.5691852e-06, ...,\n",
      "         7.1346476e-06,  9.9183098e-06, -1.4092545e-04],\n",
      "       [-7.4248061e-07,  1.1440921e-04, -7.1555780e-07, ...,\n",
      "        -7.6186564e-07, -9.9195597e-07,  2.3387662e-05],\n",
      "       [ 4.9463085e-07,  2.2669195e-04,  7.0202293e-07, ...,\n",
      "         4.4489005e-07,  9.9949602e-07,  5.7309071e-06]], dtype=float32)>, <tf.Tensor: id=902000, shape=(110,), dtype=float32, numpy=\n",
      "array([ 7.3232063e-06, -4.4487842e-04,  6.7348665e-06,  4.6220125e-04,\n",
      "       -5.6284941e-03,  3.1852149e-05,  2.8398228e-04,  8.5377242e-06,\n",
      "        8.5021811e-06,  1.2578628e-05,  9.9601038e-04,  9.9933102e-05,\n",
      "        2.1190950e-04,  2.1873346e-05,  1.5397254e-05, -8.2163402e-04,\n",
      "        4.0899601e-04, -5.9815793e-06, -9.9102741e-05, -7.8484934e-04,\n",
      "       -5.7577563e-05,  8.0318394e-05, -5.7535723e-05,  2.8281089e-05,\n",
      "        2.9428233e-04,  3.6254353e-04,  2.8757442e-05,  2.1020108e-04,\n",
      "        1.1777060e-04,  1.1542170e-04, -1.5909776e-05, -2.9998657e-04,\n",
      "        1.1911430e-05, -2.9771699e-04,  4.0193083e-04, -6.0440507e-06,\n",
      "        1.3369024e-04, -2.7941816e-04, -6.0400827e-04,  4.0413393e-04,\n",
      "       -3.8081521e-04, -4.2310852e-04, -3.7261908e-04,  1.2090149e-04,\n",
      "       -2.7212515e-04,  1.6132684e-04, -2.8361595e-04, -1.6609926e-04,\n",
      "       -7.3753625e-05,  3.1043589e-04, -1.1757244e-04,  6.0361795e-05,\n",
      "       -8.9451362e-04, -2.6369735e-04,  1.3574208e-03, -5.6262281e-05,\n",
      "        1.3370605e-04, -3.8051317e-04, -1.0426796e-04, -1.7038931e-04,\n",
      "        2.5117863e-05, -1.7901093e-05, -2.8993612e-05,  1.7384020e-05,\n",
      "        1.7291311e-05,  1.3517053e-03, -1.5252044e-03,  4.7637718e-03,\n",
      "       -3.1964248e-03, -3.0995011e-03,  2.0522268e-04, -3.2666355e-04,\n",
      "        1.1300108e-03,  1.7249859e-03, -4.5518597e-04,  9.4448210e-04,\n",
      "        2.6585818e-03, -2.9661497e-03, -9.0937642e-03,  9.9136180e-04,\n",
      "        1.4610288e-03,  8.2611659e-05, -6.7216856e-04, -5.5891844e-03,\n",
      "        1.2785067e-02,  4.1154833e-03,  3.9476165e-04,  2.2462360e-03,\n",
      "       -3.3736299e-04, -1.3749056e-03,  2.0848849e-04,  1.0170870e-05,\n",
      "        7.6452734e-06,  1.1256853e-05,  9.6136000e-06,  7.4254795e-06,\n",
      "        9.0472940e-06,  7.9424763e-06,  7.3608262e-06,  8.2318038e-06,\n",
      "        8.6632372e-06,  8.5695210e-06,  8.7666604e-06,  7.2224702e-06,\n",
      "        9.5571313e-06,  7.4200807e-06,  8.1063072e-06,  7.3487672e-06,\n",
      "        1.0179007e-05, -1.4144440e-04], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc278>, <tf.Tensor: id=902464, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-3.43708109e-07,  1.19291329e-07, -2.45781595e-09, ...,\n",
      "         2.16520802e-05, -2.30229271e-05,  2.73025144e-05],\n",
      "       [-2.47365136e-07, -1.55985262e-07,  6.40158007e-08, ...,\n",
      "        -2.32665370e-05, -1.21027333e-05, -1.40622615e-05],\n",
      "       [ 3.43115687e-07,  1.74834980e-07,  4.89826810e-08, ...,\n",
      "        -1.49122416e-05,  4.85418423e-05, -1.48092104e-05],\n",
      "       ...,\n",
      "       [-3.18492084e-07,  8.78175541e-08,  4.16574295e-07, ...,\n",
      "        -3.46881438e-06,  1.43680086e-06, -2.97495058e-06],\n",
      "       [ 2.84748953e-08,  3.80670464e-08, -3.36183414e-09, ...,\n",
      "        -1.57783961e-05, -2.65114909e-06,  1.25580473e-05],\n",
      "       [-3.68924077e-07,  6.20541769e-08,  3.65923505e-08, ...,\n",
      "        -1.29493483e-05,  5.53241407e-05,  5.61574925e-05]], dtype=float32)>, <tf.Tensor: id=902465, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 8.6944101e-07, -1.6466715e-06,  5.1526245e-06, ...,\n",
      "         7.4438081e-05, -1.8023366e-07, -4.5920508e-05],\n",
      "       [-7.1945593e-08, -3.0054136e-07, -1.4749282e-06, ...,\n",
      "        -3.8819315e-05, -3.0532003e-06, -1.6800312e-05],\n",
      "       [-1.9385206e-07, -9.0060752e-08, -5.3510041e-07, ...,\n",
      "         1.1957796e-05, -2.2967621e-05, -5.2093412e-05],\n",
      "       ...,\n",
      "       [-1.3583175e-07,  1.0665867e-07, -2.5756628e-07, ...,\n",
      "        -1.0999893e-05, -8.4891496e-07,  1.0823670e-06],\n",
      "       [-7.2591847e-06, -7.3103952e-06, -2.0000125e-06, ...,\n",
      "         7.6701435e-06, -7.5652943e-06,  1.2171146e-05],\n",
      "       [ 5.4852553e-06,  1.3449820e-05, -8.9647820e-06, ...,\n",
      "        -3.9639589e-04,  4.9748207e-05,  3.1211530e-05]], dtype=float32)>, <tf.Tensor: id=902466, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 2.0492062e-06,  3.1843140e-06,  6.8451976e-07, ...,\n",
      "       -1.1350467e-05,  6.9200396e-06, -4.0300659e-04], dtype=float32)>, <tf.Tensor: id=902369, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.59061572e-07, -1.16709387e-04, -9.78862090e-07, ...,\n",
      "        -8.92263643e-07, -1.40960890e-06, -2.19431740e-06],\n",
      "       [-1.52840073e-06, -1.94278546e-04, -1.79084429e-06, ...,\n",
      "        -1.45137290e-06, -2.64360233e-06, -4.06945946e-06],\n",
      "       [-5.35126333e-07, -1.17310803e-04, -6.23231529e-07, ...,\n",
      "        -6.77150524e-07, -8.99396809e-07, -1.30477724e-06],\n",
      "       ...,\n",
      "       [ 7.26550115e-06,  1.51830423e-03,  6.79412460e-06, ...,\n",
      "         7.15159240e-06,  1.04055835e-05,  1.50382975e-05],\n",
      "       [-6.74556077e-07, -9.32013063e-05, -6.72288934e-07, ...,\n",
      "        -6.89417675e-07, -9.37731159e-07, -1.52606458e-06],\n",
      "       [ 4.67665018e-07, -1.94249253e-04,  7.13938391e-07, ...,\n",
      "         4.42440069e-07,  1.04158084e-06,  1.66127847e-06]], dtype=float32)>, <tf.Tensor: id=902361, shape=(110,), dtype=float32, numpy=\n",
      "array([ 7.48426373e-06,  1.57919398e-03,  6.96548295e-06, -1.23890897e-03,\n",
      "       -3.33763612e-03,  3.29283939e-05, -1.26493676e-03,  8.79583695e-06,\n",
      "        8.65995207e-06,  1.32547784e-05,  8.18296336e-04,  8.09379053e-05,\n",
      "        1.90436098e-04,  2.26725078e-05, -1.39941461e-04,  1.71778817e-03,\n",
      "        1.47607119e-04,  2.77704967e-05, -9.55684081e-05, -3.57998244e-04,\n",
      "        3.63438099e-04, -1.37062033e-03,  6.32289084e-05,  1.61907992e-05,\n",
      "       -2.50438061e-05,  4.79641720e-04,  3.07413546e-04,  1.79529045e-04,\n",
      "        4.28284868e-04, -3.02533386e-04,  2.64665432e-04,  1.30928720e-05,\n",
      "        1.23424197e-05,  1.55962498e-05,  3.97236028e-04, -7.08914740e-06,\n",
      "       -6.43245527e-04, -2.40385620e-04, -9.97799580e-05,  7.83383730e-05,\n",
      "        4.40755044e-04, -2.52779253e-04,  1.32111702e-04, -3.50615766e-04,\n",
      "       -4.07373154e-04,  1.58812647e-04,  5.49121323e-05, -2.81656394e-05,\n",
      "       -5.40550100e-04, -1.52609573e-04,  1.16555493e-04,  6.33193195e-05,\n",
      "        2.57917913e-04, -1.99666392e-04, -1.50477979e-03, -5.16181863e-05,\n",
      "       -1.17998170e-05,  4.38597228e-04,  5.36902735e-05, -8.59378997e-06,\n",
      "        2.74294634e-05, -3.92827906e-06, -2.60273155e-05,  1.78112041e-05,\n",
      "        1.79850431e-05,  4.82041435e-03, -3.14510777e-04,  1.93159003e-03,\n",
      "       -2.19070259e-03, -7.85338506e-03, -9.91433975e-04, -2.06298451e-03,\n",
      "        2.48749531e-03, -2.53776554e-03, -7.22067547e-04, -2.30495527e-04,\n",
      "       -3.39690899e-03,  1.55072124e-03,  4.32207016e-03, -4.77879890e-04,\n",
      "        2.47575925e-03,  9.82662605e-05, -1.13289384e-03, -7.92052015e-04,\n",
      "        6.03287527e-03,  3.39983241e-03,  2.77874671e-04, -2.68759904e-05,\n",
      "        4.36469680e-04, -1.11091137e-03, -5.66387607e-04,  1.04949131e-05,\n",
      "        7.85472912e-06,  1.16832362e-05,  9.69267057e-06,  7.50759273e-06,\n",
      "        9.24304459e-06,  8.06781463e-06,  7.43398277e-06,  8.61138142e-06,\n",
      "        9.01405656e-06,  8.72842702e-06,  9.16582576e-06,  7.47170088e-06,\n",
      "        1.01538135e-05,  7.48460752e-06,  8.34186540e-06,  7.35924277e-06,\n",
      "        1.06741227e-05,  1.54084792e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafcfd0>, <tf.Tensor: id=902825, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 1.6570004e-06, -2.0550050e-07, -1.0349419e-06, ...,\n",
      "         3.2942877e-05, -7.9567828e-05, -2.9980496e-05],\n",
      "       [-3.6385174e-07, -1.3590068e-07,  1.4810360e-07, ...,\n",
      "        -4.3940090e-05, -3.4087204e-06, -1.2357667e-07],\n",
      "       [-6.0736505e-07,  9.7159173e-08, -1.7788453e-07, ...,\n",
      "        -3.5027628e-05, -4.5218329e-07,  1.2764635e-05],\n",
      "       ...,\n",
      "       [ 4.0170016e-07,  2.3848713e-07, -4.8095721e-08, ...,\n",
      "        -1.2932612e-06, -1.7084360e-05, -1.6170081e-05],\n",
      "       [ 1.8598197e-08,  2.8593468e-09,  6.4596706e-09, ...,\n",
      "        -9.8521545e-05,  1.9213152e-05, -8.4589292e-06],\n",
      "       [-4.3203997e-08,  2.9771016e-07,  2.3177172e-07, ...,\n",
      "         3.0904986e-05,  4.6483576e-05,  6.4581240e-05]], dtype=float32)>, <tf.Tensor: id=902826, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 1.0722032e-05,  2.7418359e-06,  1.1293581e-06, ...,\n",
      "         3.1966125e-04, -3.3532648e-05,  4.0067644e-05],\n",
      "       [ 4.0164161e-07,  8.7669383e-07,  5.2082191e-07, ...,\n",
      "        -9.8340650e-05, -9.0272060e-06, -5.4552416e-05],\n",
      "       [-6.6728703e-09,  2.2562872e-07, -8.8433370e-07, ...,\n",
      "        -7.2665454e-04,  7.3726136e-05, -1.0081161e-05],\n",
      "       ...,\n",
      "       [ 3.7479256e-07,  6.0559921e-07, -2.7797509e-07, ...,\n",
      "         3.6887895e-06,  8.6543281e-07, -4.3012130e-07],\n",
      "       [ 3.8910798e-06, -1.0205939e-05, -1.5779609e-05, ...,\n",
      "         3.7861711e-04, -5.1301082e-05,  2.9971587e-05],\n",
      "       [ 6.2323438e-06,  6.3252564e-06,  5.6271629e-06, ...,\n",
      "        -1.0587722e-03,  9.3489420e-05,  6.7586603e-05]], dtype=float32)>, <tf.Tensor: id=902827, shape=(6144,), dtype=float32, numpy=\n",
      "array([-2.6691428e-06, -3.0399412e-06, -1.4274843e-06, ...,\n",
      "        3.8550866e-06,  3.8433462e-04, -1.0755486e-03], dtype=float32)>, <tf.Tensor: id=902730, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.68550330e-07,  1.16901974e-04, -9.90552735e-07, ...,\n",
      "        -8.96826521e-07, -1.43492912e-06, -2.23188090e-06],\n",
      "       [-1.36778704e-06, -4.89513259e-05, -1.68424015e-06, ...,\n",
      "        -1.29862678e-06, -2.50993662e-06, -3.91155390e-06],\n",
      "       [-6.65237394e-07,  8.27925687e-05, -7.57069870e-07, ...,\n",
      "        -7.95447306e-07, -1.09698192e-06, -1.53391431e-06],\n",
      "       ...,\n",
      "       [ 8.00348789e-06, -3.50040529e-04,  7.17203875e-06, ...,\n",
      "         7.86939381e-06,  1.13375399e-05,  1.58772727e-05],\n",
      "       [-8.31229499e-07,  1.05160208e-04, -7.62454761e-07, ...,\n",
      "        -8.44970714e-07, -1.10169412e-06, -1.69538544e-06],\n",
      "       [ 1.60058818e-07, -1.31515262e-04,  5.11449571e-07, ...,\n",
      "         1.45779453e-07,  6.87359659e-07,  1.27579210e-06]], dtype=float32)>, <tf.Tensor: id=902722, shape=(110,), dtype=float32, numpy=\n",
      "array([ 8.28522116e-06, -3.81309568e-04,  7.37323717e-06, -4.31794266e-04,\n",
      "        8.29489436e-03,  3.53451069e-05, -4.18476557e-05,  9.49820515e-06,\n",
      "        9.38381618e-06,  1.32263276e-05,  2.52254715e-04, -2.69157841e-04,\n",
      "       -1.04975596e-04,  2.30091555e-05,  1.74666020e-05,  1.45934848e-03,\n",
      "        1.42781262e-03,  1.72172941e-03, -9.03238688e-05,  7.24685087e-04,\n",
      "       -3.12560529e-04,  2.85598857e-04,  2.26663891e-04,  3.45327135e-04,\n",
      "       -9.71741218e-04, -7.44157005e-05, -6.36911194e-04,  3.12616554e-04,\n",
      "       -1.61428361e-05,  5.77895029e-04,  9.46933142e-05, -2.97858845e-04,\n",
      "        1.31001434e-05, -2.95235513e-04, -6.91906898e-05, -1.44049831e-04,\n",
      "       -7.82807765e-04,  8.99227161e-05, -2.39579691e-04,  7.30276661e-05,\n",
      "       -1.71478780e-04, -9.23764819e-05,  4.84433025e-04,  2.76812527e-04,\n",
      "        7.81895142e-05, -1.81756059e-06,  7.19454838e-05, -2.91282049e-05,\n",
      "        8.83142857e-05, -1.24546685e-04,  4.73515072e-04,  6.42943924e-05,\n",
      "       -1.81566837e-04, -1.31818029e-04,  6.98929653e-05, -4.98055560e-05,\n",
      "        1.49080413e-04, -4.50577441e-04,  5.55553706e-05,  1.58009876e-04,\n",
      "        2.79225296e-05,  2.00477871e-06, -1.58108105e-05,  1.92946554e-05,\n",
      "        1.85947993e-05,  3.04009649e-03, -2.21428880e-03, -4.70543950e-04,\n",
      "       -3.07093398e-03, -1.61967073e-02,  3.13597266e-04,  1.47124403e-03,\n",
      "       -8.85019370e-04, -7.43890787e-03, -3.98014934e-04,  8.89228599e-04,\n",
      "        1.91208930e-03,  4.31436766e-03,  1.10845501e-02,  3.00753675e-03,\n",
      "       -4.61801275e-04, -3.70460446e-04, -1.07048666e-02,  6.29492896e-03,\n",
      "       -3.06796655e-03,  1.69596996e-03, -2.87889922e-03,  1.14059506e-03,\n",
      "        1.18574430e-03, -6.25248882e-04,  5.89828007e-04,  1.11619438e-05,\n",
      "        8.37368407e-06,  1.29226546e-05,  1.10800220e-05,  8.41510973e-06,\n",
      "        1.01928754e-05,  9.22716390e-06,  8.28215161e-06,  9.45937245e-06,\n",
      "        9.81794437e-06,  9.90269837e-06,  9.71819190e-06,  8.13857787e-06,\n",
      "        1.16236452e-05,  8.34229104e-06,  8.90944011e-06,  8.13820952e-06,\n",
      "        1.16729552e-05,  1.63185741e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafcb70>, <tf.Tensor: id=903186, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-3.08473204e-06, -8.32584249e-07,  1.27246196e-07, ...,\n",
      "         4.63202423e-06, -9.30258830e-05, -3.13199707e-05],\n",
      "       [-1.36601429e-07,  1.95863791e-07,  1.43149293e-07, ...,\n",
      "        -6.75692399e-06, -2.58848650e-05,  1.49393281e-06],\n",
      "       [-1.09456153e-07, -8.70244099e-10, -2.61852506e-07, ...,\n",
      "        -1.10670162e-05, -1.69712494e-05, -1.41843557e-05],\n",
      "       ...,\n",
      "       [-5.03302601e-07,  6.80937617e-08,  1.47438939e-07, ...,\n",
      "         2.55091618e-06, -5.24263487e-06, -1.36356550e-06],\n",
      "       [ 1.15712304e-07,  4.06155891e-08,  3.64791397e-08, ...,\n",
      "        -8.92143144e-05,  4.76940586e-05, -1.17875043e-05],\n",
      "       [-5.67114512e-07,  1.06809509e-07,  4.35872408e-07, ...,\n",
      "         3.76634998e-05, -1.35496784e-05,  4.81580682e-05]], dtype=float32)>, <tf.Tensor: id=903187, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-4.2505835e-06, -6.5697914e-06,  3.1606924e-06, ...,\n",
      "        -1.8004746e-04,  4.5065521e-05, -4.0463881e-05],\n",
      "       [-4.9517291e-07, -5.7593093e-07, -3.2896082e-07, ...,\n",
      "         7.5126867e-05, -1.5225475e-05,  2.4500296e-05],\n",
      "       [ 1.2244902e-06,  1.1148898e-06,  2.0132052e-06, ...,\n",
      "        -1.2575666e-04, -4.4584076e-06, -5.6486482e-05],\n",
      "       ...,\n",
      "       [ 4.2035009e-07,  6.2461169e-07, -4.6433269e-07, ...,\n",
      "        -3.8352300e-07,  9.1758420e-07, -1.9630374e-06],\n",
      "       [ 2.4775313e-06,  3.2307885e-06,  8.7381250e-06, ...,\n",
      "        -4.5682056e-05, -6.6355160e-06, -7.5376265e-06],\n",
      "       [ 6.1432388e-06, -1.8571703e-06, -1.5289475e-06, ...,\n",
      "        -6.4498855e-04,  3.9580791e-05,  7.4844524e-05]], dtype=float32)>, <tf.Tensor: id=903188, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 1.9300714e-05,  1.4797009e-06, -9.0395151e-06, ...,\n",
      "       -3.4644074e-07, -4.6716184e-05, -6.5870781e-04], dtype=float32)>, <tf.Tensor: id=903091, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.6335587e-07, -7.6543409e-05, -1.0054761e-06, ...,\n",
      "        -8.9199727e-07, -1.4666070e-06, -2.3432628e-06],\n",
      "       [-1.4129971e-06, -9.2594229e-05, -1.7260039e-06, ...,\n",
      "        -1.3424625e-06, -2.6029561e-06, -4.1284870e-06],\n",
      "       [-7.5921457e-07, -2.5259284e-04, -8.1271554e-07, ...,\n",
      "        -9.0659273e-07, -1.2354567e-06, -1.7658751e-06],\n",
      "       ...,\n",
      "       [ 7.9718593e-06,  1.3568150e-03,  7.2574244e-06, ...,\n",
      "         7.9118354e-06,  1.1497497e-05,  1.6241283e-05],\n",
      "       [-6.8939852e-07, -1.2569161e-04, -6.9223847e-07, ...,\n",
      "        -7.1884165e-07, -9.6309827e-07, -1.5984064e-06],\n",
      "       [ 3.0618742e-07, -4.8689049e-04,  6.0312038e-07, ...,\n",
      "         2.6975187e-07,  8.3495547e-07,  1.5413477e-06]], dtype=float32)>, <tf.Tensor: id=903083, shape=(110,), dtype=float32, numpy=\n",
      "array([ 8.2189454e-06,  1.4181217e-03,  7.4421491e-06, -1.1731095e-03,\n",
      "        2.3836060e-03,  3.5882651e-05, -6.8808476e-05,  9.3509898e-06,\n",
      "        9.4376601e-06,  1.3150909e-05,  3.6667180e-04, -4.2998820e-04,\n",
      "       -2.9045605e-04,  2.2967521e-05,  1.7960201e-05,  3.9397525e-03,\n",
      "        1.0249157e-03, -5.5164710e-04,  6.9718895e-05, -2.5906082e-04,\n",
      "        1.8055580e-06, -3.1869658e-04, -5.1044772e-04, -8.2704377e-05,\n",
      "        3.6882400e-04, -2.3982566e-04,  5.2597024e-04,  5.0567102e-04,\n",
      "        3.3053482e-04, -1.8250810e-04,  6.5660228e-05,  1.4663696e-05,\n",
      "        1.3472353e-05,  1.8186340e-05,  7.0791124e-05,  1.3903307e-05,\n",
      "        1.5956638e-04, -5.8786893e-05, -3.8861076e-04, -4.2692735e-04,\n",
      "        1.3457841e-04, -3.9474000e-04,  1.4003056e-04,  2.8450828e-05,\n",
      "       -3.7138321e-04, -1.3426823e-05,  6.2930529e-05, -6.7799498e-05,\n",
      "       -8.1824175e-05, -4.5788020e-04,  6.0414511e-04, -9.0681897e-05,\n",
      "        6.0813228e-04, -5.9854594e-04,  8.2579226e-04, -4.5682111e-05,\n",
      "       -2.8200038e-06, -3.0090112e-05, -4.1470991e-04,  1.4752084e-04,\n",
      "        2.9038682e-05, -1.5639568e-04, -1.7254886e-05,  1.9327839e-05,\n",
      "        1.8522562e-05,  1.0287696e-03, -2.1601177e-03, -4.6767215e-03,\n",
      "       -2.9049790e-04, -2.8756419e-03,  5.0515286e-05,  1.4165533e-03,\n",
      "       -4.0683951e-03, -3.1881635e-03,  4.3858678e-04,  4.8626278e-04,\n",
      "       -3.3734615e-03, -1.3358517e-03, -3.9199978e-04,  4.6494543e-03,\n",
      "        2.4792675e-03,  4.3545756e-04, -6.9820411e-03,  1.1676393e-02,\n",
      "       -5.5687572e-04, -5.2053109e-04, -1.8606079e-03,  1.6276597e-03,\n",
      "        5.9141679e-04,  1.0626846e-03, -1.6360554e-04,  1.1066852e-05,\n",
      "        8.3518089e-06,  1.2935402e-05,  1.0771696e-05,  8.2733577e-06,\n",
      "        1.0214418e-05,  8.9527657e-06,  8.2512834e-06,  9.7096781e-06,\n",
      "        9.9480867e-06,  9.6560616e-06,  9.9245990e-06,  8.2970373e-06,\n",
      "        1.1521314e-05,  8.1006510e-06,  9.0218873e-06,  8.1494863e-06,\n",
      "        1.1800636e-05,  1.6636535e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc400>, <tf.Tensor: id=903547, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-1.5145777e-06,  1.1308683e-06, -1.8463992e-06, ...,\n",
      "         3.4694385e-05, -6.4742235e-05, -3.0172392e-05],\n",
      "       [-3.0251081e-07, -2.3303519e-07,  1.2660499e-07, ...,\n",
      "         2.1906972e-05, -3.5917157e-05, -1.4130585e-05],\n",
      "       [ 3.2515285e-07,  6.6217581e-08, -8.8652286e-08, ...,\n",
      "        -2.0383375e-05,  1.1446887e-05, -4.2443703e-06],\n",
      "       ...,\n",
      "       [ 3.0897814e-07,  2.9754924e-07, -3.3623282e-09, ...,\n",
      "        -8.1342896e-06,  5.9404392e-06, -1.9923129e-06],\n",
      "       [ 4.2008402e-08,  6.5169155e-09,  6.2613283e-09, ...,\n",
      "        -3.4755642e-05,  2.6714055e-05,  2.9139013e-05],\n",
      "       [ 4.2063513e-07,  3.3034621e-07,  2.2935518e-09, ...,\n",
      "         1.3530132e-05,  9.3338676e-06, -1.0688867e-05]], dtype=float32)>, <tf.Tensor: id=903548, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 6.55727035e-06,  1.52277153e-05,  1.03801631e-05, ...,\n",
      "        -6.44149724e-04,  5.96795398e-05, -1.99781047e-04],\n",
      "       [-8.35971576e-08, -6.89492026e-07, -1.52211021e-06, ...,\n",
      "        -8.77072380e-05,  6.28993621e-06,  1.56273363e-05],\n",
      "       [-8.16241936e-07,  7.28058922e-07, -3.39586222e-06, ...,\n",
      "        -3.07017792e-04,  4.99849048e-05, -4.80512717e-05],\n",
      "       ...,\n",
      "       [-2.73153631e-07, -1.25976101e-07, -5.25865573e-07, ...,\n",
      "         1.28020531e-07,  1.85673201e-07, -1.43658247e-06],\n",
      "       [ 8.27375607e-06,  1.46382918e-05,  2.19780595e-05, ...,\n",
      "        -1.31614419e-04,  4.76922469e-06, -5.31997139e-05],\n",
      "       [ 3.58156444e-06,  6.65643893e-06,  5.75876220e-06, ...,\n",
      "        -1.25807695e-04, -1.13790775e-05,  2.20811235e-05]], dtype=float32)>, <tf.Tensor: id=903549, shape=(6144,), dtype=float32, numpy=\n",
      "array([-2.7976772e-05,  3.3224310e-07, -2.8358795e-07, ...,\n",
      "        6.0328148e-08, -1.3326138e-04, -1.3240262e-04], dtype=float32)>, <tf.Tensor: id=903452, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.79583695e-07,  2.37006770e-05, -1.01714704e-06, ...,\n",
      "        -9.10674316e-07, -1.48277036e-06, -2.40583154e-06],\n",
      "       [-1.46779576e-06,  1.46801158e-05, -1.77558593e-06, ...,\n",
      "        -1.40375994e-06, -2.64496748e-06, -4.20123843e-06],\n",
      "       [-4.47379875e-07, -1.39573513e-05, -5.81610777e-07, ...,\n",
      "        -5.74846808e-07, -8.69137011e-07, -1.28698923e-06],\n",
      "       ...,\n",
      "       [ 6.55132180e-06,  1.35649549e-04,  6.36448067e-06, ...,\n",
      "         6.36214463e-06,  9.82450547e-06,  1.46242655e-05],\n",
      "       [-6.65142124e-07,  3.23374024e-06, -6.88346972e-07, ...,\n",
      "        -6.72845715e-07, -9.51941615e-07, -1.62733966e-06],\n",
      "       [ 6.53339555e-07,  2.97572406e-05,  8.88697969e-07, ...,\n",
      "         6.25776920e-07,  1.23710970e-06,  2.10102758e-06]], dtype=float32)>, <tf.Tensor: id=903444, shape=(110,), dtype=float32, numpy=\n",
      "array([ 6.75010051e-06,  1.69144187e-04,  6.51956088e-06,  4.23350459e-04,\n",
      "        4.28508094e-04,  3.24927314e-05, -1.54887617e-04,  8.07730066e-06,\n",
      "        7.77353125e-06,  1.21802186e-05, -1.52648907e-04,  2.08185724e-04,\n",
      "        1.37489696e-04,  2.02158335e-05,  1.56320293e-05, -3.52826726e-04,\n",
      "        1.53747550e-03, -7.61935662e-05, -7.14545255e-04, -1.90712148e-04,\n",
      "        3.77389544e-04, -3.87455191e-04,  1.44288470e-05, -4.38220050e-05,\n",
      "        4.10621054e-04, -8.16573738e-05,  9.57871453e-05,  4.05233841e-05,\n",
      "       -1.00293932e-04,  1.43530429e-04, -2.81331624e-04,  1.29961609e-05,\n",
      "        1.15666890e-05,  1.59504398e-05, -2.65880517e-04,  3.67530243e-04,\n",
      "        1.22692818e-05,  1.12841510e-04, -2.15571737e-04, -2.96549406e-04,\n",
      "       -3.47778478e-05, -6.91443129e-05, -1.97531321e-04,  2.45289411e-04,\n",
      "       -3.12817647e-05,  1.30795321e-04,  3.60815844e-04, -7.27415143e-04,\n",
      "        1.97138332e-04,  3.01615772e-04,  8.63436653e-05, -9.13211770e-05,\n",
      "        2.80059787e-04,  3.70171212e-04, -1.02535239e-03,  1.09424087e-04,\n",
      "        1.50906169e-04,  9.91184133e-05,  4.88810110e-05, -2.31427584e-05,\n",
      "        2.90220778e-05, -1.55602203e-04, -1.85460653e-04,  1.63155983e-05,\n",
      "        1.64378052e-05, -8.11084826e-03,  4.80483024e-04, -2.85376213e-03,\n",
      "       -4.74338012e-04,  4.47439775e-03, -1.72695058e-04,  2.23847828e-03,\n",
      "        1.02893054e-03,  6.47903234e-03,  6.03544875e-04,  5.18902321e-04,\n",
      "       -4.95602237e-03,  1.28065911e-03, -5.70255611e-03,  5.58329839e-03,\n",
      "        3.27943848e-03, -6.68351495e-05, -3.35119851e-03,  4.74505592e-03,\n",
      "       -4.45500435e-03, -1.38622127e-03, -1.07965694e-04, -8.15657724e-04,\n",
      "       -5.98077677e-05,  8.31915182e-04, -4.06285137e-04,  9.63038656e-06,\n",
      "        7.00238616e-06,  1.05104555e-05,  8.52276116e-06,  6.71156931e-06,\n",
      "        8.42362624e-06,  6.81028541e-06,  6.59517627e-06,  7.60936200e-06,\n",
      "        8.17262480e-06,  7.69530016e-06,  8.50399738e-06,  6.80982566e-06,\n",
      "        1.00596189e-05,  6.63540004e-06,  7.61860701e-06,  6.54702308e-06,\n",
      "        1.00745719e-05,  1.49703110e-05], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafce80>, <tf.Tensor: id=903908, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-2.0680843e-06, -1.4654297e-06,  9.5836697e-08, ...,\n",
      "        -8.9271984e-05,  2.9639898e-05,  4.3116381e-07],\n",
      "       [-3.5091057e-07,  7.3477437e-08, -4.5992781e-08, ...,\n",
      "        -1.3242803e-06,  4.3568827e-05,  7.5754701e-06],\n",
      "       [ 3.3979381e-07,  1.8783443e-07,  1.3222332e-08, ...,\n",
      "         7.3876843e-05, -2.0835694e-05,  4.1140806e-06],\n",
      "       ...,\n",
      "       [ 1.0388258e-06,  4.9699185e-07, -1.2194350e-07, ...,\n",
      "        -6.5623444e-06,  9.5219675e-06,  9.6750364e-06],\n",
      "       [-1.6092471e-08,  4.9690907e-08,  4.4201589e-09, ...,\n",
      "         4.3297041e-05, -5.4945945e-06, -1.7716678e-05],\n",
      "       [ 1.8518726e-08,  3.3687169e-07, -2.8100919e-07, ...,\n",
      "         2.8746372e-06, -5.9382062e-05,  3.1596264e-05]], dtype=float32)>, <tf.Tensor: id=903909, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 2.1313247e-06,  1.8510660e-06,  1.5206937e-06, ...,\n",
      "        -1.5907736e-04,  2.9410066e-06, -1.3457936e-04],\n",
      "       [-3.3993155e-07, -7.9891299e-07, -1.7130901e-06, ...,\n",
      "         3.4419672e-06, -3.9204861e-06, -1.3788582e-06],\n",
      "       [ 8.5007639e-07,  6.6653928e-07,  2.2002516e-06, ...,\n",
      "         1.8845148e-04, -1.7845598e-05, -4.1059425e-06],\n",
      "       ...,\n",
      "       [ 9.8181761e-07,  1.2769511e-06, -4.5067944e-07, ...,\n",
      "         1.3774647e-05, -1.3496370e-06,  2.5168622e-06],\n",
      "       [ 4.7106037e-06,  1.9002653e-05,  2.9314248e-05, ...,\n",
      "        -4.0800794e-04,  3.9191549e-05, -7.3177114e-05],\n",
      "       [ 4.8936899e-06,  1.5742280e-06, -2.4397145e-06, ...,\n",
      "        -2.1023203e-04,  1.7284792e-05,  6.8052759e-05]], dtype=float32)>, <tf.Tensor: id=903910, shape=(6144,), dtype=float32, numpy=\n",
      "array([-5.3499198e-06,  9.7744271e-07, -5.9840722e-06, ...,\n",
      "        1.4101026e-05, -4.1477743e-04, -2.1626083e-04], dtype=float32)>, <tf.Tensor: id=903813, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.59137878e-07, -3.62256578e-05, -1.07302321e-06, ...,\n",
      "        -9.74231853e-07, -1.57315435e-06, -2.52373320e-06],\n",
      "       [-1.37555753e-06, -8.65873444e-05, -1.65356846e-06, ...,\n",
      "        -1.25209306e-06, -2.50903281e-06, -4.12384270e-06],\n",
      "       [-5.21104198e-07,  1.11404006e-04, -6.40706048e-07, ...,\n",
      "        -6.99922225e-07, -9.77265017e-07, -1.31443744e-06],\n",
      "       ...,\n",
      "       [ 8.12001326e-06,  9.44942003e-04,  7.42375232e-06, ...,\n",
      "         7.87783756e-06,  1.16666370e-05,  1.68472397e-05],\n",
      "       [-9.80674145e-07, -1.00983874e-04, -9.00843929e-07, ...,\n",
      "        -9.75244575e-07, -1.29095736e-06, -2.04938169e-06],\n",
      "       [ 4.56585724e-07, -7.90694321e-05,  7.27858605e-07, ...,\n",
      "         3.55260283e-07,  9.85134534e-07,  1.92783818e-06]], dtype=float32)>, <tf.Tensor: id=903805, shape=(110,), dtype=float32, numpy=\n",
      "array([ 8.4223511e-06,  9.9716953e-04,  7.6510032e-06,  9.0945145e-04,\n",
      "       -5.2278601e-03,  3.4844004e-05,  8.9055538e-07,  9.5332989e-06,\n",
      "        9.5056021e-06,  1.3430033e-05, -1.1515036e-03, -1.0534277e-04,\n",
      "       -3.2661390e-04,  2.3077784e-05,  1.8309087e-05,  1.0101021e-03,\n",
      "       -1.1132741e-03, -3.1190480e-03, -8.0366517e-05, -1.2511793e-04,\n",
      "        6.6589954e-04, -7.5801025e-04, -3.2958837e-05, -3.8999552e-04,\n",
      "       -4.0445538e-04, -2.2447048e-04,  1.5754580e-04,  2.1988009e-04,\n",
      "        4.0305709e-04,  5.9752073e-04, -4.4496865e-06,  1.6318158e-05,\n",
      "        1.4127055e-05,  2.0136766e-05,  3.1027048e-05,  6.1946770e-04,\n",
      "       -7.4942247e-05, -1.2573494e-04,  3.1366741e-04,  3.5899732e-04,\n",
      "        3.3024966e-04, -1.9814773e-04, -4.4891066e-04, -2.3435859e-05,\n",
      "        3.3029169e-04, -2.2637496e-05,  2.1525059e-04,  9.0955713e-05,\n",
      "        5.6838304e-05, -2.3831033e-04,  2.5217305e-04,  7.0572976e-05,\n",
      "        6.4057624e-04,  1.7596688e-04, -6.4544118e-04,  1.2511459e-04,\n",
      "        1.1986895e-05,  3.7724790e-04,  5.5398348e-05, -2.3283087e-06,\n",
      "        3.1683725e-05, -1.3984853e-04, -1.7716151e-04,  1.9886447e-05,\n",
      "        1.8392364e-05,  8.0988655e-05,  1.1579897e-03,  6.4531370e-04,\n",
      "        3.4886962e-03,  1.1632812e-02,  2.1165570e-03,  5.6506129e-04,\n",
      "        2.6027833e-03,  8.4782875e-04,  3.9565790e-04, -1.0266703e-03,\n",
      "       -5.0918357e-03,  1.0569644e-03, -4.1695959e-03,  2.1959695e-03,\n",
      "        5.0829651e-05, -1.1718931e-03,  2.8325729e-03, -2.9175989e-03,\n",
      "       -8.3474275e-03, -2.6850956e-03, -1.0328884e-03, -7.8097265e-04,\n",
      "       -2.3759535e-04,  3.1229670e-03,  3.7156852e-04,  1.1335930e-05,\n",
      "        8.2450943e-06,  1.3216265e-05,  1.0972378e-05,  8.3849336e-06,\n",
      "        1.0627706e-05,  9.1372922e-06,  8.2600855e-06,  9.2310665e-06,\n",
      "        9.7525372e-06,  9.7747761e-06,  9.9467406e-06,  8.3923733e-06,\n",
      "        1.2514613e-05,  8.3397599e-06,  9.0909871e-06,  8.1672015e-06,\n",
      "        1.2038702e-05,  1.7333314e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdae6748>, <tf.Tensor: id=904269, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-6.25901748e-06,  8.07552055e-07,  1.40965710e-06, ...,\n",
      "        -1.07932619e-05, -7.51138077e-06,  7.81264043e-06],\n",
      "       [ 1.54932081e-07,  1.02856525e-07, -2.94149434e-08, ...,\n",
      "         4.52167296e-06,  1.01242949e-05,  1.22731944e-05],\n",
      "       [ 2.78408066e-07, -4.80090030e-08,  7.64450618e-08, ...,\n",
      "         1.24385188e-05, -2.18013952e-06,  4.89324430e-06],\n",
      "       ...,\n",
      "       [ 1.41585076e-06,  5.16241698e-07, -5.22942344e-07, ...,\n",
      "         5.07239429e-06, -4.94095457e-06, -4.36808841e-06],\n",
      "       [-5.12235232e-08,  3.52876910e-08,  3.70214726e-09, ...,\n",
      "         2.30690548e-05, -1.31118759e-05,  1.43479629e-05],\n",
      "       [-1.06461528e-06, -6.52649348e-08,  3.30358546e-07, ...,\n",
      "        -1.58711264e-05,  2.92665800e-06,  2.92350978e-05]], dtype=float32)>, <tf.Tensor: id=904270, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-2.9930384e-06,  1.9196725e-06,  9.5533333e-06, ...,\n",
      "        -7.6633564e-04,  9.8093929e-05, -2.3056798e-04],\n",
      "       [-1.2603568e-06, -2.5081870e-06, -1.1300650e-06, ...,\n",
      "        -3.2076967e-04,  3.6856403e-05,  1.7844724e-05],\n",
      "       [ 2.1374915e-07,  9.1064476e-07,  5.4667407e-07, ...,\n",
      "         4.1862315e-04, -6.1552979e-05, -1.2110067e-05],\n",
      "       ...,\n",
      "       [ 6.1893900e-07,  1.9519791e-07, -2.7331805e-07, ...,\n",
      "         1.0399641e-05,  2.3896899e-07,  3.3732888e-06],\n",
      "       [ 6.9357161e-06,  2.8259974e-05,  3.2597232e-05, ...,\n",
      "        -6.3295313e-04,  7.0524969e-05, -1.8098502e-04],\n",
      "       [ 2.2623097e-05,  3.2583484e-05, -2.4213916e-05, ...,\n",
      "        -2.8102103e-04,  1.6377589e-05, -4.4590284e-05]], dtype=float32)>, <tf.Tensor: id=904271, shape=(6144,), dtype=float32, numpy=\n",
      "array([-1.55798953e-05, -1.90153656e-07,  7.53724805e-07, ...,\n",
      "        1.04984765e-05, -6.41499297e-04, -2.81526212e-04], dtype=float32)>, <tf.Tensor: id=904174, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.5063280e-07,  5.7769685e-06, -1.0721189e-06, ...,\n",
      "        -9.7320185e-07, -1.5734479e-06, -2.5771828e-06],\n",
      "       [-1.3797301e-06, -9.3864328e-06, -1.6440148e-06, ...,\n",
      "        -1.2736484e-06, -2.5031131e-06, -4.1441831e-06],\n",
      "       [-5.8651608e-07,  1.0519288e-06, -6.8824238e-07, ...,\n",
      "        -7.7123138e-07, -1.0288197e-06, -1.5054131e-06],\n",
      "       ...,\n",
      "       [ 8.0287728e-06,  4.2711146e-04,  7.4960026e-06, ...,\n",
      "         7.8650810e-06,  1.1674845e-05,  1.7162885e-05],\n",
      "       [-8.2846009e-07,  2.7411272e-05, -8.2311874e-07, ...,\n",
      "        -8.3925693e-07, -1.1674430e-06, -1.8920729e-06],\n",
      "       [ 3.9722443e-07, -1.7752440e-04,  6.9365922e-07, ...,\n",
      "         3.4585858e-07,  9.5636688e-07,  1.8713779e-06]], dtype=float32)>, <tf.Tensor: id=904166, shape=(110,), dtype=float32, numpy=\n",
      "array([ 8.2571642e-06,  4.4678507e-04,  7.6823717e-06, -3.4016211e-04,\n",
      "       -5.5844118e-03,  3.4705743e-05,  1.6162205e-04,  9.5602927e-06,\n",
      "        9.6888134e-06,  1.3441029e-05,  9.8677177e-04, -4.2139829e-04,\n",
      "       -1.1079914e-03,  2.3171604e-05,  1.8777691e-05, -1.3578474e-04,\n",
      "       -3.8090622e-04, -2.0979994e-03,  8.2606755e-05, -1.1011182e-04,\n",
      "       -5.1290175e-04, -1.1407145e-03,  5.7882734e-04,  2.6529484e-05,\n",
      "        3.3167264e-04, -8.6152315e-05,  4.2320852e-04, -9.0409903e-06,\n",
      "        1.7218497e-04, -2.0025975e-04, -6.4436812e-04,  1.6526063e-05,\n",
      "       -1.4224053e-04, -1.3629642e-04, -4.4499550e-04,  5.8368052e-04,\n",
      "       -1.8004645e-04,  2.1498554e-04, -2.7421859e-04, -1.0868979e-04,\n",
      "        1.2099246e-05,  1.3957430e-04,  2.1166357e-04,  4.9915107e-04,\n",
      "       -1.1080978e-04,  1.3043448e-04,  3.7610118e-04,  1.0719618e-04,\n",
      "        5.9127284e-05,  2.6528578e-04,  8.3044244e-05, -8.4830237e-05,\n",
      "        6.2483491e-04,  8.6458072e-05, -6.9997867e-04,  1.2511680e-04,\n",
      "        1.7153818e-04, -4.8075915e-05,  5.6300760e-05, -4.3297428e-06,\n",
      "        3.2021740e-05,  1.8719884e-04,  1.3685126e-04, -9.1667380e-04,\n",
      "        1.8701863e-05, -3.6565266e-03,  2.5343799e-03, -1.0982836e-03,\n",
      "        4.5396565e-03,  1.0599348e-02, -2.1652728e-03,  7.7457371e-06,\n",
      "       -2.8657124e-03,  8.5305749e-03,  3.9673905e-04,  6.4429245e-04,\n",
      "       -4.5788335e-03,  1.0808897e-03,  2.9262521e-03,  3.4805778e-03,\n",
      "       -1.1981468e-03,  9.2684219e-05,  7.6980791e-03, -5.6619253e-03,\n",
      "       -4.5022918e-03, -2.9603560e-03, -1.8507864e-03, -4.7477465e-03,\n",
      "       -1.7186046e-04,  1.6711412e-03, -4.9038790e-04,  1.1380023e-05,\n",
      "        8.3331024e-06,  1.3243356e-05,  1.1131700e-05,  8.3180339e-06,\n",
      "        1.0657492e-05,  9.1437869e-06,  8.1435892e-06,  9.4766092e-06,\n",
      "        1.0004237e-05,  9.6976592e-06,  1.0032603e-05,  8.4228641e-06,\n",
      "        1.2687142e-05,  8.4206331e-06,  9.0703170e-06,  8.0837635e-06,\n",
      "        1.1970303e-05,  1.7562596e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc3c8>, <tf.Tensor: id=904630, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-1.84871567e-06, -2.76941773e-06,  5.90436514e-07, ...,\n",
      "        -7.45747730e-06,  3.18460879e-05,  7.65471559e-05],\n",
      "       [ 1.18881459e-07,  2.30074676e-07, -1.26929635e-07, ...,\n",
      "        -2.87253170e-06,  3.12640054e-06,  1.29093278e-05],\n",
      "       [-4.71325592e-07, -2.27979626e-07,  3.22573470e-07, ...,\n",
      "         1.36840945e-05,  2.53550279e-05,  1.59861593e-05],\n",
      "       ...,\n",
      "       [ 7.47171441e-07,  2.35980679e-07, -1.35556107e-07, ...,\n",
      "        -6.89314038e-06,  4.25330791e-06,  2.55240184e-06],\n",
      "       [-1.05636602e-07,  7.15841750e-08, -6.56955823e-10, ...,\n",
      "         5.97293001e-05,  1.20358891e-05,  4.57086608e-05],\n",
      "       [-6.92561514e-07,  2.19796163e-08,  2.45511984e-07, ...,\n",
      "         6.62645743e-06, -5.86158349e-05, -2.80787990e-05]], dtype=float32)>, <tf.Tensor: id=904631, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-2.2736360e-06, -6.2555805e-06, -3.0606459e-06, ...,\n",
      "        -1.3724423e-04,  8.7476637e-06, -5.1399216e-05],\n",
      "       [-1.0846993e-06, -2.5870613e-06, -3.1567490e-06, ...,\n",
      "        -1.2334665e-04,  3.7770405e-05, -3.8197417e-05],\n",
      "       [-7.5423964e-07, -2.1751849e-07,  6.5943240e-08, ...,\n",
      "         3.4489983e-04, -3.8761926e-05,  4.7956335e-05],\n",
      "       ...,\n",
      "       [-6.9191799e-08,  4.7741861e-07,  6.1589668e-07, ...,\n",
      "        -5.1260922e-06,  4.8743408e-07, -7.4843382e-07],\n",
      "       [ 4.7939520e-06,  1.6801057e-06,  2.8845066e-06, ...,\n",
      "         2.3584490e-04, -2.4145707e-05,  1.2620322e-04],\n",
      "       [-1.3198702e-05, -4.3501230e-05,  2.9394652e-05, ...,\n",
      "         5.7525613e-04, -1.0959866e-04,  2.0195042e-05]], dtype=float32)>, <tf.Tensor: id=904632, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 1.0630241e-05,  3.8828930e-06,  4.9746827e-06, ...,\n",
      "       -5.3094554e-06,  2.3620360e-04,  5.8383332e-04], dtype=float32)>, <tf.Tensor: id=904535, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.2831772e-07,  9.2382637e-05, -1.0345123e-06, ...,\n",
      "        -9.4303817e-07, -1.5726112e-06, -2.5605286e-06],\n",
      "       [-1.4572413e-06, -8.2444356e-05, -1.7185992e-06, ...,\n",
      "        -1.3546937e-06, -2.7127608e-06, -4.3748605e-06],\n",
      "       [-5.1791096e-07, -5.0401912e-05, -6.1158812e-07, ...,\n",
      "        -7.1508111e-07, -1.0254943e-06, -1.3936976e-06],\n",
      "       ...,\n",
      "       [ 7.9028232e-06,  2.0596117e-04,  7.2534995e-06, ...,\n",
      "         7.7134555e-06,  1.1826325e-05,  1.7215920e-05],\n",
      "       [-7.7535822e-07,  2.4949899e-05, -7.4166769e-07, ...,\n",
      "        -7.7690322e-07, -1.0847182e-06, -1.7716429e-06],\n",
      "       [ 7.2277908e-07,  1.7874164e-04,  9.8569728e-07, ...,\n",
      "         6.3449852e-07,  1.4522143e-06,  2.5471215e-06]], dtype=float32)>, <tf.Tensor: id=904527, shape=(110,), dtype=float32, numpy=\n",
      "array([ 8.15149724e-06,  1.96071866e-04,  7.43882492e-06, -8.98990547e-05,\n",
      "       -3.20447725e-03,  3.47090318e-05, -1.07047393e-03,  9.36986635e-06,\n",
      "        9.45338797e-06,  1.35812461e-05, -1.18967984e-03, -6.95568509e-04,\n",
      "       -7.91607657e-04, -2.89341522e-04,  1.94042641e-05,  3.77212360e-04,\n",
      "        1.22818642e-03, -2.10184650e-03,  8.70524673e-05,  1.97734596e-04,\n",
      "        2.51363643e-04, -9.51352413e-05,  4.91296523e-04, -2.02152005e-05,\n",
      "       -3.71047354e-04, -5.28727076e-04,  3.41143343e-04,  2.20520102e-04,\n",
      "       -2.68261181e-04,  1.63940407e-04, -1.09489501e-05,  1.73122189e-05,\n",
      "        1.44093483e-05,  2.02977953e-05,  4.02710757e-05,  1.09527062e-03,\n",
      "       -5.80291846e-04,  4.22217359e-04,  2.48083234e-04,  4.10004068e-05,\n",
      "        1.91818821e-04,  1.68348546e-04,  2.28943027e-04, -4.28082945e-04,\n",
      "       -2.21611102e-04,  1.27618099e-04,  5.17976150e-05,  2.68739561e-04,\n",
      "        2.20005750e-04, -1.67253980e-04,  3.73136194e-04,  7.47445010e-05,\n",
      "        2.60536006e-04,  4.71586769e-04, -9.78798722e-04,  1.30131317e-04,\n",
      "        1.74557586e-04,  2.86997471e-04,  5.63976064e-05,  1.98170528e-06,\n",
      "       -1.22632860e-04,  2.01740928e-04,  1.38075338e-04, -2.91758915e-04,\n",
      "        1.84429518e-05,  3.39447724e-04,  1.44037697e-03,  4.85113473e-04,\n",
      "        1.26567716e-03, -1.69537481e-04, -4.56038048e-04, -8.04172945e-04,\n",
      "        1.69877685e-03,  4.74311551e-03,  6.97679294e-04, -1.35992916e-04,\n",
      "       -3.54729593e-04,  1.28161511e-03, -1.34497345e-03, -9.41780768e-03,\n",
      "       -2.89737800e-04, -6.04289584e-04,  1.15967561e-02, -7.37846876e-03,\n",
      "        5.18539175e-03,  1.05832180e-03,  3.26662295e-04, -4.33279900e-03,\n",
      "       -2.50348239e-04, -1.88657723e-05, -2.39714369e-04,  1.07283831e-05,\n",
      "        8.52551057e-06,  1.32050081e-05,  1.09540470e-05,  8.30971931e-06,\n",
      "        1.03727543e-05,  8.80793232e-06,  8.38011510e-06,  9.59114641e-06,\n",
      "        1.02934073e-05,  9.57029897e-06,  9.99068288e-06,  8.26830819e-06,\n",
      "        1.29974605e-05,  8.26832547e-06,  9.11374900e-06,  7.94503012e-06,\n",
      "        1.21368157e-05,  1.76431331e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc668>, <tf.Tensor: id=904991, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-9.8227440e-07, -1.8016774e-06, -7.9633310e-07, ...,\n",
      "        -2.2602560e-05, -1.0083775e-05,  1.8564357e-05],\n",
      "       [-2.8827023e-07,  3.0609097e-09,  2.5122296e-07, ...,\n",
      "        -7.3450219e-06, -1.1786245e-05,  2.2752897e-06],\n",
      "       [-2.5385057e-07,  2.3697638e-07,  2.1480884e-07, ...,\n",
      "        -1.5785070e-05,  2.2716347e-05,  5.3405379e-06],\n",
      "       ...,\n",
      "       [-2.1356085e-08, -1.3240081e-07,  2.2257495e-07, ...,\n",
      "        -1.3300834e-05,  6.1858503e-07, -2.8082040e-06],\n",
      "       [ 1.0109132e-07,  7.7067185e-08, -5.2201345e-09, ...,\n",
      "         3.1489664e-05,  8.9187970e-06,  6.4411324e-05],\n",
      "       [-5.7653989e-07,  1.2627461e-07,  3.4703731e-07, ...,\n",
      "         3.3584911e-05, -4.6760080e-05, -4.3798473e-06]], dtype=float32)>, <tf.Tensor: id=904992, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 1.1745036e-05,  3.2314681e-06,  1.1889026e-05, ...,\n",
      "         8.7522130e-05, -8.1579037e-06, -2.0863548e-05],\n",
      "       [-6.8780179e-07, -1.0431385e-06, -1.0164053e-06, ...,\n",
      "         3.1458194e-04, -3.1277668e-05, -4.6399546e-05],\n",
      "       [ 7.9925871e-07,  9.2397261e-07,  1.6034141e-06, ...,\n",
      "        -1.7116286e-04, -5.7347565e-06, -4.8354043e-05],\n",
      "       ...,\n",
      "       [ 4.2224295e-07,  1.7587604e-07, -3.2199743e-08, ...,\n",
      "         6.4441874e-06, -1.6384836e-06,  1.9370473e-06],\n",
      "       [-1.1285888e-05, -2.5885140e-05, -2.0717598e-05, ...,\n",
      "         7.8454998e-04, -7.1167546e-05,  2.5883992e-04],\n",
      "       [-4.1609565e-05, -7.4508425e-05,  6.4223641e-05, ...,\n",
      "         2.8204665e-04, -3.4151904e-05,  2.1021227e-04]], dtype=float32)>, <tf.Tensor: id=904993, shape=(6144,), dtype=float32, numpy=\n",
      "array([-2.0657280e-05,  4.4709291e-06, -4.0205477e-06, ...,\n",
      "        6.3685848e-06,  7.9646835e-04,  2.9050763e-04], dtype=float32)>, <tf.Tensor: id=904896, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.6040048e-07,  5.0953968e-06, -1.0663314e-06, ...,\n",
      "        -1.0026322e-06,  2.5627520e-05, -2.6622065e-06],\n",
      "       [-1.3969977e-06, -1.2246623e-04, -1.7101556e-06, ...,\n",
      "        -1.3334734e-06,  8.0632279e-05, -4.4265616e-06],\n",
      "       [-9.2710843e-07, -1.3094814e-05, -9.3769376e-07, ...,\n",
      "        -1.1548548e-06,  6.6636057e-05, -2.1151895e-06],\n",
      "       ...,\n",
      "       [ 9.5323085e-06,  1.0416084e-03,  8.5329330e-06, ...,\n",
      "         9.6833992e-06, -1.4040645e-04,  2.0081261e-05],\n",
      "       [-8.4006786e-07, -3.6243993e-05, -7.9395545e-07, ...,\n",
      "        -8.7602552e-07,  1.2544323e-05, -1.9200390e-06],\n",
      "       [ 6.4975302e-07,  1.6570160e-04,  9.5476355e-07, ...,\n",
      "         5.6754612e-07, -1.8600051e-05,  2.5877478e-06]], dtype=float32)>, <tf.Tensor: id=904888, shape=(110,), dtype=float32, numpy=\n",
      "array([ 9.81752692e-06,  1.06696959e-03,  8.75447950e-06,  1.04967365e-03,\n",
      "        9.01841465e-03,  3.98119992e-05,  1.18192300e-04,  1.07655042e-05,\n",
      "        1.12875314e-05,  1.51090953e-05,  3.57169163e-04,  1.26449449e-04,\n",
      "        4.01736528e-04,  2.67177875e-05,  2.20967504e-05,  2.61402281e-04,\n",
      "        3.74660420e-04, -6.00625016e-03, -2.10867773e-04, -1.06240704e-03,\n",
      "       -6.65529980e-04, -2.38095017e-05, -6.10722462e-04,  3.30936658e-04,\n",
      "        6.86784915e-07,  6.20587263e-04, -1.14534902e-04,  8.11640784e-05,\n",
      "       -2.13305728e-04,  3.61683808e-04,  1.69094274e-04,  2.10370526e-05,\n",
      "        1.69571831e-05,  2.34984473e-05,  3.83084262e-04,  1.00748555e-03,\n",
      "        2.02370720e-04,  4.25551989e-04,  4.77788970e-04, -2.57457345e-04,\n",
      "        2.15209962e-04, -1.17957221e-04, -2.44927742e-05, -1.05442025e-03,\n",
      "        3.00518906e-04, -2.80114255e-05, -4.00822348e-04, -1.85278564e-04,\n",
      "        2.39294066e-04,  2.02737676e-04, -1.01183366e-04,  8.10691708e-05,\n",
      "        7.13820482e-05,  2.49937671e-04, -1.75099180e-04, -1.77921684e-04,\n",
      "       -1.36724353e-04,  4.09225177e-05,  6.13862721e-05,  2.23628122e-05,\n",
      "        3.72580871e-05,  2.28605015e-04,  1.56674650e-04,  2.65878807e-05,\n",
      "        2.09603495e-05, -4.36937204e-03, -3.63059546e-04, -5.98253263e-03,\n",
      "        9.81542282e-04,  4.85774036e-03,  2.51035322e-04,  2.20251968e-04,\n",
      "        5.32946456e-03, -2.30749464e-03,  5.58652100e-04, -4.63622127e-04,\n",
      "       -1.83743599e-04, -2.94974842e-03,  1.83084700e-03, -1.25719579e-02,\n",
      "        2.71303230e-03,  5.86407550e-04,  7.06067018e-04,  1.62007974e-03,\n",
      "        4.29866463e-03,  7.20633951e-04, -7.08754931e-04,  3.45246517e-05,\n",
      "       -1.51967467e-03, -6.82275859e-04, -1.14452669e-04,  1.29212031e-05,\n",
      "        1.00450488e-05,  1.56772894e-05,  1.36759545e-05,  9.72098223e-06,\n",
      "        1.27475523e-05,  1.10666415e-05,  1.03496695e-05,  1.17840627e-05,\n",
      "        1.30189837e-05,  1.15461371e-05,  1.15737930e-05,  9.93763024e-06,\n",
      "        1.53584588e-05,  9.76508090e-06,  1.08470986e-05,  9.96881317e-06,\n",
      "       -1.41903773e-04,  2.05792312e-05], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafcba8>, <tf.Tensor: id=905352, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 2.3749167e-06,  4.8852684e-07, -2.6493192e-06, ...,\n",
      "        -2.6354496e-06,  2.5518197e-05,  1.2538050e-05],\n",
      "       [-3.7930886e-07,  9.2304340e-08,  1.5227906e-07, ...,\n",
      "        -6.2955141e-06, -6.1299477e-07,  3.6024761e-05],\n",
      "       [-6.8811732e-07, -3.5427283e-07,  3.7574068e-08, ...,\n",
      "        -8.3497061e-06,  3.7427657e-05,  1.4443307e-07],\n",
      "       ...,\n",
      "       [ 1.2971848e-06,  5.5151327e-07, -6.0250267e-07, ...,\n",
      "        -3.0768351e-06, -3.1868769e-06, -4.8909242e-06],\n",
      "       [ 2.4293572e-08,  4.7425541e-08,  7.4302431e-09, ...,\n",
      "        -7.1450064e-05,  2.1066608e-05,  1.3008388e-05],\n",
      "       [-2.7328082e-07,  2.9357258e-07,  1.8727117e-07, ...,\n",
      "        -1.3163490e-05, -2.0419993e-06, -2.4494762e-05]], dtype=float32)>, <tf.Tensor: id=905353, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 1.2752924e-05,  8.5100219e-06,  1.0640517e-05, ...,\n",
      "         8.9298096e-04, -9.7957061e-05,  1.9600659e-04],\n",
      "       [-3.8619146e-07, -4.9731926e-07, -1.4300239e-06, ...,\n",
      "         6.7367546e-05, -2.0600979e-05,  9.9875560e-06],\n",
      "       [ 3.6246649e-07,  3.6447946e-07,  1.4197326e-06, ...,\n",
      "         1.1650700e-04, -1.9538025e-05, -4.6177906e-06],\n",
      "       ...,\n",
      "       [ 2.4081581e-07, -4.9151001e-08, -7.1909670e-07, ...,\n",
      "        -1.7723201e-05,  1.7123953e-06, -5.7393199e-06],\n",
      "       [-1.2507433e-05, -2.7202341e-05, -3.1348485e-05, ...,\n",
      "         6.8164850e-04, -7.2458432e-05,  1.9901431e-04],\n",
      "       [-1.8053452e-06, -4.9952505e-06, -5.9293384e-06, ...,\n",
      "         2.5579004e-04, -2.7402097e-05,  3.0066392e-05]], dtype=float32)>, <tf.Tensor: id=905354, shape=(6144,), dtype=float32, numpy=\n",
      "array([-2.4527004e-05,  1.1182788e-06, -1.9881613e-06, ...,\n",
      "       -1.8213217e-05,  6.9317146e-04,  2.5753555e-04], dtype=float32)>, <tf.Tensor: id=905257, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.8437702e-07,  7.4611693e-05, -1.0100345e-06, ...,\n",
      "        -9.2149048e-07, -1.5590520e-06, -2.4582641e-06],\n",
      "       [-1.4872569e-06, -2.5017540e-05, -1.7645558e-06, ...,\n",
      "        -1.4610924e-06, -2.8641593e-06, -4.4291314e-06],\n",
      "       [-6.3564642e-07, -4.8185844e-05, -7.2328135e-07, ...,\n",
      "        -7.7639766e-07, -1.2446253e-06, -1.6976780e-06],\n",
      "       ...,\n",
      "       [ 7.6780043e-06, -2.8837391e-04,  7.3677193e-06, ...,\n",
      "         7.6026768e-06,  1.2040404e-05,  1.7245948e-05],\n",
      "       [-7.5170931e-07,  7.0791604e-05, -7.5019494e-07, ...,\n",
      "        -7.6125417e-07, -1.1060649e-06, -1.7556370e-06],\n",
      "       [ 7.8828009e-07, -1.3058561e-05,  1.0726003e-06, ...,\n",
      "         7.8347261e-07,  1.5898609e-06,  2.5878389e-06]], dtype=float32)>, <tf.Tensor: id=905249, shape=(110,), dtype=float32, numpy=\n",
      "array([ 7.89563273e-06, -2.64732807e-04,  7.54693929e-06,  4.43552155e-04,\n",
      "        6.22193050e-03, -2.76863721e-04,  6.14922261e-04,  9.12571977e-06,\n",
      "        9.25824679e-06,  1.37632760e-05, -6.60698453e-04,  1.73494162e-04,\n",
      "       -3.94256378e-04,  2.39255132e-05,  1.96243600e-05, -1.85334531e-03,\n",
      "        9.31762275e-04, -8.28126504e-04, -5.63558569e-05, -2.98959931e-05,\n",
      "        3.24933266e-04,  6.48812274e-04,  1.23964768e-04,  2.53918755e-04,\n",
      "       -2.35637854e-04,  7.20274169e-04,  1.18834083e-04, -6.22611726e-04,\n",
      "       -4.84449381e-04,  1.87552228e-04,  1.66116035e-04, -2.95161561e-04,\n",
      "        1.44100241e-05, -2.91875505e-04, -9.27201545e-05,  7.14577385e-04,\n",
      "       -1.74532426e-04,  3.81206686e-04,  2.83109519e-04, -1.32162662e-04,\n",
      "        2.91567500e-04,  1.95379936e-04,  1.87898520e-04,  3.42327374e-04,\n",
      "        2.87316914e-04, -4.23875208e-05,  1.92138759e-04, -2.24618532e-04,\n",
      "        4.58919239e-05,  6.02220651e-04,  4.27820429e-04, -7.91244238e-05,\n",
      "        4.53841814e-04,  6.04055123e-04,  8.18059954e-04,  1.26479441e-04,\n",
      "        1.63523349e-04,  6.50317059e-04,  5.65775299e-05, -1.25991064e-05,\n",
      "        3.55392549e-05,  4.75003144e-05,  1.49328043e-04, -6.02757384e-04,\n",
      "       -2.93588091e-04,  2.82394397e-03, -6.46419940e-04, -1.88707170e-04,\n",
      "        2.69436883e-03, -1.19579164e-02,  2.80071213e-03, -2.13710149e-03,\n",
      "        2.72705872e-03, -6.27801195e-03,  9.69405053e-04, -5.41034155e-04,\n",
      "        5.30488603e-03, -3.06092598e-03,  3.69748985e-03, -5.61466627e-03,\n",
      "        2.24689534e-03,  3.58087651e-04, -4.61880909e-03,  3.16128047e-04,\n",
      "        6.96262810e-04, -9.40823229e-04, -1.19969554e-05,  1.39503076e-03,\n",
      "       -7.63364311e-04, -3.11008189e-04,  7.04768230e-04,  1.10705705e-05,\n",
      "        8.36457639e-06,  1.28414185e-05,  1.05245217e-05,  8.08286768e-06,\n",
      "        1.03090233e-05,  8.31259331e-06,  8.02776412e-06,  9.31085833e-06,\n",
      "        1.01769838e-05,  9.26627399e-06,  9.83199971e-06,  8.11477548e-06,\n",
      "        1.29407881e-05,  8.06944354e-06,  8.91385480e-06,  7.81106792e-06,\n",
      "        1.23390118e-05,  1.76419380e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc470>, <tf.Tensor: id=905713, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-2.0224443e-06, -1.5259736e-06,  1.9846349e-08, ...,\n",
      "        -1.0082740e-04, -7.4690925e-06, -8.5951046e-07],\n",
      "       [ 3.1197114e-08,  1.9787716e-08,  1.8832779e-08, ...,\n",
      "        -1.6625718e-05,  4.1292093e-05,  2.8526109e-05],\n",
      "       [-4.4982121e-07,  1.7776000e-07,  2.0344793e-07, ...,\n",
      "         9.1841366e-06, -1.3097683e-05, -2.9239365e-05],\n",
      "       ...,\n",
      "       [ 1.6037203e-06,  9.3101853e-08, -1.6401614e-07, ...,\n",
      "         4.2372176e-06, -7.3212568e-08, -1.5454887e-06],\n",
      "       [ 1.5708613e-08, -2.2523642e-08,  3.9888363e-08, ...,\n",
      "         2.1141657e-06, -9.1491966e-05, -2.0778735e-07],\n",
      "       [ 4.7066912e-08,  4.7651031e-07,  9.1034387e-08, ...,\n",
      "        -6.1269820e-05,  1.4399114e-05, -4.8829943e-06]], dtype=float32)>, <tf.Tensor: id=905714, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 1.6079615e-06, -5.3458616e-06,  6.0900305e-07, ...,\n",
      "         5.3478306e-04, -6.7919675e-05,  1.2489549e-04],\n",
      "       [ 6.0247794e-07,  8.3074264e-07,  2.2368301e-06, ...,\n",
      "         1.9803438e-04, -4.2497941e-05,  2.1099999e-05],\n",
      "       [-8.7374298e-07, -2.4361475e-07, -8.5159070e-07, ...,\n",
      "         2.0492048e-04, -8.1235466e-06,  5.3313415e-05],\n",
      "       ...,\n",
      "       [ 8.6516746e-07,  1.2073530e-06,  1.7734902e-07, ...,\n",
      "        -1.6701113e-05,  1.7718284e-06, -4.6247910e-06],\n",
      "       [-7.0542287e-06, -1.0609388e-05, -1.7657504e-06, ...,\n",
      "         1.9130670e-04, -1.5750451e-05,  4.5480032e-05],\n",
      "       [-1.3871238e-05, -9.0921312e-06,  2.0370640e-06, ...,\n",
      "         2.2384078e-04, -9.5293517e-06, -2.3459083e-06]], dtype=float32)>, <tf.Tensor: id=905715, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 1.1982396e-05, -2.8239440e-06,  3.9582601e-06, ...,\n",
      "       -1.6989583e-05,  1.9459912e-04,  2.2986744e-04], dtype=float32)>, <tf.Tensor: id=905618, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-7.4415215e-07,  1.5988022e-05, -8.8516555e-07, ...,\n",
      "        -7.9492429e-07, -1.3907556e-06, -2.2077602e-06],\n",
      "       [-1.3533431e-06, -6.7295863e-05, -1.6256967e-06, ...,\n",
      "        -1.3497262e-06, -2.6625662e-06, -4.0814766e-06],\n",
      "       [-4.6024991e-07,  8.2663406e-05, -5.9122260e-07, ...,\n",
      "        -5.9367301e-07, -1.0391556e-06, -1.4198928e-06],\n",
      "       ...,\n",
      "       [ 6.1795149e-06,  1.5812091e-04,  6.0536318e-06, ...,\n",
      "         6.3611942e-06,  1.0141575e-05,  1.4613517e-05],\n",
      "       [-6.3836467e-07,  1.7109252e-05, -6.3367492e-07, ...,\n",
      "        -6.7287056e-07, -9.7576844e-07, -1.5470548e-06],\n",
      "       [ 7.8411870e-07, -1.3904422e-04,  1.0254693e-06, ...,\n",
      "         8.1411133e-07,  1.5253424e-06,  2.4913884e-06]], dtype=float32)>, <tf.Tensor: id=905610, shape=(110,), dtype=float32, numpy=\n",
      "array([ 6.33798481e-06,  1.65705220e-04,  6.18048989e-06,  3.72482791e-05,\n",
      "       -5.74859045e-03,  3.18937782e-05, -3.23262706e-04,  7.70043152e-06,\n",
      "        7.56702275e-06,  1.14444356e-05,  1.14814189e-04,  4.48817649e-04,\n",
      "        3.79125006e-04,  2.09161790e-05,  1.68656607e-05, -4.72121930e-04,\n",
      "       -1.24637503e-03,  1.22069649e-03, -6.12226868e-05,  5.17684792e-04,\n",
      "        2.64877199e-05,  5.27637385e-05,  2.00318482e-05,  2.55561918e-05,\n",
      "        1.88368504e-04,  2.69690529e-04,  4.92450636e-05,  2.51852500e-04,\n",
      "       -1.33977024e-04,  2.73808924e-04, -1.15842031e-05,  1.48896406e-05,\n",
      "        1.21763851e-05,  1.75416299e-05, -1.13132053e-04,  1.00311846e-03,\n",
      "        2.56987289e-04,  1.09419540e-04, -9.59287281e-05,  1.50250024e-04,\n",
      "        2.22646631e-04,  7.36543279e-06, -3.38603859e-04,  9.16460587e-04,\n",
      "        8.14815867e-05, -5.95757047e-05, -4.74097586e-04, -2.69816170e-04,\n",
      "        1.66506943e-05, -2.64672475e-04,  1.66182697e-04,  6.95690542e-05,\n",
      "       -1.33825422e-04, -1.70569518e-04, -2.83861445e-04,  1.09599197e-04,\n",
      "       -1.59032952e-05,  1.76906993e-04, -1.07795684e-04, -1.82596952e-04,\n",
      "        3.07895389e-05,  2.42650531e-05, -2.19564772e-05,  1.98895759e-05,\n",
      "       -2.96012266e-04,  5.83064882e-03,  3.60792619e-04, -1.78835681e-03,\n",
      "        7.57090631e-04, -8.88523366e-03,  5.06548909e-04,  1.66120473e-04,\n",
      "        1.21669087e-03, -6.73960475e-03, -7.33047898e-04, -1.94661005e-03,\n",
      "        7.75953289e-03, -1.54314050e-03,  6.50321657e-04,  7.56380032e-05,\n",
      "       -1.56286254e-03,  2.21303519e-04, -3.66657972e-03,  5.18615311e-03,\n",
      "        6.12145616e-03, -3.32833338e-03,  1.53880077e-03, -1.08073535e-03,\n",
      "        2.21823051e-04,  3.14602675e-03,  6.25300861e-04,  9.00155374e-06,\n",
      "        7.05345337e-06,  1.04956898e-05,  8.55829694e-06,  6.60121896e-06,\n",
      "        8.42429745e-06,  6.76032005e-06,  6.61405102e-06,  7.70739825e-06,\n",
      "        8.27780786e-06,  7.54803114e-06,  8.18243734e-06,  6.74828561e-06,\n",
      "        1.09106204e-05,  6.56099564e-06,  7.50624804e-06,  6.51745768e-06,\n",
      "        1.03624270e-05,  1.49099924e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc780>, <tf.Tensor: id=906074, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 1.2946228e-07, -9.7483496e-07, -9.8841008e-07, ...,\n",
      "        -5.8052785e-05,  2.6487744e-05,  4.0713181e-05],\n",
      "       [-3.1074268e-07, -2.2163837e-07,  1.2806112e-07, ...,\n",
      "        -1.0841044e-05,  9.4485331e-06,  2.6616835e-05],\n",
      "       [-5.3022842e-07, -1.8358031e-07,  1.2695787e-07, ...,\n",
      "        -5.7367248e-05,  4.1848209e-05, -4.6236513e-05],\n",
      "       ...,\n",
      "       [ 3.9616856e-07,  1.7018027e-07, -4.1065235e-08, ...,\n",
      "        -7.0783271e-06,  1.0495731e-05, -1.7519355e-06],\n",
      "       [-1.5773169e-07, -2.3447150e-08,  1.0660577e-07, ...,\n",
      "        -5.9880989e-05, -5.0948121e-05, -8.7959394e-05],\n",
      "       [-8.0301260e-07,  4.4759804e-07, -1.2999448e-07, ...,\n",
      "        -2.8001381e-05,  3.0614576e-06,  4.3213554e-06]], dtype=float32)>, <tf.Tensor: id=906075, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-5.20325466e-06, -9.86391115e-06, -3.04443256e-06, ...,\n",
      "         4.31408349e-04, -6.74582552e-05,  1.68900209e-04],\n",
      "       [ 8.29969395e-07,  1.58484056e-06,  1.52053360e-06, ...,\n",
      "        -3.62364139e-04,  4.67008831e-05, -1.50429129e-04],\n",
      "       [-1.11815893e-06,  1.59733773e-08, -3.03030060e-06, ...,\n",
      "         2.51522986e-04, -2.30634814e-05,  5.06237047e-05],\n",
      "       ...,\n",
      "       [-3.50038732e-07,  2.43968799e-07,  1.37689980e-07, ...,\n",
      "        -2.03999662e-05,  2.11824886e-06, -2.28086014e-06],\n",
      "       [ 3.97697295e-06,  7.45548448e-07,  1.49916341e-05, ...,\n",
      "        -3.27421207e-04,  2.81490647e-05, -6.73563482e-05],\n",
      "       [ 1.25457145e-05,  2.01349812e-05, -5.40673773e-06, ...,\n",
      "        -4.16089839e-04,  3.35902841e-05,  7.42150587e-05]], dtype=float32)>, <tf.Tensor: id=906076, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 2.5540176e-05, -6.7935771e-06,  5.0761032e-06, ...,\n",
      "       -2.0457892e-05, -3.3226237e-04, -4.2144829e-04], dtype=float32)>, <tf.Tensor: id=905979, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-7.67715221e-07,  2.82340297e-05, -9.08778929e-07, ...,\n",
      "        -8.35634694e-07, -1.45633339e-06, -2.24949690e-06],\n",
      "       [-1.35483731e-06,  5.59690307e-06, -1.60602895e-06, ...,\n",
      "        -1.34981269e-06, -2.69076372e-06, -4.04655293e-06],\n",
      "       [-4.84465602e-07,  1.92284897e-05, -6.28696341e-07, ...,\n",
      "        -6.65328571e-07, -1.12970110e-06, -1.47976209e-06],\n",
      "       ...,\n",
      "       [ 6.17095338e-06, -1.30409026e-03,  6.10673169e-06, ...,\n",
      "         6.32383399e-06,  1.04371575e-05,  1.46064576e-05],\n",
      "       [-5.79591415e-07,  1.28580345e-04, -6.19693765e-07, ...,\n",
      "        -6.15972056e-07, -9.55994437e-07, -1.52875077e-06],\n",
      "       [ 7.66119740e-07,  6.66406355e-04,  9.98727273e-07, ...,\n",
      "         7.79667289e-07,  1.46468199e-06,  2.50512676e-06]], dtype=float32)>, <tf.Tensor: id=905971, shape=(110,), dtype=float32, numpy=\n",
      "array([ 6.3156481e-06, -1.3737064e-03,  6.2282438e-06,  1.5483683e-03,\n",
      "       -9.6448790e-03,  3.2250606e-05,  1.8387004e-04,  7.6173383e-06,\n",
      "        7.6293436e-06,  1.1533237e-05,  8.0881895e-05, -2.2091492e-05,\n",
      "        2.7080087e-04,  2.1398500e-05,  1.6584901e-05, -1.3680626e-03,\n",
      "        6.3641564e-05, -1.6873399e-03,  1.0048994e-04,  4.4789173e-05,\n",
      "        3.6992034e-04,  5.4501416e-04,  3.1146010e-06, -4.0510535e-05,\n",
      "        1.1213091e-04,  2.0671246e-04, -1.5515400e-04,  2.2840736e-04,\n",
      "       -3.8574892e-04,  3.7687042e-04, -4.8923946e-04,  1.5118670e-05,\n",
      "        1.2030229e-05,  1.7989965e-05,  3.7954429e-05, -5.0413568e-04,\n",
      "        3.7540571e-04, -2.0160271e-04,  2.3958497e-04,  1.6021005e-04,\n",
      "       -7.1576494e-04,  3.3032824e-04, -9.1107754e-04,  1.6733873e-04,\n",
      "        2.4365998e-04, -2.1912841e-04,  3.0844763e-04, -5.5711734e-04,\n",
      "       -1.4089535e-04,  8.7064574e-05, -3.4000460e-04,  6.9761867e-05,\n",
      "        1.5367016e-04,  4.7993995e-04,  6.0042762e-04, -4.5034991e-05,\n",
      "        1.4242581e-04, -7.3305226e-04,  4.8973696e-05, -1.8046350e-05,\n",
      "        3.1879394e-05, -2.7808148e-04, -3.3509178e-04,  2.1636435e-05,\n",
      "        1.7218812e-05,  1.0791327e-02, -1.9191068e-03,  2.9071614e-03,\n",
      "       -2.0150566e-03, -1.2538172e-02,  5.5412999e-03, -6.0473260e-04,\n",
      "       -1.4628918e-03,  6.6216855e-04, -6.3647237e-04, -1.3480179e-03,\n",
      "        2.5936360e-03,  1.9089296e-03,  2.6506092e-04,  7.0709079e-03,\n",
      "        8.1963453e-04,  2.0129589e-04, -7.5320820e-03,  9.5248281e-05,\n",
      "        3.3981006e-03,  2.5289999e-03,  9.6425100e-04,  2.6089256e-06,\n",
      "       -5.2725489e-04,  7.5679558e-04,  2.7827852e-04,  9.3519229e-06,\n",
      "        6.9625116e-06,  1.0479480e-05,  8.2830657e-06,  6.5901472e-06,\n",
      "        8.3642026e-06,  6.4632927e-06,  6.5413842e-06,  7.5605826e-06,\n",
      "        8.2008373e-06,  7.4254722e-06,  8.3200284e-06,  6.7090996e-06,\n",
      "        1.0788141e-05,  6.6040170e-06,  7.5619250e-06,  6.4658539e-06,\n",
      "        1.0656430e-05,  1.4883807e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc198>, <tf.Tensor: id=906435, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-2.1212340e-06, -2.9876944e-07,  3.1086842e-07, ...,\n",
      "        -3.7631959e-05, -6.2884152e-05,  3.7008667e-05],\n",
      "       [ 2.5491403e-07,  2.9928117e-07,  1.6564231e-07, ...,\n",
      "        -1.3553894e-05,  4.1997671e-05,  2.3274461e-05],\n",
      "       [-5.8662937e-07, -4.3207677e-07, -2.7402510e-07, ...,\n",
      "        -2.2501697e-05,  1.8125291e-05,  3.8855123e-07],\n",
      "       ...,\n",
      "       [-6.1723381e-07, -4.5682420e-08,  8.1348318e-08, ...,\n",
      "        -4.8659949e-06,  2.4644839e-06,  8.3738669e-06],\n",
      "       [-5.3378418e-08,  2.5602093e-08,  2.9256167e-08, ...,\n",
      "        -3.8578703e-05,  3.1060074e-05, -5.7190591e-05],\n",
      "       [-7.7137219e-07,  2.0604867e-07,  1.6982243e-07, ...,\n",
      "         2.0606352e-05, -6.1454797e-05,  1.8113311e-05]], dtype=float32)>, <tf.Tensor: id=906436, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 1.7021880e-06, -6.5378040e-06,  2.3560328e-06, ...,\n",
      "        -4.5446624e-04,  4.0328436e-05, -1.0731393e-04],\n",
      "       [ 7.1559960e-08, -1.0614908e-07,  4.4124154e-06, ...,\n",
      "         1.1777189e-05,  1.8542758e-06, -1.5127940e-05],\n",
      "       [-9.5631594e-07, -6.3255459e-07, -2.8770223e-06, ...,\n",
      "        -3.5731599e-04,  4.2638520e-05,  6.3123734e-05],\n",
      "       ...,\n",
      "       [-1.6736001e-07, -3.8364286e-07, -4.4738692e-07, ...,\n",
      "        -3.1018415e-05,  3.6601668e-06, -4.6329319e-06],\n",
      "       [-4.4262620e-06,  1.2723257e-06,  6.1236001e-06, ...,\n",
      "        -5.1386334e-04,  7.4890784e-05, -1.8259113e-04],\n",
      "       [ 3.7935086e-05,  7.2850351e-05, -5.8553607e-05, ...,\n",
      "        -2.1623883e-04,  1.7308777e-05, -1.5601994e-05]], dtype=float32)>, <tf.Tensor: id=906437, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 1.9074238e-05, -6.1288706e-06,  3.9463694e-06, ...,\n",
      "       -3.1976375e-05, -5.2556675e-04, -2.1921530e-04], dtype=float32)>, <tf.Tensor: id=906340, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.51393304e-07,  1.65680176e-05, -1.06867856e-06, ...,\n",
      "        -9.91786237e-07, -1.68983047e-06, -2.54430256e-06],\n",
      "       [-1.31304762e-06, -1.05737621e-04, -1.57969384e-06, ...,\n",
      "        -1.26247346e-06, -2.61280775e-06, -4.00962290e-06],\n",
      "       [-7.59537102e-07, -1.52726861e-05, -8.76671038e-07, ...,\n",
      "        -9.42662837e-07, -1.58055218e-06, -1.91088793e-06],\n",
      "       ...,\n",
      "       [ 8.13848419e-06,  4.02478559e-04,  7.64655215e-06, ...,\n",
      "         7.99820464e-06,  1.28338479e-05,  1.72010623e-05],\n",
      "       [-8.17249656e-07, -2.93746052e-06, -8.17952241e-07, ...,\n",
      "        -8.28008979e-07, -1.24365715e-06, -1.87301077e-06],\n",
      "       [ 4.58453400e-07, -1.12197566e-04,  7.94639959e-07, ...,\n",
      "         4.84759539e-07,  1.03661955e-06,  2.13142971e-06]], dtype=float32)>, <tf.Tensor: id=906332, shape=(110,), dtype=float32, numpy=\n",
      "array([ 8.45045815e-06,  4.21959790e-04,  7.90227114e-06, -8.88812792e-05,\n",
      "       -3.28735332e-04,  3.84196537e-05, -8.31793644e-04,  9.39903657e-06,\n",
      "        9.83212249e-06,  1.30401231e-05, -5.56477753e-04, -8.01309943e-04,\n",
      "       -7.25128746e-04, -1.31166569e-04,  1.92852513e-05,  3.14156502e-03,\n",
      "       -3.69765854e-04,  2.78345798e-03,  1.18079195e-04, -9.24905180e-05,\n",
      "       -2.75802537e-04, -3.28875409e-04, -2.11157021e-04,  5.58224390e-04,\n",
      "        3.00242216e-04, -4.12818597e-04, -3.64161795e-04, -7.93283514e-04,\n",
      "       -3.25282861e-04,  5.52520796e-04,  1.78152841e-04,  1.95910379e-05,\n",
      "        1.46740222e-05,  2.29855523e-05,  3.83015053e-04, -3.13455093e-04,\n",
      "        2.04530283e-04, -5.20220674e-05,  8.09372941e-05, -2.90726341e-04,\n",
      "       -8.65433831e-05,  2.04027929e-05,  5.22603339e-04,  5.71631186e-04,\n",
      "        8.27851836e-05, -2.14265092e-04, -4.67739708e-04, -3.64858453e-04,\n",
      "        1.79544601e-04,  4.14236390e-04, -4.83301701e-05, -8.18361950e-05,\n",
      "       -3.30894400e-04,  1.21404606e-04, -7.91042985e-04,  1.15498355e-04,\n",
      "        1.46997467e-04, -2.88823183e-04,  5.47610070e-05, -6.13972679e-06,\n",
      "        3.41123632e-05,  4.03575614e-05, -6.75407136e-07,  2.88826323e-05,\n",
      "        2.05112501e-05,  3.85320513e-04,  9.98663250e-04, -1.82874931e-03,\n",
      "       -6.17028796e-04, -3.46208457e-03,  2.57565815e-04,  1.07405474e-03,\n",
      "       -5.82272513e-03,  2.74626841e-03,  1.79036972e-04, -1.30446372e-03,\n",
      "        2.01740768e-03,  2.09593825e-04, -3.03248363e-03,  1.17267072e-02,\n",
      "       -3.15972348e-03,  3.27258080e-04, -7.16435490e-03, -9.88282845e-04,\n",
      "       -2.32716498e-04,  4.38925670e-03,  5.26431948e-04,  1.85083156e-03,\n",
      "       -3.11816664e-04,  3.67656205e-04, -5.98353741e-04,  1.19464175e-05,\n",
      "        8.56519546e-06,  1.35289429e-05,  1.11417112e-05,  8.42064128e-06,\n",
      "        1.11654508e-05,  8.80275093e-06,  8.72230248e-06,  9.64809806e-06,\n",
      "        1.02634813e-05,  9.90866738e-06,  9.91359320e-06,  8.65193397e-06,\n",
      "        1.35501514e-05,  8.43266207e-06,  9.32637795e-06,  8.28825887e-06,\n",
      "        1.32466103e-05,  1.76976628e-05], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafcd68>, <tf.Tensor: id=906796, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-3.7095547e-06, -1.5464244e-06,  8.5828452e-07, ...,\n",
      "         1.3570619e-05, -4.8026992e-05,  1.7781982e-05],\n",
      "       [ 5.0436546e-07,  3.2037758e-07, -8.6281730e-08, ...,\n",
      "        -1.8491337e-05, -2.2845070e-05, -7.3473348e-06],\n",
      "       [-2.4627019e-07,  7.4877704e-08,  2.8932753e-08, ...,\n",
      "         5.0602066e-06,  3.4749919e-05,  2.4964602e-05],\n",
      "       ...,\n",
      "       [ 4.7561326e-07,  4.7790952e-08,  2.6118789e-08, ...,\n",
      "        -8.9730092e-06, -2.4203305e-07, -7.4695649e-06],\n",
      "       [ 1.8934742e-08,  1.4350778e-08,  3.7759738e-09, ...,\n",
      "         5.1170962e-05,  3.8672813e-05,  3.2510110e-05],\n",
      "       [-3.9497260e-07,  8.2805116e-08,  1.9404442e-07, ...,\n",
      "        -4.0780486e-07, -3.7250513e-05, -2.2366088e-05]], dtype=float32)>, <tf.Tensor: id=906797, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-8.97606060e-06, -2.76679589e-06, -8.82748009e-06, ...,\n",
      "        -3.02487286e-04,  2.75465572e-05, -1.00375721e-04],\n",
      "       [-3.21964194e-07, -1.28428405e-06, -8.68112181e-07, ...,\n",
      "         2.93081219e-04, -4.29786196e-05,  1.20743134e-04],\n",
      "       [ 7.36842367e-07,  1.15976277e-07, -2.93586510e-07, ...,\n",
      "        -5.59063221e-04,  7.77107562e-05, -8.08731929e-05],\n",
      "       ...,\n",
      "       [ 2.24273265e-07,  1.64271654e-07,  7.08070544e-08, ...,\n",
      "         2.30914338e-05, -9.97667939e-07,  6.88583168e-06],\n",
      "       [-3.63321504e-07,  6.16687089e-07, -3.48550884e-06, ...,\n",
      "        -1.08855129e-04, -1.65497340e-07,  1.64508538e-06],\n",
      "       [-2.86985710e-06, -2.94069196e-06,  1.60246182e-05, ...,\n",
      "         5.11583523e-04, -7.75430890e-05,  8.93193283e-05]], dtype=float32)>, <tf.Tensor: id=906798, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 1.9105031e-05,  8.9565719e-07, -1.3703947e-06, ...,\n",
      "        2.3368679e-05, -1.1143711e-04,  5.1931513e-04], dtype=float32)>, <tf.Tensor: id=906701, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.8954050e-07, -9.1658920e-05, -1.0296327e-06, ...,\n",
      "        -9.3328794e-07, -1.6284970e-06,  4.4094464e-05],\n",
      "       [-1.4065999e-06, -1.0501109e-04, -1.6805520e-06, ...,\n",
      "        -1.3492649e-06, -2.7887322e-06,  8.7192057e-05],\n",
      "       [-5.9350998e-07, -8.7506807e-05, -7.1877218e-07, ...,\n",
      "        -7.3924400e-07, -1.3247874e-06,  9.8821176e-05],\n",
      "       ...,\n",
      "       [ 6.9049293e-06,  9.8682998e-04,  6.6347975e-06, ...,\n",
      "         6.7379960e-06,  1.1242237e-05, -1.3676177e-04],\n",
      "       [-7.5425930e-07, -1.3881843e-04, -7.4231662e-07, ...,\n",
      "        -7.4798072e-07, -1.1398954e-06,  2.5929603e-05],\n",
      "       [ 7.8007486e-07, -5.9791164e-05,  1.0920378e-06, ...,\n",
      "         7.8405003e-07,  1.5196438e-06,  4.2389016e-05]], dtype=float32)>, <tf.Tensor: id=906693, shape=(110,), dtype=float32, numpy=\n",
      "array([ 7.07405752e-06,  1.02318439e-03,  6.77240860e-06, -8.84680543e-04,\n",
      "        8.24220758e-03,  3.70090165e-05,  5.66743256e-04,  8.17855926e-06,\n",
      "       -1.48011764e-04,  1.18689650e-05, -1.08738418e-03, -5.95986086e-04,\n",
      "       -8.75772268e-04, -6.01565233e-04,  1.74500437e-05,  1.41828263e-03,\n",
      "       -1.43870583e-03,  5.43411216e-03,  1.19722579e-04,  4.90497856e-04,\n",
      "       -2.43928313e-04,  3.25818197e-04, -1.00381796e-04,  2.18925725e-05,\n",
      "       -3.98513454e-04,  2.55546154e-04,  4.21318255e-05, -6.53034076e-05,\n",
      "        1.51760862e-04,  3.31538438e-04, -2.69663462e-04,  1.77378079e-05,\n",
      "        1.33179910e-05,  2.12318264e-05,  7.06898863e-05, -9.34737036e-05,\n",
      "        3.90308196e-06, -5.88433759e-05,  6.94128976e-05, -1.38098592e-04,\n",
      "        3.74551950e-04, -2.86966097e-04, -2.58641725e-04,  5.99868305e-04,\n",
      "        2.26935925e-04, -3.72449722e-04,  1.45761369e-04, -3.40353407e-04,\n",
      "        1.25772222e-05,  8.56295301e-05, -7.49695391e-05,  7.30695610e-05,\n",
      "       -1.98824651e-04, -2.37528031e-04, -7.56815134e-05, -2.02036230e-04,\n",
      "        1.41248936e-04,  4.82018979e-04,  4.93992193e-05, -3.28637514e-04,\n",
      "        3.38517639e-05,  3.35355980e-05, -8.10165784e-06,  2.60390188e-05,\n",
      "        1.91591353e-05, -8.47445615e-03,  2.77177955e-04,  4.82775550e-03,\n",
      "       -7.81176891e-03,  7.89022446e-03,  4.36567207e-05, -4.12840222e-04,\n",
      "       -3.22275329e-03,  1.82507618e-03, -1.58501207e-05,  2.69013719e-04,\n",
      "        2.92453729e-03,  3.52758798e-06, -8.39272141e-03,  7.07243336e-03,\n",
      "       -5.09443693e-03, -3.09217285e-04,  3.77042289e-03, -2.51618586e-03,\n",
      "       -6.15812279e-03,  2.32245875e-04,  2.15180009e-03,  2.29946314e-03,\n",
      "       -7.15970469e-04, -1.38160377e-03, -4.12895257e-04,  1.02817085e-05,\n",
      "        7.50149593e-06, -1.44542500e-04,  9.05965771e-06,  7.10180529e-06,\n",
      "        9.41591770e-06, -1.48984473e-04,  7.18638739e-06,  8.07141078e-06,\n",
      "        8.86780253e-06,  8.24965264e-06,  8.77123966e-06,  7.19867012e-06,\n",
      "        1.20387140e-05,  7.23668745e-06,  8.32672049e-06,  6.89651279e-06,\n",
      "        1.14810091e-05, -1.39899246e-04], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafce48>, <tf.Tensor: id=907157, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 6.9590163e-07, -2.1512474e-06,  8.0104115e-07, ...,\n",
      "        -1.0926700e-05, -9.9481571e-05, -8.7271828e-06],\n",
      "       [-3.3856662e-07, -8.0639069e-08,  2.8053268e-08, ...,\n",
      "         2.8283209e-06, -1.5483354e-05, -1.6921925e-06],\n",
      "       [-4.1374622e-07, -1.4351755e-07, -1.6006364e-07, ...,\n",
      "         2.0592635e-05,  4.0214265e-05,  3.1642521e-05],\n",
      "       ...,\n",
      "       [ 3.3751621e-07,  3.4311014e-08, -4.2098154e-08, ...,\n",
      "         1.2934171e-05, -7.5068697e-06, -1.1985285e-05],\n",
      "       [ 1.4489730e-08,  1.0857577e-07,  3.2928710e-08, ...,\n",
      "         4.2008967e-05,  2.2364324e-05,  4.6408692e-05],\n",
      "       [-3.0200617e-08,  4.0110419e-07, -4.0314819e-07, ...,\n",
      "        -2.9661887e-05, -2.1149481e-05,  4.1916901e-05]], dtype=float32)>, <tf.Tensor: id=907158, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-1.3067406e-06, -2.3712487e-06, -4.3965733e-06, ...,\n",
      "        -3.4948793e-04,  2.8338134e-05, -1.1184397e-04],\n",
      "       [-2.0856451e-06, -3.6477811e-06, -2.4063165e-06, ...,\n",
      "         3.6239100e-04, -3.2867392e-05, -1.2254721e-05],\n",
      "       [ 7.3238752e-08, -1.3636613e-06, -1.2051222e-06, ...,\n",
      "        -3.4893796e-04,  2.3083441e-05,  6.5153778e-05],\n",
      "       ...,\n",
      "       [ 8.1151813e-07,  6.4150373e-07,  6.9324460e-07, ...,\n",
      "         4.9134422e-05, -4.7838221e-06,  1.1123084e-05],\n",
      "       [-1.5313958e-06, -1.7956190e-05, -1.3388967e-05, ...,\n",
      "         5.3624997e-05, -1.1036354e-05,  1.6279637e-05],\n",
      "       [-3.3191172e-06, -1.1694141e-05,  1.8196402e-05, ...,\n",
      "        -2.5511609e-04,  1.7722183e-05, -2.0416268e-05]], dtype=float32)>, <tf.Tensor: id=907159, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 1.9665074e-05,  9.9495983e-06,  1.2363163e-06, ...,\n",
      "        4.9951934e-05,  5.3738790e-05, -2.5902412e-04], dtype=float32)>, <tf.Tensor: id=907062, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.5862990e-07, -2.5402802e-05, -1.0338688e-06, ...,\n",
      "        -8.9660875e-07, -1.6575416e-06, -2.5855938e-06],\n",
      "       [-1.3466714e-06, -2.1764287e-04, -1.6768313e-06, ...,\n",
      "        -1.3053456e-06, -2.8478248e-06, -4.3177643e-06],\n",
      "       [-5.1823565e-07,  1.5724718e-04, -6.8176473e-07, ...,\n",
      "        -6.5302885e-07, -1.2748510e-06, -1.7040784e-06],\n",
      "       ...,\n",
      "       [ 6.2959434e-06,  1.4420605e-04,  6.4101296e-06, ...,\n",
      "         6.1938035e-06,  1.0979372e-05,  1.5463171e-05],\n",
      "       [-6.4620855e-07,  4.8973394e-05, -6.8501407e-07, ...,\n",
      "        -6.5551058e-07, -1.0423122e-06, -1.7038278e-06],\n",
      "       [ 8.5656518e-07,  1.3383824e-04,  1.1198736e-06, ...,\n",
      "         8.4069313e-07,  1.6927154e-06,  2.7864544e-06]], dtype=float32)>, <tf.Tensor: id=907054, shape=(110,), dtype=float32, numpy=\n",
      "array([ 6.43547446e-06,  1.35331124e-04,  6.53197822e-06, -2.66495394e-04,\n",
      "        6.05097041e-03,  3.60479353e-05,  8.37479660e-04,  7.77713012e-06,\n",
      "        7.88597754e-06,  1.18768712e-05,  7.03068145e-05,  4.64000797e-04,\n",
      "       -1.04993276e-04,  2.36084688e-05,  1.68543647e-05, -1.83572178e-03,\n",
      "       -3.33489850e-04,  2.63919681e-03,  1.12959911e-04,  2.53084174e-04,\n",
      "       -4.21523699e-04,  9.23168031e-04,  5.02564581e-05, -1.14590157e-05,\n",
      "        2.29206271e-04,  4.07835847e-04, -5.96504600e-04, -3.79492354e-04,\n",
      "        1.77778347e-05,  2.06104305e-04,  2.01270974e-04,  1.68985171e-05,\n",
      "        1.24209364e-05,  2.07247467e-05, -1.05525134e-04,  8.07918877e-06,\n",
      "        3.42005776e-04,  4.10811830e-04,  2.20229267e-04, -2.87724484e-04,\n",
      "        2.10530154e-04,  1.98438764e-04,  7.54063512e-05, -1.40016747e-03,\n",
      "       -9.95407754e-05, -2.13705454e-04, -2.32624698e-05, -4.67391419e-06,\n",
      "        4.72062629e-06,  2.17724199e-04,  3.83855513e-04,  7.44295467e-05,\n",
      "        9.71376503e-05,  3.56951758e-04,  4.56842303e-04,  1.12117988e-04,\n",
      "       -3.30556388e-04,  4.62381140e-04,  4.74091612e-05, -4.81188006e-04,\n",
      "        3.49301226e-05,  1.93881249e-04,  1.40075979e-04,  2.48765919e-05,\n",
      "        1.87852220e-05, -7.60632893e-03, -6.64287247e-04,  5.14292996e-03,\n",
      "       -2.72563030e-03,  1.76153295e-02, -3.45611596e-03,  1.24655978e-03,\n",
      "       -7.76238856e-04,  1.44857529e-03, -2.81675602e-04,  4.83756419e-04,\n",
      "       -3.62368557e-03,  4.54448193e-04,  2.28234380e-03, -4.52041114e-03,\n",
      "       -2.39429879e-03,  4.89043654e-04,  2.70815031e-03, -3.02924099e-03,\n",
      "       -1.14455279e-02, -2.73887790e-03,  1.09321799e-03,  1.17802864e-03,\n",
      "        9.34861891e-04, -2.46358942e-03,  5.24031871e-04,  9.86691430e-06,\n",
      "        7.11103212e-06,  1.09530856e-05,  8.26958512e-06,  6.76339278e-06,\n",
      "        8.67128074e-06,  6.57114924e-06,  6.77179651e-06,  7.51929429e-06,\n",
      "        8.34505954e-06,  7.52310734e-06,  8.33158902e-06,  6.85536997e-06,\n",
      "        1.15637495e-05,  6.81688925e-06,  7.83091673e-06,  6.32627780e-06,\n",
      "        1.11947365e-05,  1.57425220e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafce10>, <tf.Tensor: id=907518, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-5.21308948e-06,  1.22992697e-06,  9.41323549e-07, ...,\n",
      "        -1.77006987e-05, -5.60388326e-05,  6.16786565e-06],\n",
      "       [ 1.75978911e-07,  6.36322170e-07, -4.32127848e-07, ...,\n",
      "         1.89572238e-05, -6.34957687e-05, -2.87899347e-05],\n",
      "       [-1.69585110e-06,  3.47393666e-07,  5.71698081e-07, ...,\n",
      "         2.58413329e-05, -2.79308806e-05,  2.02310293e-05],\n",
      "       ...,\n",
      "       [ 1.02030974e-07, -4.56325836e-07,  1.86360182e-07, ...,\n",
      "         2.84347188e-05, -3.07999726e-05,  1.31240286e-05],\n",
      "       [-1.67092253e-07, -1.13444266e-07,  6.71583180e-08, ...,\n",
      "         9.26391003e-05, -2.05906190e-06,  3.98301017e-05],\n",
      "       [ 1.05164389e-07,  3.82887251e-08, -4.76348021e-07, ...,\n",
      "         4.61209584e-06, -1.85646422e-05, -5.03711135e-07]], dtype=float32)>, <tf.Tensor: id=907519, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-5.7827651e-06, -6.5914915e-06,  2.6332964e-06, ...,\n",
      "        -7.6752953e-04,  9.3266135e-05, -7.0986433e-05],\n",
      "       [-5.5378860e-07, -4.4174622e-06,  9.8924329e-07, ...,\n",
      "        -4.0179619e-04,  1.7395751e-05,  8.4851628e-05],\n",
      "       [-3.4265528e-07,  6.6089098e-07, -7.7692994e-07, ...,\n",
      "        -1.1825757e-04,  1.5512380e-06, -1.6294201e-05],\n",
      "       ...,\n",
      "       [-3.2546936e-06,  1.8002387e-06, -4.7879377e-07, ...,\n",
      "         3.7252909e-04, -3.5673474e-05,  7.3737254e-05],\n",
      "       [ 8.7326825e-06,  5.9968461e-06,  2.4333865e-05, ...,\n",
      "        -4.8175271e-04,  3.9445542e-05, -6.5227368e-06],\n",
      "       [-5.7938628e-06, -9.3057020e-07, -1.2861011e-06, ...,\n",
      "        -1.5404963e-04, -1.1120346e-05, -5.9109652e-06]], dtype=float32)>, <tf.Tensor: id=907520, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 9.8662449e-06, -4.2086026e-06,  3.4375141e-06, ...,\n",
      "        4.8742018e-04, -6.6767476e-04, -1.4170122e-04], dtype=float32)>, <tf.Tensor: id=907423, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.88854619e-06,  5.23680428e-05, -1.76763842e-06, ...,\n",
      "        -1.75977573e-06, -2.73843375e-06, -4.11148449e-06],\n",
      "       [-2.00572708e-06, -5.78759136e-05, -2.18055470e-06, ...,\n",
      "        -1.74760044e-06, -3.67646044e-06, -5.55171164e-06],\n",
      "       [-1.16249430e-06,  1.47981773e-04, -1.32106061e-06, ...,\n",
      "        -1.36672770e-06, -2.27796704e-06, -2.88918136e-06],\n",
      "       ...,\n",
      "       [ 1.60840355e-05, -3.41638981e-04,  1.34374768e-05, ...,\n",
      "         1.46294815e-05,  2.12571067e-05,  2.89486652e-05],\n",
      "       [-1.53245060e-06,  1.09904657e-04, -1.35362745e-06, ...,\n",
      "        -1.39670465e-06, -2.03124455e-06, -3.16416572e-06],\n",
      "       [ 7.46150306e-07, -1.77185982e-04,  1.01077103e-06, ...,\n",
      "         6.35050924e-07,  1.51622135e-06,  2.48098650e-06]], dtype=float32)>, <tf.Tensor: id=907415, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.97127774e-05, -4.61480027e-04,  1.60545642e-05, -6.32516167e-04,\n",
      "       -2.40085064e-05,  5.76794009e-05,  6.02111511e-04,  2.06390851e-05,\n",
      "        2.07432204e-05,  2.41519901e-05, -2.68830627e-04, -5.25309995e-04,\n",
      "       -3.87606997e-04,  4.59397197e-05,  3.40283332e-05, -6.45179127e-04,\n",
      "       -7.89010432e-04,  1.57918001e-03, -1.55737856e-04, -5.32293780e-05,\n",
      "        4.56736569e-04,  1.54368812e-03,  2.86193623e-04,  3.19193467e-04,\n",
      "        1.52836205e-04,  3.53076903e-04, -8.46365583e-06, -3.17716447e-04,\n",
      "       -2.66042101e-04,  2.10904527e-05, -9.15788623e-05,  3.79966623e-05,\n",
      "        2.93230605e-05,  4.68048856e-05, -8.82973909e-05, -1.73604593e-03,\n",
      "        1.21006509e-04, -8.71419325e-04, -1.81903888e-04, -2.18296936e-03,\n",
      "       -3.46221532e-05, -2.21198192e-04, -4.17241594e-04, -2.85802584e-04,\n",
      "        1.12738198e-04, -1.04544470e-05, -8.82761495e-04, -7.87634985e-04,\n",
      "       -1.33265590e-03,  3.70076858e-04, -8.11501639e-04,  1.02287544e-04,\n",
      "       -1.19646953e-03, -9.73729300e-04, -3.17602302e-04, -1.55189075e-04,\n",
      "        4.05548781e-05,  6.01944455e-04,  8.21291906e-05, -9.12527903e-05,\n",
      "        5.29557547e-05,  7.92181017e-05,  3.93956798e-05, -7.30890140e-04,\n",
      "        3.77493234e-05, -3.18937283e-03,  1.42491702e-03,  1.99023471e-03,\n",
      "       -4.97875316e-03,  1.22619802e-02, -7.68603932e-04, -1.73037013e-04,\n",
      "        6.63416367e-03, -5.97449858e-03,  1.10305678e-04, -3.17440223e-04,\n",
      "       -6.09491765e-03, -1.04269770e-03,  1.22887641e-03,  1.39195635e-03,\n",
      "       -1.20665354e-03,  5.28166303e-04,  8.55323114e-03,  4.24528634e-03,\n",
      "       -3.19257565e-03, -3.82667116e-04, -1.25413970e-03,  9.55868454e-05,\n",
      "        3.44056927e-04, -5.96338359e-04,  5.71143406e-04,  2.51574074e-05,\n",
      "        1.95132161e-05,  2.69340308e-05,  2.32813290e-05,  1.96715009e-05,\n",
      "        2.39267974e-05,  1.87406386e-05,  1.92947791e-05,  2.00464401e-05,\n",
      "        1.94793756e-05,  2.07026296e-05,  2.16403732e-05,  2.10388571e-05,\n",
      "        2.71545123e-05,  1.79664839e-05,  1.97216159e-05,  1.77727761e-05,\n",
      "        2.49707846e-05,  3.38925674e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc320>, <tf.Tensor: id=907879, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 2.4046476e-06,  2.1291423e-06, -1.2777223e-06, ...,\n",
      "        -2.7696438e-05,  1.7304134e-06,  3.2445281e-05],\n",
      "       [-5.0315532e-07, -2.3772782e-07,  1.7734030e-07, ...,\n",
      "         2.2882528e-05, -2.5930804e-05,  9.9043600e-06],\n",
      "       [-5.0814629e-07, -6.6486956e-07,  5.0871546e-07, ...,\n",
      "         3.1961477e-05, -2.3044069e-05,  1.1706189e-05],\n",
      "       ...,\n",
      "       [ 1.7689549e-06,  5.8874440e-07, -4.2520887e-07, ...,\n",
      "         2.9007638e-06, -6.1645028e-06, -4.9857790e-06],\n",
      "       [ 1.4208466e-08,  2.2745811e-08,  1.1327472e-08, ...,\n",
      "        -1.0093869e-05,  2.5819209e-05, -2.5223224e-05],\n",
      "       [ 9.5453254e-07,  1.0431630e-06, -7.0082916e-07, ...,\n",
      "        -4.8938386e-05,  5.6570650e-05,  3.5147893e-06]], dtype=float32)>, <tf.Tensor: id=907880, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 3.16436876e-06, -6.03710760e-06,  2.67942687e-06, ...,\n",
      "         4.46227437e-04, -4.63944998e-05,  9.52925402e-05],\n",
      "       [-6.27058000e-07,  1.21894772e-07,  1.04973526e-06, ...,\n",
      "         8.74917168e-06,  4.03338345e-06, -1.13173715e-04],\n",
      "       [-3.18329569e-07,  4.39476864e-08,  2.78836001e-06, ...,\n",
      "         2.17564258e-04, -5.46373121e-05,  6.86626445e-05],\n",
      "       ...,\n",
      "       [ 6.31174828e-07,  1.32213745e-06, -3.86186957e-06, ...,\n",
      "        -5.05244025e-05,  3.47758396e-06, -1.62917113e-05],\n",
      "       [-9.99857730e-06, -2.12044633e-05, -1.85385616e-05, ...,\n",
      "         5.18321816e-04, -5.91967146e-05,  1.32603134e-04],\n",
      "       [ 8.15810745e-06,  7.24047231e-06, -3.08290100e-06, ...,\n",
      "        -1.34682588e-04,  4.24221216e-05,  7.89810292e-05]], dtype=float32)>, <tf.Tensor: id=907881, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 8.3039477e-06, -2.9088376e-06, -6.0494476e-06, ...,\n",
      "       -5.2070969e-05,  5.2754133e-04, -1.3359319e-04], dtype=float32)>, <tf.Tensor: id=907784, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.2384071e-07,  1.2171189e-05, -1.0016132e-06, ...,\n",
      "        -8.4504859e-07, -1.6519657e-06, -2.6031112e-06],\n",
      "       [-1.5227943e-06,  5.8874335e-05, -1.8845985e-06, ...,\n",
      "        -1.4144091e-06, -3.2699268e-06, -4.8972865e-06],\n",
      "       [-4.2981895e-07, -1.5297305e-04, -5.9102098e-07, ...,\n",
      "        -5.8968760e-07, -1.2601104e-06, -1.6110137e-06],\n",
      "       ...,\n",
      "       [ 6.7679575e-06,  3.7198994e-04,  6.7027963e-06, ...,\n",
      "         6.4747528e-06,  1.1821299e-05,  1.6603753e-05],\n",
      "       [-7.2187993e-07, -1.0487103e-05, -7.2776777e-07, ...,\n",
      "        -6.9309914e-07, -1.1693480e-06, -1.8192043e-06],\n",
      "       [ 1.0467766e-06, -4.9816340e-04,  1.2870623e-06, ...,\n",
      "         9.3700118e-07,  1.9165857e-06,  3.1621155e-06]], dtype=float32)>, <tf.Tensor: id=907776, shape=(110,), dtype=float32, numpy=\n",
      "array([ 6.98458553e-06,  4.12355992e-04,  6.88079353e-06, -8.67049748e-05,\n",
      "        2.32609618e-03,  3.67505527e-05,  4.33758658e-04,  8.28716020e-06,\n",
      "        8.64407957e-06,  1.26126952e-05, -3.73209798e-04,  7.58571114e-05,\n",
      "        3.12487187e-04,  2.70105465e-05,  1.83291522e-05, -1.93080760e-03,\n",
      "        5.58546162e-04, -6.96486823e-05,  1.19434895e-04,  6.65350643e-04,\n",
      "        8.47100164e-05,  4.78384056e-04,  1.53981222e-04,  5.02276816e-05,\n",
      "        6.45219116e-04,  1.23611098e-04, -5.27861179e-04, -3.35011951e-04,\n",
      "        6.85896201e-04, -6.99780183e-04, -1.09135945e-04,  1.84255223e-05,\n",
      "        1.36913159e-05,  2.36701635e-05,  1.87275276e-04, -1.12667170e-04,\n",
      "        5.42149472e-04,  1.25454288e-04,  2.13846492e-04,  8.67854469e-05,\n",
      "       -9.19381928e-05,  3.79907171e-04,  3.43359337e-04,  1.47356803e-03,\n",
      "        2.04313576e-04, -4.01640427e-05,  3.03618959e-04,  2.56563857e-04,\n",
      "       -4.47554659e-04, -3.93579016e-04,  3.78535624e-04,  7.72292842e-05,\n",
      "       -1.81421899e-04,  4.96812048e-04,  4.90652106e-04, -3.16949518e-05,\n",
      "        1.49541433e-04, -3.18307546e-04,  5.21878392e-05, -1.19981145e-04,\n",
      "        3.71974784e-05,  2.01594128e-04, -1.71716674e-05,  2.86606646e-05,\n",
      "        2.06870009e-05, -1.66543387e-03, -8.28856719e-04, -1.58866087e-03,\n",
      "       -1.05280150e-03, -5.87946735e-03, -3.16261710e-03, -5.23433206e-04,\n",
      "        2.69989646e-03, -2.84088100e-03, -1.68362036e-04,  7.17120070e-04,\n",
      "       -6.22484600e-03,  1.22209394e-03,  3.93782835e-03, -7.76515901e-03,\n",
      "        4.84669022e-03,  5.04634343e-04,  7.17565045e-03, -9.39734222e-04,\n",
      "        3.98291368e-03,  2.14570761e-03,  1.71686790e-03, -7.11247849e-04,\n",
      "        1.10134599e-03, -3.60799674e-03, -7.38718081e-04,  1.01204896e-05,\n",
      "        7.85797329e-06,  1.21228322e-05,  8.91334184e-06,  7.47371359e-06,\n",
      "        9.27620476e-06,  7.47095919e-06,  7.38553354e-06,  8.12602138e-06,\n",
      "        8.95548601e-06,  8.16228385e-06,  9.13752956e-06,  7.52143751e-06,\n",
      "        1.27684243e-05,  7.47968625e-06,  8.36077743e-06,  6.67260019e-06,\n",
      "        1.21419298e-05,  1.70204949e-05], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc630>, <tf.Tensor: id=908240, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-9.4951531e-09,  5.6382234e-08, -1.0893188e-07, ...,\n",
      "        -1.6264185e-06, -2.9588829e-05, -8.9986128e-07],\n",
      "       [-1.9174312e-08,  7.0308765e-08, -8.9607724e-08, ...,\n",
      "        -1.3242555e-05,  4.7532096e-05,  1.7185163e-05],\n",
      "       [-1.2683271e-06, -5.3850721e-07,  4.8221096e-07, ...,\n",
      "        -4.6254750e-05,  1.8701516e-05, -2.8269857e-05],\n",
      "       ...,\n",
      "       [ 1.4078609e-06,  3.9360270e-07, -4.7312372e-07, ...,\n",
      "        -1.3644007e-05,  1.6242369e-05,  2.8195750e-06],\n",
      "       [-2.4374986e-08, -7.6373851e-08,  2.2828040e-08, ...,\n",
      "         5.3490480e-06, -1.5428830e-05, -3.5671175e-05],\n",
      "       [-1.2311067e-06,  4.8743368e-07,  3.7553835e-07, ...,\n",
      "        -4.5425448e-05,  2.4256256e-05, -5.5547430e-06]], dtype=float32)>, <tf.Tensor: id=908241, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 5.2084242e-06, -9.6435051e-06, -7.7826280e-06, ...,\n",
      "         5.2365346e-04, -4.3791442e-05,  1.7537734e-04],\n",
      "       [ 7.2148805e-07,  8.8611006e-07,  2.6124819e-06, ...,\n",
      "        -3.6257802e-04,  6.1384068e-05, -6.9678536e-05],\n",
      "       [-6.0650569e-07, -1.7666570e-06,  1.4635742e-06, ...,\n",
      "         4.3130390e-04, -5.1605082e-05,  8.7305954e-05],\n",
      "       ...,\n",
      "       [-2.4042205e-07, -2.4269079e-07, -1.0615613e-06, ...,\n",
      "        -2.7897820e-05,  2.7888832e-06, -7.4427517e-06],\n",
      "       [-7.1242090e-07, -8.8439601e-06, -1.4825713e-05, ...,\n",
      "        -1.9914334e-04,  3.0502784e-05, -1.0025204e-05],\n",
      "       [ 9.1599236e-07,  1.4314540e-05, -1.3656028e-05, ...,\n",
      "        -1.9936176e-04,  1.0541446e-05,  1.2264598e-05]], dtype=float32)>, <tf.Tensor: id=908242, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 1.6989365e-05, -6.0943730e-06, -6.5179347e-07, ...,\n",
      "       -2.8499546e-05, -2.0243521e-04, -2.0302966e-04], dtype=float32)>, <tf.Tensor: id=908145, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.57720658e-07,  1.59673684e-04, -1.10143651e-06, ...,\n",
      "        -9.63455250e-07, -1.82404244e-06,  3.72162976e-05],\n",
      "       [-1.54831844e-06,  9.88602278e-06, -1.87908267e-06, ...,\n",
      "        -1.37832239e-06, -3.34514743e-06,  7.17390649e-05],\n",
      "       [-6.28872556e-07,  1.02222271e-04, -7.57891485e-07, ...,\n",
      "        -8.30322449e-07, -1.48606387e-06,  1.10576926e-04],\n",
      "       ...,\n",
      "       [ 9.62167178e-06, -8.60558299e-04,  8.50849119e-06, ...,\n",
      "         8.85288409e-06,  1.49926027e-05, -1.34627990e-04],\n",
      "       [-1.14254806e-06,  1.91202533e-04, -1.02709464e-06, ...,\n",
      "        -1.06247649e-06, -1.66857387e-06,  2.70226901e-05],\n",
      "       [ 8.19167781e-07,  1.58065188e-04,  1.14833006e-06, ...,\n",
      "         6.90933291e-07,  1.69322800e-06, -6.73432032e-06]], dtype=float32)>, <tf.Tensor: id=908137, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.00028010e-05, -9.41041508e-04,  8.78885567e-06,  2.17881025e-04,\n",
      "       -1.61525756e-02,  4.03755766e-05,  1.82501346e-04,  1.08846125e-05,\n",
      "        1.14304094e-05,  1.49408152e-05,  6.60788850e-04,  5.25193231e-04,\n",
      "        4.05840052e-04,  3.40410334e-05,  2.24597661e-05, -1.10294798e-03,\n",
      "       -1.04584615e-04,  1.10270828e-03,  1.32701098e-04,  3.89399473e-04,\n",
      "        4.34510061e-04,  1.00066140e-03,  1.84234144e-04, -1.15361647e-04,\n",
      "       -9.51981056e-05, -1.64780024e-04, -2.02780066e-04, -3.94654526e-05,\n",
      "       -1.15759794e-04,  4.38513125e-05, -1.21543859e-03,  2.37650183e-05,\n",
      "        1.73869830e-05,  3.03874494e-05, -2.97160179e-04, -2.13601510e-04,\n",
      "        3.39851715e-04,  3.23232089e-04, -4.11097571e-04, -1.46451639e-04,\n",
      "       -2.26437478e-04,  8.11697319e-05, -3.97851720e-04,  1.14536379e-03,\n",
      "        3.58671241e-04, -2.20072579e-05,  1.86190646e-05, -1.22529775e-04,\n",
      "       -2.55967781e-04,  4.43061144e-04, -2.29650934e-04,  8.21546273e-05,\n",
      "        3.39177321e-04, -1.11638627e-04,  1.07151945e-03,  1.40995326e-04,\n",
      "        1.24253493e-05, -5.83022658e-04, -9.33902338e-05, -7.06158680e-05,\n",
      "        4.01166290e-05,  5.11926992e-05, -1.56654889e-04,  3.99881756e-05,\n",
      "        2.53435355e-05,  6.56100409e-03, -2.11626687e-03,  1.80418813e-03,\n",
      "        1.99277001e-03, -7.69318501e-03, -4.58662491e-03,  1.76791486e-03,\n",
      "        4.73181857e-03,  6.10246439e-04,  2.82345194e-04,  1.98082533e-03,\n",
      "       -1.53692102e-03,  2.30523851e-03,  3.92366853e-03, -3.20315035e-03,\n",
      "        1.01432612e-03,  1.13923464e-03, -1.55936298e-03, -4.30731522e-03,\n",
      "        7.05998018e-03, -2.65512499e-04,  1.66384387e-03,  1.68843754e-03,\n",
      "        1.74524891e-03, -1.81766856e-03,  3.07271926e-04,  1.34134734e-05,\n",
      "        1.00256129e-05,  1.61717235e-05,  1.25720526e-05,  1.01208198e-05,\n",
      "        1.25116958e-05,  1.08744880e-05,  1.02719541e-05,  1.11393065e-05,\n",
      "        1.15257362e-05,  1.13382303e-05,  1.16222427e-05,  1.02269942e-05,\n",
      "        1.61593107e-05,  9.97749248e-06,  1.07258966e-05,  9.19259037e-06,\n",
      "        1.54875925e-05, -1.35315408e-04], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc860>, <tf.Tensor: id=908601, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 2.8450827e-06, -5.9048409e-07, -2.0462514e-06, ...,\n",
      "        -4.7391623e-05, -3.4735662e-05,  1.7531280e-05],\n",
      "       [ 1.5673337e-06, -1.5696223e-06,  1.2261132e-06, ...,\n",
      "        -2.7866252e-05,  6.2496823e-05,  1.0310154e-05],\n",
      "       [ 3.3534061e-06,  8.2040771e-07,  2.5794639e-06, ...,\n",
      "        -6.3449421e-05, -2.3769595e-05, -5.3772594e-05],\n",
      "       ...,\n",
      "       [ 3.6140746e-06,  7.8411222e-07, -4.1803150e-07, ...,\n",
      "        -6.8268477e-05,  2.9658046e-05, -7.2284900e-05],\n",
      "       [-5.0575800e-06,  1.2632303e-06, -5.7339445e-08, ...,\n",
      "        -1.5010731e-05,  2.4145851e-05, -1.7219438e-05],\n",
      "       [-1.3196471e-06,  1.0971837e-06,  3.4602689e-07, ...,\n",
      "         3.6450732e-05, -4.8700793e-05, -8.2333446e-07]], dtype=float32)>, <tf.Tensor: id=908602, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 1.02327449e-05, -9.51872335e-06, -5.08079438e-06, ...,\n",
      "         9.50414280e-04, -1.11522975e-04,  4.27808292e-04],\n",
      "       [ 3.06218158e-06,  2.03369200e-06, -7.45923671e-06, ...,\n",
      "        -1.25806560e-04,  1.31998586e-05,  8.55721810e-05],\n",
      "       [ 2.48307356e-06, -3.50111520e-07,  9.04147146e-06, ...,\n",
      "         2.13775216e-04, -1.18623220e-05,  3.01197324e-05],\n",
      "       ...,\n",
      "       [ 6.81504616e-06,  3.81894688e-06, -1.93323394e-05, ...,\n",
      "        -2.64801696e-04,  1.14672066e-05, -1.90481180e-04],\n",
      "       [ 5.91180537e-07, -1.07323694e-05, -1.31180504e-05, ...,\n",
      "         2.81433109e-04, -1.86857615e-05,  2.04335069e-04],\n",
      "       [ 3.50005007e-06, -5.67011102e-06,  2.90382759e-05, ...,\n",
      "         6.93975890e-05, -5.30410471e-05,  8.69483629e-05]], dtype=float32)>, <tf.Tensor: id=908603, shape=(6144,), dtype=float32, numpy=\n",
      "array([-9.8454293e-06, -3.5190380e-05, -2.2227234e-05, ...,\n",
      "       -1.1900030e-03,  1.6256372e-03, -3.6095564e-05], dtype=float32)>, <tf.Tensor: id=908506, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[ 8.7365424e-06,  2.5845683e-04,  6.4576820e-06, ...,\n",
      "         6.7824776e-06,  5.1275920e-06,  1.3088270e-06],\n",
      "       [ 2.4303503e-05,  1.3022391e-04,  1.7744051e-05, ...,\n",
      "         1.9661000e-05,  1.5321712e-05,  6.7216724e-06],\n",
      "       [ 3.5085810e-05,  1.7633598e-04,  2.6082045e-05, ...,\n",
      "         2.7789514e-05,  2.4382702e-05,  1.5332083e-05],\n",
      "       ...,\n",
      "       [-2.8348266e-05, -1.3977590e-03, -2.0125684e-05, ...,\n",
      "        -2.2242564e-05, -1.2731672e-05,  3.9003689e-06],\n",
      "       [ 5.0177905e-06,  2.7145774e-04,  4.2704105e-06, ...,\n",
      "         4.9624960e-06,  3.3355050e-06,  2.0225905e-07],\n",
      "       [ 1.3338981e-05,  1.0083566e-04,  1.0544754e-05, ...,\n",
      "         1.0683610e-05,  1.0912061e-05,  9.4989864e-06]], dtype=float32)>, <tf.Tensor: id=908498, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.34577800e-04, -1.62161724e-03,  1.02571081e-04,  2.83592148e-04,\n",
      "        1.42895023e-03,  6.92312969e-05, -7.27020320e-04,  1.65588674e-04,\n",
      "        1.14739851e-04,  1.63262550e-04,  5.55073260e-04,  5.39317669e-04,\n",
      "        1.31933339e-04,  1.06383188e-04,  1.35496972e-04, -2.96915532e-03,\n",
      "       -1.62871787e-04, -1.31649678e-04, -6.24366476e-06,  3.88946675e-04,\n",
      "        8.91035364e-04,  1.04514649e-03,  5.62180649e-04,  5.51678299e-04,\n",
      "        2.69414624e-04,  6.75448508e-04,  4.03065525e-04,  4.79818380e-04,\n",
      "        4.06340085e-04, -4.07751155e-04,  1.15407456e-04,  1.19124714e-04,\n",
      "        1.26254832e-04,  1.01623838e-04,  2.01490548e-04, -5.05769975e-04,\n",
      "       -1.12073780e-04,  2.45837902e-04,  2.63975031e-04, -1.75868289e-03,\n",
      "        1.51860644e-04,  1.46196209e-04, -1.77364127e-05, -3.54767602e-04,\n",
      "        4.25865059e-04,  5.29147947e-05, -4.13055386e-04,  1.14343718e-04,\n",
      "       -3.49367358e-04, -9.47679742e-04, -1.74649875e-04,  2.28431636e-05,\n",
      "       -1.30989822e-03, -9.76565934e-05, -1.33491505e-03, -2.23489435e-04,\n",
      "        2.98271421e-04, -2.38777080e-04,  1.68847502e-04,  1.81975018e-04,\n",
      "       -5.44206705e-06,  2.80224252e-04,  1.45717549e-05,  8.83540633e-05,\n",
      "       -2.07422258e-04,  5.54731535e-03,  2.27332162e-03,  1.66173282e-04,\n",
      "        4.57522506e-03, -1.26358261e-02,  3.89714900e-04, -2.52107787e-03,\n",
      "        1.41271763e-03,  3.65079753e-03,  4.41584794e-04,  1.47979229e-03,\n",
      "       -1.36405509e-03,  1.77275867e-03,  2.93144432e-04, -7.87328754e-04,\n",
      "        6.61055557e-04,  2.76949926e-04, -2.73535075e-03,  1.08411739e-04,\n",
      "       -2.00850237e-03, -2.09728722e-03, -2.04177457e-03,  2.76334980e-03,\n",
      "        1.13621028e-03, -1.58552942e-03, -1.39436015e-04,  1.07975509e-04,\n",
      "        1.42899225e-04,  1.10459026e-04,  1.19393415e-04,  1.69290375e-04,\n",
      "        1.10387162e-04,  1.50617430e-04,  1.56597744e-04,  1.34933929e-04,\n",
      "        9.54029383e-05,  9.90677727e-05,  1.31866822e-04,  1.41315511e-04,\n",
      "        1.28588334e-04,  1.08500477e-04,  1.12965492e-04,  1.08841297e-04,\n",
      "        1.06940432e-04,  8.55219623e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafcdd8>, <tf.Tensor: id=908962, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 3.9888377e-07,  5.5967303e-07, -7.2516372e-08, ...,\n",
      "         7.4014019e-06,  2.2276463e-05,  4.4434200e-05],\n",
      "       [ 9.0486346e-08,  3.6228056e-07,  4.0451749e-07, ...,\n",
      "        -6.3631609e-07, -8.0928630e-06,  1.8097548e-05],\n",
      "       [ 4.2852207e-07,  4.8811831e-08, -4.6732662e-08, ...,\n",
      "        -4.1453335e-05, -5.1286361e-06, -1.9114161e-05],\n",
      "       ...,\n",
      "       [ 1.6611465e-06,  5.9506897e-07, -6.8402710e-07, ...,\n",
      "         1.5581702e-05, -1.1413418e-06,  4.9767568e-06],\n",
      "       [-4.3551302e-07, -2.9712788e-07,  1.8862498e-07, ...,\n",
      "        -2.2039862e-05,  8.1078224e-06, -1.4755744e-05],\n",
      "       [-3.3156205e-08,  9.8966416e-07,  1.0563677e-07, ...,\n",
      "        -4.0599931e-05,  6.6352608e-05,  3.2556738e-06]], dtype=float32)>, <tf.Tensor: id=908963, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 1.3628143e-05,  9.0308122e-06,  3.0723804e-06, ...,\n",
      "         3.4103963e-05, -4.4311350e-06,  2.9757857e-05],\n",
      "       [ 7.8832977e-07, -1.3924466e-07,  3.5538828e-06, ...,\n",
      "        -8.2675033e-05,  4.3862938e-06,  1.0573799e-06],\n",
      "       [ 8.1036410e-07,  9.9783698e-08,  5.9933345e-06, ...,\n",
      "         2.9554070e-04, -2.6292695e-05,  2.2041744e-05],\n",
      "       ...,\n",
      "       [-3.0384763e-07, -1.3819155e-06, -7.9395437e-07, ...,\n",
      "        -1.3064977e-05,  1.9726281e-06, -2.7538204e-05],\n",
      "       [-7.8948897e-06, -2.7677370e-06,  1.0054503e-05, ...,\n",
      "        -2.8208175e-04,  3.6709051e-05, -4.5907382e-05],\n",
      "       [ 1.5801619e-05,  1.1464830e-05, -4.6199502e-06, ...,\n",
      "        -4.8541199e-04,  3.1008323e-05,  8.1930732e-05]], dtype=float32)>, <tf.Tensor: id=908964, shape=(6144,), dtype=float32, numpy=\n",
      "array([-2.21453884e-05,  1.13231981e-05, -3.82285180e-06, ...,\n",
      "       -1.13087015e-04, -1.98593494e-04, -5.05752163e-04], dtype=float32)>, <tf.Tensor: id=908867, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.3603086e-07, -2.5191428e-05, -1.0576077e-06, ...,\n",
      "        -1.0947617e-06, -1.9361646e-06,  5.4939672e-05],\n",
      "       [-8.9988646e-08, -3.3527882e-05, -5.2836810e-07, ...,\n",
      "        -9.8245437e-08, -2.2026402e-06,  5.4110940e-05],\n",
      "       [ 2.5381112e-06, -8.2265637e-05,  2.1896308e-06, ...,\n",
      "         2.2272982e-06,  1.1778368e-06,  8.3289437e-05],\n",
      "       ...,\n",
      "       [ 7.5372413e-06, -8.1616425e-04,  6.5343256e-06, ...,\n",
      "         7.1856266e-06,  1.3155909e-05, -1.3471999e-04],\n",
      "       [-1.1885955e-06,  1.5960079e-04, -1.1048976e-06, ...,\n",
      "        -1.1746894e-06, -1.7679351e-06,  6.2492927e-06],\n",
      "       [ 2.4405645e-06,  1.9703290e-04,  2.6150501e-06, ...,\n",
      "         2.2818904e-06,  3.2096273e-06,  2.1903252e-05]], dtype=float32)>, <tf.Tensor: id=908859, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.98343296e-05, -8.49968288e-04,  1.77894472e-05,  2.06668465e-03,\n",
      "       -6.59655081e-03,  4.36148803e-05, -2.23980460e-04,  2.62620651e-05,\n",
      "        2.03857671e-05,  2.70610817e-05,  7.51791988e-04,  7.26203900e-04,\n",
      "        7.53164786e-05,  4.28650674e-05,  3.28064998e-05,  1.79920543e-03,\n",
      "       -2.26078293e-04, -4.26473608e-03, -1.54629306e-05,  6.46601431e-04,\n",
      "        3.47954396e-04,  7.12978886e-04, -1.92114705e-04,  2.61034264e-04,\n",
      "        3.94290109e-04, -1.32434128e-04,  1.50488806e-04,  1.16549090e-04,\n",
      "       -2.07022109e-04,  2.05969744e-04,  1.54421112e-04, -2.80094449e-04,\n",
      "        2.97456500e-05, -2.72786361e-04,  1.96309236e-04, -1.55923059e-04,\n",
      "        4.28770756e-04, -5.64275724e-05,  3.37289122e-04, -6.01895772e-05,\n",
      "       -2.33550032e-04,  6.22143853e-05,  4.64817742e-04, -6.30202121e-04,\n",
      "        3.03394132e-04, -3.18483624e-04,  2.35177286e-05,  9.71312984e-05,\n",
      "       -2.29423822e-04, -5.27218217e-04,  2.29357611e-04,  9.13309559e-05,\n",
      "       -1.15648974e-04, -3.29374801e-04, -9.50477988e-05,  1.52230976e-04,\n",
      "        1.88455451e-05,  3.65780579e-04,  7.43641649e-05,  3.11401876e-04,\n",
      "        5.39063149e-05, -2.61528126e-04, -3.06632370e-04,  4.80221534e-05,\n",
      "        3.42546009e-05,  4.01369855e-03, -1.50422694e-03, -2.87605659e-03,\n",
      "        6.05657510e-03, -9.46637895e-03, -2.47682387e-04,  2.28790595e-05,\n",
      "       -6.65295403e-03,  2.75369245e-03,  1.34701753e-04,  7.96302629e-05,\n",
      "        2.09975988e-05,  3.31681804e-06,  3.86130298e-04,  2.80140201e-03,\n",
      "        3.55125358e-03,  4.80885094e-04, -6.54501887e-03,  2.72510457e-03,\n",
      "        3.47162178e-03,  2.04033940e-03,  1.43395364e-03, -1.00193836e-03,\n",
      "        1.63986802e-03,  1.38296268e-03, -3.52276838e-04,  2.11193747e-05,\n",
      "        2.24427149e-05,  2.47714124e-05,  2.18473979e-05,  2.45093725e-05,\n",
      "        2.07219709e-05,  2.35026255e-05,  2.11757379e-05,  2.13073054e-05,\n",
      "        1.89922539e-05,  1.88542381e-05,  2.23424540e-05,  2.24465584e-05,\n",
      "        2.62248232e-05,  1.85347271e-05,  2.10696508e-05,  1.91597483e-05,\n",
      "        2.42087717e-05, -1.27628969e-04], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc400>, <tf.Tensor: id=909323, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 1.1394490e-06,  1.8162739e-06, -4.3821613e-09, ...,\n",
      "        -3.9539009e-05,  4.2799900e-05,  4.1117335e-05],\n",
      "       [ 3.9824013e-07,  9.4562438e-08,  8.5157843e-08, ...,\n",
      "        -4.0619350e-05,  2.6824393e-05,  2.9293922e-05],\n",
      "       [-2.0763645e-08,  5.6387826e-08,  3.0389273e-09, ...,\n",
      "        -1.4225601e-05,  2.5494519e-05, -1.1156742e-05],\n",
      "       ...,\n",
      "       [ 5.6305345e-07,  3.5572339e-07, -5.5357413e-07, ...,\n",
      "        -3.4940733e-06,  7.0500715e-07,  2.1416472e-06],\n",
      "       [-6.4425102e-08, -4.8728541e-08,  3.5539841e-08, ...,\n",
      "        -1.2011051e-05, -3.1611173e-05, -4.1678159e-06],\n",
      "       [-7.3234884e-07,  7.0907049e-07,  3.1329159e-07, ...,\n",
      "         8.2106351e-05, -1.0406573e-05, -2.6730118e-05]], dtype=float32)>, <tf.Tensor: id=909324, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-1.03975651e-06, -5.25707947e-06,  3.58147986e-07, ...,\n",
      "        -3.09434545e-04,  1.99599381e-05, -1.33395864e-04],\n",
      "       [ 3.49480302e-08, -6.43765020e-07,  1.06809875e-06, ...,\n",
      "        -4.86672434e-05, -4.10500434e-06,  1.21961630e-05],\n",
      "       [ 9.23008884e-07,  2.36468168e-06,  1.60542115e-07, ...,\n",
      "        -2.11362887e-04,  3.81617356e-05, -9.23366606e-05],\n",
      "       ...,\n",
      "       [ 1.82426696e-09,  4.68722590e-08,  6.32478248e-07, ...,\n",
      "        -3.66664608e-05,  4.24534301e-06, -1.06832140e-05],\n",
      "       [ 1.48109175e-05,  1.62644392e-05,  2.22076105e-05, ...,\n",
      "        -2.26803182e-04,  1.06731331e-05, -5.18652814e-05],\n",
      "       [-4.31178887e-06,  6.38638539e-06,  6.09415474e-06, ...,\n",
      "         5.31804224e-04, -8.09207268e-05,  2.57641339e-04]], dtype=float32)>, <tf.Tensor: id=909325, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 2.1210653e-05,  7.1721917e-07, -5.0494668e-06, ...,\n",
      "       -3.7035566e-05, -2.3395030e-04,  5.3870532e-04], dtype=float32)>, <tf.Tensor: id=909228, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.3620986e-07, -2.9714618e-05, -1.0675452e-06, ...,\n",
      "        -9.3154392e-07, -1.8080067e-06, -2.8557963e-06],\n",
      "       [-1.5452633e-06, -1.3664408e-04, -1.8450048e-06, ...,\n",
      "        -1.4278014e-06, -3.2557880e-06, -4.9728669e-06],\n",
      "       [-5.1342784e-07,  1.2944332e-04, -6.4292641e-07, ...,\n",
      "        -6.2316099e-07, -1.3787726e-06, -1.9365202e-06],\n",
      "       ...,\n",
      "       [ 6.3992284e-06,  2.8837653e-04,  6.3014304e-06, ...,\n",
      "         6.0720618e-06,  1.1374463e-05,  1.6180260e-05],\n",
      "       [-7.2487865e-07,  1.3292738e-06, -7.6186404e-07, ...,\n",
      "        -7.2122390e-07, -1.2487351e-06, -1.9983797e-06],\n",
      "       [ 1.0918657e-06,  1.4310647e-04,  1.3116394e-06, ...,\n",
      "         1.0184319e-06,  2.0062744e-06,  3.1882155e-06]], dtype=float32)>, <tf.Tensor: id=909220, shape=(110,), dtype=float32, numpy=\n",
      "array([ 6.5777740e-06,  2.8009966e-04,  6.4465476e-06,  6.0157891e-04,\n",
      "        3.3387914e-03,  3.4544857e-05, -1.2783001e-03,  7.3779183e-06,\n",
      "        7.9222991e-06,  1.1144224e-05,  2.9029578e-04,  6.1772595e-04,\n",
      "        6.7151099e-04,  2.8323451e-05,  1.6454880e-05,  1.3337668e-03,\n",
      "        6.9230871e-04, -1.9734527e-04,  1.2342597e-04,  1.7278746e-04,\n",
      "        2.4301127e-04,  3.2650406e-04, -8.9003443e-06, -1.0601812e-03,\n",
      "        2.8185773e-04, -2.2898916e-04, -6.0332066e-05,  2.8513177e-04,\n",
      "        3.0875730e-04,  3.4888761e-04, -1.1968781e-03,  1.6905080e-05,\n",
      "        1.3075089e-05,  2.3750863e-05,  3.3969650e-04, -4.9740000e-04,\n",
      "       -4.3233635e-04,  1.3053830e-04,  2.2638495e-04,  2.9662880e-05,\n",
      "       -1.6584944e-04, -3.0121460e-04,  1.3136513e-04, -1.2238155e-03,\n",
      "        6.4108586e-05, -2.9847713e-05, -1.9181012e-04,  2.9504014e-04,\n",
      "       -4.3570698e-04,  1.1562988e-04,  4.0380100e-06,  6.8637441e-05,\n",
      "       -6.7068433e-04, -6.4721331e-04,  2.9069887e-04,  1.1634714e-04,\n",
      "        1.2956996e-04,  6.1730208e-04,  4.8059930e-05,  2.3929689e-04,\n",
      "       -2.7928627e-04, -1.5135622e-04, -1.2756433e-05,  3.3938577e-05,\n",
      "        2.1707823e-05, -3.3252800e-03,  1.6390870e-03, -3.1283945e-03,\n",
      "        2.7595479e-03, -6.5861256e-03,  1.7277247e-04, -1.7072659e-03,\n",
      "       -7.8644175e-03,  2.9472448e-03,  3.6500231e-04,  3.1754730e-04,\n",
      "        3.5683690e-03, -2.6321234e-04, -6.0653011e-03,  7.1857711e-03,\n",
      "        2.4028923e-03,  3.6044361e-04, -4.4513466e-03,  1.5029253e-04,\n",
      "        6.8651107e-03, -9.3485322e-04, -3.6935922e-04,  3.3832213e-05,\n",
      "       -2.4725395e-04,  2.8088673e-03, -7.1865291e-04,  9.9326962e-06,\n",
      "        7.0789188e-06,  1.1196153e-05,  8.0373184e-06,  6.7754982e-06,\n",
      "        8.3187224e-06,  6.9251878e-06,  6.7213691e-06,  7.5529483e-06,\n",
      "        8.1547450e-06,  7.3270389e-06,  8.5367965e-06,  6.8307609e-06,\n",
      "        1.1415302e-05,  6.6223756e-06,  8.0645805e-06,  6.2327763e-06,\n",
      "        1.1643864e-05,  1.6532813e-05], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafcb38>, <tf.Tensor: id=909684, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 9.26233497e-07, -1.98208204e-06, -1.55500027e-06, ...,\n",
      "        -2.53019552e-05, -1.05352165e-05,  3.44114937e-06],\n",
      "       [ 1.11027339e-06, -1.93202618e-06,  1.33368599e-06, ...,\n",
      "         6.21855179e-06, -3.08464332e-05,  2.27574014e-06],\n",
      "       [ 5.11381359e-07, -6.38497056e-07,  8.55665974e-08, ...,\n",
      "        -1.73427179e-05,  1.90598735e-06, -3.95957359e-05],\n",
      "       ...,\n",
      "       [-1.13610429e-06,  6.59348643e-07,  8.87277338e-07, ...,\n",
      "        -7.90900231e-05, -2.73894075e-05, -5.96601312e-05],\n",
      "       [-7.37461187e-06, -1.78577045e-06,  1.31509552e-07, ...,\n",
      "         2.13323583e-05, -3.66612585e-05,  1.68492807e-05],\n",
      "       [-2.51878078e-06, -4.16169144e-07, -1.39418205e-07, ...,\n",
      "         4.00841382e-05, -9.30340029e-05, -6.14581677e-06]], dtype=float32)>, <tf.Tensor: id=909685, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 1.5869175e-06, -6.7241167e-06,  3.8405233e-07, ...,\n",
      "        -1.1566015e-05,  4.4477216e-05,  2.6627524e-05],\n",
      "       [ 7.0610467e-06,  6.6933285e-06,  1.5976391e-06, ...,\n",
      "         2.2253020e-04, -2.2070838e-06,  2.3960014e-04],\n",
      "       [ 1.5956311e-06,  4.7066962e-08,  3.7251900e-06, ...,\n",
      "        -1.3225625e-04,  3.0413201e-05,  3.2986864e-05],\n",
      "       ...,\n",
      "       [ 1.3846881e-05, -5.9797731e-07, -8.3870464e-06, ...,\n",
      "        -2.7108603e-04, -9.3893750e-06, -1.9212920e-04],\n",
      "       [-4.5121232e-07, -7.7693157e-06, -2.2150753e-05, ...,\n",
      "         3.7354903e-04,  2.6773971e-06,  2.1900784e-04],\n",
      "       [-8.5906804e-06, -2.5660203e-05,  1.8680976e-05, ...,\n",
      "         1.9016526e-04, -8.4013584e-05,  1.6324186e-04]], dtype=float32)>, <tf.Tensor: id=909686, shape=(6144,), dtype=float32, numpy=\n",
      "array([-7.1145987e-06, -2.4687521e-05, -1.6650290e-05, ...,\n",
      "       -1.4695278e-03,  2.0046148e-03,  2.0055921e-05], dtype=float32)>, <tf.Tensor: id=909589, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[ 4.3878399e-06,  2.2108645e-04,  3.1582881e-06, ...,\n",
      "         3.6157385e-06,  2.4158608e-06, -4.7389483e-07],\n",
      "       [ 1.1678465e-05,  1.9124813e-05,  9.0869426e-06, ...,\n",
      "         1.0492135e-05,  7.5523922e-06,  2.4688493e-06],\n",
      "       [ 2.6664082e-05,  1.4098117e-04,  2.1601081e-05, ...,\n",
      "         2.3167686e-05,  2.0270534e-05,  1.3477596e-05],\n",
      "       ...,\n",
      "       [-1.6501923e-05, -1.8571698e-03, -1.3580361e-05, ...,\n",
      "        -1.4869773e-05, -7.2515149e-06,  6.2938529e-06],\n",
      "       [ 3.6236747e-06,  2.6290980e-04,  3.3061447e-06, ...,\n",
      "         4.0805571e-06,  2.9635016e-06,  1.2258977e-07],\n",
      "       [ 9.7946513e-06,  9.3678129e-05,  8.3756549e-06, ...,\n",
      "         8.2224033e-06,  8.8243351e-06,  8.4373669e-06]], dtype=float32)>, <tf.Tensor: id=909581, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.22788944e-04, -2.26205331e-03,  1.00284182e-04,  2.20186921e-05,\n",
      "        6.82072481e-03,  7.47153899e-05,  1.08320965e-03, -1.22051670e-05,\n",
      "        1.10379719e-04,  1.49945423e-04, -6.13916782e-04,  4.63831064e-04,\n",
      "        4.35268681e-04,  1.17254705e-04,  1.39889424e-04,  3.88625544e-04,\n",
      "        6.09467446e-04, -3.43337213e-03,  1.59823263e-04,  5.02495095e-04,\n",
      "        1.03073253e-03,  2.80129083e-04,  9.98972391e-05,  3.66069697e-04,\n",
      "        4.91273007e-04,  4.27261257e-04,  4.66102210e-04,  5.11425198e-04,\n",
      "        8.82776803e-04, -2.54570274e-04, -1.40209551e-04,  1.24748854e-04,\n",
      "        1.27446328e-04,  1.05363419e-04,  3.56146687e-04, -1.07730017e-03,\n",
      "       -5.57522755e-04, -6.17963145e-04, -4.19841264e-04, -1.72755797e-03,\n",
      "       -3.06529400e-04, -3.50353628e-04,  5.75985250e-05, -1.03900419e-03,\n",
      "        3.55849297e-05,  2.42195980e-04, -8.48514173e-05, -3.91107926e-04,\n",
      "       -1.69625843e-03, -7.28200597e-04,  2.91829259e-04,  1.94187269e-05,\n",
      "       -1.39571348e-04, -7.42229982e-04, -1.20428938e-03, -3.82888567e-04,\n",
      "       -1.77579510e-04,  2.46512122e-04,  1.97157660e-05,  2.62038491e-04,\n",
      "        1.51597778e-04,  1.23730468e-04,  2.90795888e-05,  1.05371997e-04,\n",
      "        1.10389519e-04, -5.24482969e-03, -1.48970739e-03, -1.90244219e-03,\n",
      "       -5.77010855e-04,  7.91675225e-03,  1.18161098e-03,  2.65534851e-03,\n",
      "       -2.69548572e-03,  2.29198765e-03,  8.60479311e-04, -3.41330597e-04,\n",
      "        6.97512878e-04, -1.83065725e-03, -5.33187762e-03,  4.18153417e-04,\n",
      "        4.13020665e-04,  1.25224018e-04,  2.92996032e-04,  2.58170441e-03,\n",
      "       -1.72220194e-03, -8.99388688e-04,  6.23881351e-04, -5.96140628e-04,\n",
      "        2.08227313e-04, -4.58950642e-04,  6.65925618e-04,  1.00748453e-04,\n",
      "        1.36172093e-04,  1.18193166e-04,  1.16905612e-04,  1.57648872e-04,\n",
      "        1.05527382e-04,  1.31652356e-04,  1.43188794e-04,  1.32791683e-04,\n",
      "        9.74363866e-05,  1.00519035e-04,  1.27364416e-04,  1.27776802e-04,\n",
      "        1.27064952e-04,  1.10920781e-04,  1.13106340e-04,  1.08833061e-04,\n",
      "        1.05695523e-04,  8.98739308e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdae6748>, <tf.Tensor: id=910045, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-6.9183916e-06, -3.3198680e-06,  1.5783752e-06, ...,\n",
      "        -5.6159119e-05, -2.6555348e-05, -1.1664557e-05],\n",
      "       [-1.5052881e-07,  3.9236261e-07, -9.4667847e-08, ...,\n",
      "         4.3816919e-05,  1.8708691e-06,  1.0909039e-05],\n",
      "       [-5.1221338e-07,  3.1384556e-07, -4.5031359e-07, ...,\n",
      "         4.2381664e-05,  1.3836330e-06,  1.0837990e-05],\n",
      "       ...,\n",
      "       [ 1.3504135e-06,  1.9781378e-07, -4.3706731e-07, ...,\n",
      "        -1.4165181e-07, -8.1783173e-06, -1.6167726e-06],\n",
      "       [-6.4234379e-10,  1.9470423e-07,  2.6652277e-08, ...,\n",
      "        -7.2725920e-06,  3.6199253e-05, -1.8171928e-05],\n",
      "       [ 1.1048915e-06,  6.4176237e-07, -2.9873482e-07, ...,\n",
      "         1.4061510e-05, -1.6789903e-05,  3.2570410e-05]], dtype=float32)>, <tf.Tensor: id=910046, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-1.62858796e-05, -1.40404545e-05, -1.77500101e-06, ...,\n",
      "        -7.88844074e-04,  1.17830270e-04, -1.76086949e-04],\n",
      "       [ 7.39470977e-07, -1.10505835e-06,  3.02213084e-06, ...,\n",
      "         2.63560098e-04, -5.68806754e-05,  2.06798461e-04],\n",
      "       [ 8.48879608e-07,  1.16941840e-06, -4.56380064e-07, ...,\n",
      "        -1.54976791e-04,  1.02572794e-05,  2.02320575e-06],\n",
      "       ...,\n",
      "       [-4.06358026e-07, -1.39079521e-06,  1.86929935e-06, ...,\n",
      "         6.33396339e-05, -4.78233687e-06,  1.30272001e-05],\n",
      "       [ 3.52632424e-06,  1.25856404e-05,  2.28395820e-05, ...,\n",
      "        -4.32991015e-04,  5.35725230e-05, -1.09342975e-04],\n",
      "       [ 9.61571550e-06,  3.01089094e-05, -1.29929904e-05, ...,\n",
      "        -6.10096671e-04,  2.35372172e-05,  1.00156808e-06]], dtype=float32)>, <tf.Tensor: id=910047, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 3.0819541e-05, -5.1400530e-06,  1.1677504e-07, ...,\n",
      "        4.7855196e-05, -4.3521865e-04, -6.1512057e-04], dtype=float32)>, <tf.Tensor: id=909950, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.0556672e-06,  1.5424250e-04, -1.1641287e-06, ...,\n",
      "        -1.0393829e-06, -1.9150170e-06, -3.1493651e-06],\n",
      "       [-1.4553706e-06, -4.9054790e-05, -1.7530820e-06, ...,\n",
      "        -1.3248936e-06, -2.9312580e-06, -4.8515026e-06],\n",
      "       [-3.1334503e-07,  1.7416332e-04, -4.8366826e-07, ...,\n",
      "        -4.6141292e-07, -1.2857296e-06, -2.1014744e-06],\n",
      "       ...,\n",
      "       [ 7.5642470e-06, -1.7451851e-03,  6.9725452e-06, ...,\n",
      "         7.1306931e-06,  1.2217282e-05,  1.8234807e-05],\n",
      "       [-1.0288629e-06,  2.7757767e-04, -9.7422094e-07, ...,\n",
      "        -1.0034098e-06, -1.5511484e-06, -2.4423948e-06],\n",
      "       [ 9.6182998e-07,  7.0187845e-04,  1.2953606e-06, ...,\n",
      "         9.4962871e-07,  1.7161337e-06,  2.8374502e-06]], dtype=float32)>, <tf.Tensor: id=909942, shape=(110,), dtype=float32, numpy=\n",
      "array([ 8.77449202e-06, -1.79066672e-03,  8.04155752e-06,  7.80384289e-04,\n",
      "        5.26096672e-03,  3.56924866e-05, -1.00697158e-03,  1.04828960e-05,\n",
      "        1.02862368e-05,  1.26580344e-05,  9.67613887e-05,  3.88471992e-04,\n",
      "        2.34224863e-04,  3.15100006e-05,  1.92398111e-05,  1.18675805e-03,\n",
      "       -1.62782118e-04,  1.23812686e-04,  1.27740728e-04, -3.57565295e-04,\n",
      "       -2.93538469e-04, -3.55887641e-05, -7.96839071e-04,  1.13545022e-04,\n",
      "        2.92779587e-04,  2.87481613e-04,  4.60766023e-05,  2.66859861e-04,\n",
      "        3.26809095e-04, -1.18676573e-04,  5.67671552e-04,  2.22281051e-05,\n",
      "        1.61243679e-05,  2.83398986e-05,  3.75964009e-06, -6.89000648e-04,\n",
      "       -5.53462480e-04, -1.53697518e-04, -2.50339101e-04,  5.39292058e-04,\n",
      "        1.92623687e-04,  1.65955804e-04,  1.97776142e-04,  3.06799018e-04,\n",
      "       -1.14806273e-04, -3.14807781e-04, -2.79299093e-06,  2.96194077e-04,\n",
      "       -6.89582317e-04,  7.45700236e-05, -1.32832880e-04, -8.72283854e-05,\n",
      "       -8.68418720e-05, -8.46435796e-05,  8.53245438e-04, -4.12023983e-05,\n",
      "       -2.66069692e-05, -3.94123665e-04,  5.47962118e-05,  1.22228244e-04,\n",
      "        3.77938813e-05,  1.13526785e-05,  5.18583875e-06,  4.42164928e-05,\n",
      "        2.41692360e-05, -1.82946469e-03, -1.68494633e-04, -2.15462781e-03,\n",
      "        1.62913883e-03,  2.19569951e-02,  2.65095010e-03,  5.03340561e-04,\n",
      "       -2.55809329e-03, -3.65937036e-03, -3.34991128e-05, -3.15024750e-04,\n",
      "        4.86022420e-03, -1.91401865e-04, -3.19235213e-03, -7.44080229e-04,\n",
      "        1.10071676e-04, -3.04837537e-04, -8.64187256e-03,  2.36621883e-04,\n",
      "       -8.17183405e-03,  8.80539650e-04, -2.23632692e-03, -2.88612605e-03,\n",
      "       -1.41395931e-03,  5.32542705e-04, -1.15350296e-04,  1.13134211e-05,\n",
      "        9.28040936e-06,  1.41904447e-05,  1.08663307e-05,  9.33239153e-06,\n",
      "        1.08240902e-05,  9.72636735e-06,  8.77155071e-06,  9.72115322e-06,\n",
      "        9.96597828e-06,  9.38645371e-06,  1.00376074e-05,  9.49486730e-06,\n",
      "        1.42171320e-05,  8.87605256e-06,  9.98536325e-06,  8.29209512e-06,\n",
      "        1.33678632e-05,  1.95555895e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdae7908>, <tf.Tensor: id=910406, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-2.60219622e-06, -1.53303540e-06,  1.08236998e-06, ...,\n",
      "         1.07043625e-05, -8.36796789e-06,  2.30378628e-05],\n",
      "       [ 1.14853776e-07,  4.65291350e-07,  5.50744605e-09, ...,\n",
      "         2.49501154e-05, -8.85918962e-06,  3.36141711e-05],\n",
      "       [ 4.59055684e-07,  1.60140686e-07, -8.76111770e-08, ...,\n",
      "        -3.07691989e-05,  1.74058550e-05,  4.92274012e-06],\n",
      "       ...,\n",
      "       [ 1.74862589e-06,  1.08136862e-06, -4.66682025e-07, ...,\n",
      "        -1.39648873e-05,  3.46433421e-06, -6.02020464e-07],\n",
      "       [-2.69823062e-08,  6.02491355e-08,  4.96065198e-08, ...,\n",
      "         1.11566624e-04, -1.25031002e-05,  8.90376577e-06],\n",
      "       [-6.36415677e-07,  1.11028373e-06,  7.42734414e-07, ...,\n",
      "        -1.78077007e-05, -1.89495095e-05, -1.80192255e-05]], dtype=float32)>, <tf.Tensor: id=910407, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-1.1832939e-05, -1.2629031e-05, -1.2286091e-07, ...,\n",
      "        -2.4562000e-04,  2.6650136e-05, -1.1305498e-04],\n",
      "       [-1.4246484e-06, -3.1809827e-06, -1.5188194e-06, ...,\n",
      "         8.2037783e-05, -9.6574367e-06,  3.7042544e-05],\n",
      "       [ 6.0987043e-07,  6.3978263e-07, -7.0067540e-07, ...,\n",
      "         1.7918147e-04, -2.6052139e-05,  1.0431435e-05],\n",
      "       ...,\n",
      "       [ 6.1653810e-07,  9.4141302e-07, -9.5123966e-07, ...,\n",
      "        -4.1692747e-06,  2.6634352e-06, -7.4945649e-07],\n",
      "       [ 2.2176123e-06,  4.3057080e-06,  1.4770445e-05, ...,\n",
      "        -3.0207057e-06, -1.4517271e-05,  1.1376685e-05],\n",
      "       [ 7.5674125e-06, -4.4685412e-06,  1.2157881e-07, ...,\n",
      "         8.7610482e-05, -6.9942784e-05,  1.9334783e-04]], dtype=float32)>, <tf.Tensor: id=910408, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 3.4694065e-05,  3.4007885e-06, -3.2675302e-07, ...,\n",
      "       -4.2115248e-06, -3.4449622e-06,  8.6898996e-05], dtype=float32)>, <tf.Tensor: id=910311, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.7686632e-07, -7.6690660e-05, -1.1165993e-06, ...,\n",
      "        -9.6804899e-07, -1.8792911e-06,  4.2277850e-05],\n",
      "       [-1.2945547e-06, -4.0460589e-05, -1.6216343e-06, ...,\n",
      "        -1.1847003e-06, -2.8047054e-06,  5.7237641e-05],\n",
      "       [-5.0577933e-07, -5.5401524e-05, -6.5765408e-07, ...,\n",
      "        -5.9958313e-07, -1.3250447e-06,  8.3338469e-05],\n",
      "       ...,\n",
      "       [ 6.1870269e-06,  1.2104106e-03,  6.1754613e-06, ...,\n",
      "         5.7797006e-06,  1.0823087e-05, -1.3918591e-04],\n",
      "       [-7.7710621e-07, -1.3424923e-04, -7.9179995e-07, ...,\n",
      "        -7.5838057e-07, -1.2997484e-06,  2.0207504e-05],\n",
      "       [ 8.0858814e-07,  5.2890136e-05,  1.1302099e-06, ...,\n",
      "         8.1346224e-07,  1.6371598e-06, -1.7060222e-05]], dtype=float32)>, <tf.Tensor: id=910303, shape=(110,), dtype=float32, numpy=\n",
      "array([ 6.33074069e-06,  1.24878855e-03,  6.29742817e-06, -1.34312862e-03,\n",
      "       -4.84652305e-03,  3.21116822e-05, -6.12485572e-04,  7.58247006e-06,\n",
      "        7.68310838e-06,  1.03745715e-05, -6.09891082e-04,  3.08051734e-04,\n",
      "        4.45414014e-04, -2.85308168e-04,  1.54488025e-05,  9.23717860e-04,\n",
      "       -1.03739696e-03,  3.31173214e-04,  1.17050920e-04, -3.80968231e-07,\n",
      "       -1.44430262e-04, -6.12682925e-05,  3.66753142e-04, -2.58614280e-04,\n",
      "       -5.78180305e-04,  6.64625841e-05, -1.81932017e-04, -4.21613455e-04,\n",
      "        1.79994284e-04,  3.15260782e-04,  4.51445114e-04,  1.81693467e-05,\n",
      "        1.19942069e-05,  2.34416257e-05, -2.30543628e-05,  2.01183342e-04,\n",
      "        3.42155603e-04,  2.79948086e-04, -1.32106885e-04,  4.84708988e-04,\n",
      "        3.14497098e-04,  1.65236663e-04, -6.91730063e-04,  2.08714424e-04,\n",
      "        1.66422004e-04, -1.65214748e-04,  2.91369477e-04, -2.42227950e-04,\n",
      "        2.60088709e-04,  1.28343701e-04,  3.20866035e-04,  6.16718171e-05,\n",
      "        3.55339434e-04,  4.76602640e-04,  8.07020988e-05,  9.97585812e-05,\n",
      "       -1.99779926e-04,  2.41369475e-04,  4.69185870e-05, -8.19663765e-05,\n",
      "        3.41909472e-05,  1.55164467e-04,  1.55662841e-04,  3.66419845e-05,\n",
      "        2.12085397e-05, -5.34126768e-03,  1.37381582e-03, -1.70668558e-04,\n",
      "        1.46503840e-03,  4.31422936e-03,  2.19302438e-03, -1.18576258e-03,\n",
      "        3.34795192e-03,  1.36119896e-03,  2.06071418e-04,  5.82318986e-04,\n",
      "        1.17838220e-03, -1.79720507e-03,  5.66208595e-03, -2.39061285e-03,\n",
      "       -2.64110253e-03, -3.06801230e-05,  3.73791903e-03, -4.31774557e-03,\n",
      "       -2.46298593e-03, -5.82475914e-05, -1.11296494e-03,  2.90621014e-04,\n",
      "       -1.23708730e-03, -1.06662977e-03,  3.50429269e-04,  9.14534849e-06,\n",
      "        6.73431805e-06,  1.08001241e-05,  7.91692491e-06,  6.57113378e-06,\n",
      "        8.27220993e-06,  6.74129569e-06,  6.30242812e-06,  7.01350791e-06,\n",
      "        7.76733941e-06,  7.06794435e-06,  7.67347228e-06,  6.74012063e-06,\n",
      "       -1.44885227e-04,  6.73510658e-06,  7.31889622e-06,  5.90754826e-06,\n",
      "        1.10388155e-05, -1.40066084e-04], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc320>, <tf.Tensor: id=910767, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-3.8572125e-06, -1.4436366e-06,  1.8343950e-06, ...,\n",
      "        -4.3649266e-06, -3.9772731e-05, -1.0687370e-05],\n",
      "       [ 1.5008398e-07,  2.1189321e-07,  1.6479547e-08, ...,\n",
      "         7.9131551e-06,  1.3312561e-05,  9.3704693e-06],\n",
      "       [ 4.1548088e-07,  2.6009366e-07, -1.4288845e-07, ...,\n",
      "         5.2986036e-07, -6.3018761e-06, -2.6464027e-06],\n",
      "       ...,\n",
      "       [-1.5386405e-07,  1.9565686e-08, -2.3277795e-07, ...,\n",
      "        -5.5366618e-06,  5.7949514e-06,  9.9200156e-07],\n",
      "       [-7.2189465e-08, -2.9539460e-09,  4.9062809e-08, ...,\n",
      "        -3.7371239e-05, -3.8093299e-06,  6.0542170e-05],\n",
      "       [-5.3858986e-07,  2.6254637e-07,  7.8496521e-07, ...,\n",
      "         8.9441419e-06,  8.0262043e-07,  5.4681645e-06]], dtype=float32)>, <tf.Tensor: id=910768, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-7.45977559e-06, -8.00594444e-06,  9.83805057e-06, ...,\n",
      "        -6.28549024e-04,  8.18893604e-05, -1.08186345e-04],\n",
      "       [-7.01336489e-07, -1.65129552e-06, -2.92127879e-06, ...,\n",
      "         2.03488496e-04, -3.12824195e-05, -1.40793163e-05],\n",
      "       [-1.37784821e-06, -8.80655534e-07, -1.84547594e-06, ...,\n",
      "         2.41950274e-05, -5.18332490e-06,  6.46762637e-05],\n",
      "       ...,\n",
      "       [-1.14280837e-07, -1.66219394e-07,  7.23602398e-07, ...,\n",
      "         3.96741634e-05, -3.63386425e-06,  1.02415534e-05],\n",
      "       [-3.67238727e-07, -6.69440124e-06, -9.24084179e-06, ...,\n",
      "        -2.64798255e-05,  1.42129211e-05, -2.31832855e-05],\n",
      "       [-6.30680370e-06,  1.28836234e-06, -1.20432951e-05, ...,\n",
      "         2.10834027e-04, -2.92675722e-05,  1.51439017e-04]], dtype=float32)>, <tf.Tensor: id=910769, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 1.1295940e-05,  2.1340588e-06,  2.4023993e-06, ...,\n",
      "        4.0216408e-05, -2.6335483e-05,  2.1403839e-04], dtype=float32)>, <tf.Tensor: id=910672, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.9291296e-07, -6.9578869e-05, -1.0175154e-06, ...,\n",
      "        -8.9164291e-07, -1.7369680e-06, -2.9120981e-06],\n",
      "       [-1.3158824e-06, -8.6537211e-06, -1.6224120e-06, ...,\n",
      "        -1.2187675e-06, -2.8599775e-06, -4.7790204e-06],\n",
      "       [-5.6606086e-07, -2.3428432e-04, -7.0311518e-07, ...,\n",
      "        -6.5299201e-07, -1.3934105e-06, -2.3634454e-06],\n",
      "       ...,\n",
      "       [ 6.2914833e-06,  6.3159259e-04,  6.0772891e-06, ...,\n",
      "         5.9615713e-06,  1.0896471e-05,  1.6666112e-05],\n",
      "       [-7.3569845e-07, -1.0662951e-04, -7.6103686e-07, ...,\n",
      "        -7.3634493e-07, -1.2527373e-06, -2.1383707e-06],\n",
      "       [ 8.7292790e-07, -1.4656861e-04,  1.1440147e-06, ...,\n",
      "         8.8104542e-07,  1.7356340e-06,  2.9306775e-06]], dtype=float32)>, <tf.Tensor: id=910664, shape=(110,), dtype=float32, numpy=\n",
      "array([ 6.4057817e-06,  6.4381224e-04,  6.1737856e-06, -1.0625465e-03,\n",
      "        3.5502572e-04, -1.2395340e-04, -2.4910949e-04,  7.6073802e-06,\n",
      "        7.8322955e-06,  1.0069789e-05,  1.2187811e-03,  3.1948104e-05,\n",
      "        5.6347274e-04,  2.8047605e-05,  1.5431844e-05, -1.3413017e-03,\n",
      "        3.0905486e-04,  3.5618231e-04,  1.1631059e-04, -8.1597129e-04,\n",
      "       -7.2154944e-05, -1.6691776e-03,  4.0245987e-04, -2.3416399e-04,\n",
      "        2.8710758e-05,  1.9085013e-04,  1.4559846e-04,  5.3832424e-05,\n",
      "        5.3694041e-04,  4.7650939e-04,  6.9922593e-04,  1.9539246e-05,\n",
      "        1.2120223e-05,  2.3428700e-05,  2.8560002e-04, -1.5159155e-04,\n",
      "       -2.0003495e-04, -6.4561312e-04, -2.7115908e-04, -7.1599250e-05,\n",
      "       -4.4398580e-04,  7.2094290e-06,  1.6923684e-04, -4.5115731e-04,\n",
      "       -4.6574106e-04,  3.3950741e-06,  2.9735634e-04, -1.8436296e-04,\n",
      "        8.6396485e-06,  3.4439220e-04, -1.4821452e-04,  6.0155089e-05,\n",
      "        6.7181245e-05, -2.5666651e-04,  8.6834814e-05,  9.8302473e-05,\n",
      "       -2.0291650e-04, -3.3763368e-04, -1.0995818e-04, -3.9350492e-05,\n",
      "        3.4312838e-05,  5.3904432e-06, -1.5214774e-04,  3.9300437e-05,\n",
      "        2.1338084e-05, -6.3494765e-03,  1.2205509e-03,  7.9798512e-05,\n",
      "       -4.7138785e-03,  1.1862451e-02,  3.6320069e-03, -2.0753390e-03,\n",
      "        4.7554373e-04, -5.6358078e-03, -1.6369979e-04, -7.3563971e-04,\n",
      "        1.9050259e-03, -1.2842477e-03,  2.1006190e-03, -6.6342403e-04,\n",
      "       -1.7545988e-03,  3.8589461e-04,  3.4097291e-03,  2.3553157e-03,\n",
      "       -1.4612261e-03, -1.0703718e-03, -2.5574579e-03,  1.1420663e-03,\n",
      "       -6.5786287e-04,  1.9615169e-03,  3.3817458e-04,  8.8924025e-06,\n",
      "        6.7790879e-06,  1.1214954e-05,  8.1279368e-06,  6.5548847e-06,\n",
      "        8.3237537e-06,  7.0771734e-06,  6.4226597e-06,  7.2315170e-06,\n",
      "        8.0695854e-06,  7.2633293e-06,  7.7571194e-06,  6.8908630e-06,\n",
      "        1.1812859e-05,  6.8465160e-06,  7.5052171e-06,  6.0666753e-06,\n",
      "        1.1073583e-05,  1.6921042e-05], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc7f0>, <tf.Tensor: id=911128, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 9.77839022e-07, -2.16720650e-06, -7.21907497e-07, ...,\n",
      "        -1.34424345e-05, -1.39435297e-05,  2.13335170e-05],\n",
      "       [ 5.67183520e-07, -3.15812443e-08, -9.82861579e-08, ...,\n",
      "         5.41195368e-06, -1.58126950e-05, -1.05768981e-06],\n",
      "       [-3.19930848e-07,  7.07399010e-08,  7.24937834e-08, ...,\n",
      "         2.35790540e-05, -2.09142945e-05, -2.21106238e-05],\n",
      "       ...,\n",
      "       [ 8.72750888e-07,  6.43337216e-07, -4.56401409e-07, ...,\n",
      "         3.52860434e-06, -1.54388490e-05, -1.38730002e-05],\n",
      "       [ 8.43464889e-08,  3.02983523e-08, -8.73113082e-10, ...,\n",
      "         7.58020406e-06,  7.46780934e-05,  3.57165686e-06],\n",
      "       [-1.43735761e-07,  4.22794045e-07,  1.55363750e-07, ...,\n",
      "         8.58813200e-06, -4.23083657e-06,  2.16841158e-06]], dtype=float32)>, <tf.Tensor: id=911129, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 7.7784125e-06, -2.7634496e-06,  5.5535979e-06, ...,\n",
      "         6.7254342e-04, -1.1491311e-04,  7.8418700e-05],\n",
      "       [ 5.6048088e-07,  1.3905603e-06,  7.6827678e-07, ...,\n",
      "        -9.1197493e-05,  6.4431142e-06,  9.1465863e-06],\n",
      "       [-7.0407305e-07, -7.9008493e-07, -5.0329345e-07, ...,\n",
      "         4.9947016e-04, -6.3027132e-05,  1.4699105e-04],\n",
      "       ...,\n",
      "       [ 3.1777194e-08,  1.3886687e-07, -7.7110968e-07, ...,\n",
      "         3.3817483e-05, -2.9980345e-06,  1.1880978e-05],\n",
      "       [-2.4260164e-06, -1.8099467e-05, -2.4962312e-05, ...,\n",
      "         5.2998390e-04, -7.1960734e-05,  1.3989894e-04],\n",
      "       [-2.6527901e-05, -4.2749205e-05,  2.7703303e-05, ...,\n",
      "         4.7371759e-06, -4.0431580e-05, -8.3801278e-05]], dtype=float32)>, <tf.Tensor: id=911130, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 7.4127752e-06, -4.1759649e-06,  7.3315641e-06, ...,\n",
      "        3.4217614e-05,  5.3590717e-04,  6.2100553e-06], dtype=float32)>, <tf.Tensor: id=911033, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.5938666e-07,  5.5514527e-05, -1.0036703e-06, ...,\n",
      "        -8.6262071e-07, -1.7403200e-06, -2.9295090e-06],\n",
      "       [-1.4401899e-06,  8.0460770e-05, -1.7345296e-06, ...,\n",
      "        -1.3191027e-06, -3.1453364e-06, -5.1003226e-06],\n",
      "       [-4.8247733e-07,  5.3529206e-05, -6.4381277e-07, ...,\n",
      "        -6.0038258e-07, -1.3743972e-06, -2.3158434e-06],\n",
      "       ...,\n",
      "       [ 6.0679458e-06,  6.5078528e-04,  6.0198909e-06, ...,\n",
      "         5.6265408e-06,  1.1058139e-05,  1.6624235e-05],\n",
      "       [-6.5484465e-07, -7.5336684e-05, -6.9891081e-07, ...,\n",
      "        -6.5018133e-07, -1.1631271e-06, -2.0041366e-06],\n",
      "       [ 8.6819784e-07, -2.4052142e-04,  1.1425089e-06, ...,\n",
      "         8.5812752e-07,  1.7439456e-06,  2.9441997e-06]], dtype=float32)>, <tf.Tensor: id=911025, shape=(110,), dtype=float32, numpy=\n",
      "array([ 6.16349098e-06,  6.70106441e-04,  6.10533652e-06,  6.03624430e-05,\n",
      "       -9.28400201e-04,  3.15263424e-05,  2.05552235e-04,  7.62435229e-06,\n",
      "        7.60766534e-06,  1.03186103e-05,  9.54124436e-04,  4.47266822e-04,\n",
      "        4.65938094e-04, -1.27005100e-04,  1.56471433e-05, -3.34915647e-04,\n",
      "       -1.36372948e-03,  2.49082688e-04, -2.00057606e-04,  1.25486782e-04,\n",
      "        7.04647246e-05, -1.31711684e-04,  2.24817166e-04,  7.10415407e-05,\n",
      "       -3.13018973e-04, -4.33673762e-04,  1.56648253e-04,  2.04006094e-04,\n",
      "       -2.00758121e-04, -2.50564335e-05,  6.63383165e-04, -1.37452604e-04,\n",
      "        1.19803608e-05, -2.89041578e-04, -3.59649130e-04,  5.82683948e-04,\n",
      "       -1.52143330e-05,  4.80443588e-04, -2.58267275e-04, -5.23749695e-05,\n",
      "       -1.21889658e-04,  1.82389485e-04,  4.78673232e-04, -7.90954626e-04,\n",
      "        1.59089701e-04, -4.52846376e-04, -1.72676053e-04,  6.30552939e-04,\n",
      "        2.11897146e-04,  1.54411377e-04,  3.32149240e-04,  6.00431413e-05,\n",
      "        9.20817838e-05,  3.69253859e-04, -4.80371556e-04,  9.77206000e-05,\n",
      "       -3.92917718e-05,  1.18281583e-04, -1.07363972e-04,  2.54327344e-04,\n",
      "        3.57853060e-05,  6.68205030e-06,  5.59072669e-06, -4.30302462e-04,\n",
      "        2.21771807e-05,  6.60813553e-03, -5.79122454e-04, -8.42532900e-04,\n",
      "       -3.99281969e-03, -4.98173898e-03,  3.00293742e-03, -1.67812733e-03,\n",
      "        4.66366624e-03, -4.56244592e-03,  3.23785178e-04, -1.06457702e-03,\n",
      "        2.04803934e-03, -2.73901038e-04,  6.79822499e-03, -1.33659150e-02,\n",
      "        1.27844186e-03, -4.14863491e-04,  1.39372107e-02, -5.98697225e-04,\n",
      "       -1.31226587e-03, -1.12218119e-03, -1.60200987e-03, -4.12727270e-04,\n",
      "       -6.53212017e-04, -2.60263821e-03,  3.81662074e-04,  8.78936862e-06,\n",
      "        6.77703065e-06,  1.07161004e-05,  7.66384437e-06,  6.44606780e-06,\n",
      "        7.95983397e-06,  6.58460021e-06,  6.11395626e-06,  6.96783491e-06,\n",
      "        7.85527027e-06,  7.03973274e-06,  7.77107107e-06,  6.50169841e-06,\n",
      "       -1.44461344e-04, -1.49623229e-04,  7.32466833e-06,  5.71156625e-06,\n",
      "        1.12184425e-05,  1.68492497e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafcf60>, <tf.Tensor: id=911489, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-1.2791519e-06,  4.4336352e-07, -9.9798262e-07, ...,\n",
      "         8.7649742e-06, -6.6117580e-05, -1.4748464e-05],\n",
      "       [-4.3114770e-07, -8.8764700e-08,  2.6583996e-07, ...,\n",
      "        -7.6318120e-06, -1.3960653e-06,  5.9502958e-07],\n",
      "       [-3.4676466e-09, -1.5297772e-07,  2.0218560e-07, ...,\n",
      "        -8.7180197e-05,  1.2582200e-06, -4.9597693e-05],\n",
      "       ...,\n",
      "       [ 3.3203357e-06,  1.1374343e-06, -9.5662165e-07, ...,\n",
      "         1.5673697e-05, -9.7443099e-06, -2.2820789e-06],\n",
      "       [-5.7795457e-08,  4.8730925e-08,  5.9509895e-09, ...,\n",
      "         1.0457061e-05, -5.8969119e-05,  2.0240930e-05],\n",
      "       [ 7.5006386e-07,  5.3229201e-07, -8.2416722e-07, ...,\n",
      "        -9.5074640e-05,  6.1223304e-06, -1.4644158e-05]], dtype=float32)>, <tf.Tensor: id=911490, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 2.4111929e-05,  5.2919722e-06,  6.2097693e-06, ...,\n",
      "         3.4126770e-04, -1.0642806e-05,  1.9791877e-04],\n",
      "       [ 5.1949507e-07,  6.5903841e-08,  1.2398795e-06, ...,\n",
      "        -5.4780382e-04,  5.0790939e-05, -6.9677233e-05],\n",
      "       [ 1.0499650e-06, -3.0473581e-07,  4.6912655e-06, ...,\n",
      "         4.6991379e-05, -1.5968636e-06,  4.5878012e-05],\n",
      "       ...,\n",
      "       [-1.2414935e-08, -6.2602297e-08, -8.8281467e-08, ...,\n",
      "         8.2830966e-06, -4.4022531e-07,  1.0712499e-06],\n",
      "       [ 1.9612671e-06,  3.5063149e-06,  7.4523450e-06, ...,\n",
      "        -1.9038467e-04,  5.6966169e-06, -5.7683410e-05],\n",
      "       [ 3.2466880e-05,  7.2419920e-05, -4.4535220e-05, ...,\n",
      "        -4.7976253e-04,  4.0244617e-05,  4.0998322e-05]], dtype=float32)>, <tf.Tensor: id=911491, shape=(6144,), dtype=float32, numpy=\n",
      "array([-2.0031983e-05, -2.6446066e-06, -4.1127528e-06, ...,\n",
      "        8.4209105e-06, -1.9182841e-04, -4.8430561e-04], dtype=float32)>, <tf.Tensor: id=911394, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.5411818e-07,  1.3766871e-04, -9.9538374e-07, ...,\n",
      "        -8.5581689e-07, -3.6062545e-08, -3.0206456e-06],\n",
      "       [-1.4835582e-06,  1.4498348e-04, -1.7969260e-06, ...,\n",
      "        -1.3842341e-06,  7.8031830e-05, -5.4588345e-06],\n",
      "       [-4.4723677e-07, -1.5236216e-04, -6.4257404e-07, ...,\n",
      "        -5.3041822e-07,  9.7900062e-05, -2.4446190e-06],\n",
      "       ...,\n",
      "       [ 5.7381108e-06, -8.0703944e-04,  5.7005468e-06, ...,\n",
      "         5.3898434e-06, -1.4432284e-04,  1.6698465e-05],\n",
      "       [-6.7296247e-07,  3.3487431e-06, -7.1940826e-07, ...,\n",
      "        -6.6846837e-07,  1.4034927e-05, -2.1567487e-06],\n",
      "       [ 8.3156925e-07, -1.8024044e-04,  1.1593992e-06, ...,\n",
      "         8.9129594e-07,  5.7529855e-05,  2.9015380e-06]], dtype=float32)>, <tf.Tensor: id=911386, shape=(110,), dtype=float32, numpy=\n",
      "array([ 5.82669327e-06, -7.93846149e-04,  5.77915034e-06, -1.37343653e-03,\n",
      "       -6.81801513e-03,  3.14861099e-05,  2.09466278e-04,  7.52142796e-06,\n",
      "        7.22458617e-06,  9.68709264e-06,  1.44491089e-04, -2.54543353e-04,\n",
      "       -2.49897654e-04,  2.86562044e-05,  1.54891604e-05,  8.20608053e-04,\n",
      "        1.00773037e-03,  6.85375708e-04,  1.11314934e-04,  3.35556018e-04,\n",
      "       -6.77405405e-05, -2.66688061e-04,  4.15763352e-04,  4.09763219e-04,\n",
      "       -6.56677701e-04,  1.52912020e-04,  1.53392131e-04,  3.43313499e-04,\n",
      "       -1.69958570e-04, -1.92450010e-04,  3.39277176e-04,  1.90641804e-05,\n",
      "        1.19505939e-05,  2.37352888e-05,  2.65697949e-04,  1.50727457e-04,\n",
      "        3.50008981e-04, -2.73731712e-04,  2.43079325e-04,  3.51105555e-05,\n",
      "       -4.02161153e-04,  2.38221419e-05,  1.97919711e-04,  5.62494679e-04,\n",
      "       -1.49381158e-04, -2.75552418e-04, -3.27994290e-04, -2.25679541e-04,\n",
      "       -4.72569373e-05, -2.58178799e-04,  2.29931138e-05,  5.93889417e-05,\n",
      "        2.83542322e-04, -8.20260029e-04,  2.66366696e-04, -2.17344641e-04,\n",
      "       -3.86286520e-05,  4.26277329e-05,  4.91667815e-05,  1.07854750e-04,\n",
      "        3.56939345e-05,  1.59837757e-04,  1.64275232e-04,  3.84390914e-05,\n",
      "        2.16197495e-05,  8.85616709e-03, -9.39161167e-04, -1.51422911e-03,\n",
      "       -4.48314613e-03, -1.12800831e-02,  1.04462134e-03,  1.26120605e-04,\n",
      "        3.43803782e-03, -3.30871693e-03, -6.84822444e-04, -4.32352535e-04,\n",
      "       -5.33970864e-03,  3.35939950e-03,  7.14712869e-03,  5.44454949e-03,\n",
      "       -1.57973263e-04, -4.21574223e-04,  5.62680745e-03, -3.15797370e-04,\n",
      "       -2.12295074e-03, -1.27014285e-03,  8.83670407e-04, -1.17909716e-04,\n",
      "       -2.14196611e-04,  2.28417153e-03,  5.59472755e-05,  8.40263328e-06,\n",
      "        6.66297092e-06,  1.02731747e-05,  7.23626908e-06,  6.20639639e-06,\n",
      "        7.51464995e-06,  6.36419645e-06,  6.02131331e-06,  6.74644343e-06,\n",
      "        7.55533256e-06,  6.79111463e-06,  7.49832407e-06,  6.31797775e-06,\n",
      "        1.17325380e-05,  6.37268886e-06, -1.49124506e-04,  5.46956198e-06,\n",
      "       -1.45448881e-04,  1.69240120e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc748>, <tf.Tensor: id=911850, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 3.9412494e-06,  2.2884553e-06, -4.2418337e-06, ...,\n",
      "         4.0282011e-05, -6.3017200e-05, -2.4802623e-05],\n",
      "       [ 5.3926584e-07,  5.8683000e-08, -1.0471595e-08, ...,\n",
      "         3.6156827e-05, -3.8550880e-07,  3.5663453e-07],\n",
      "       [-3.4602979e-07, -6.8470804e-08, -6.0080858e-08, ...,\n",
      "         8.1840291e-05,  5.5539749e-06,  1.5147037e-05],\n",
      "       ...,\n",
      "       [ 4.0142191e-07, -2.2127725e-07, -7.4889172e-08, ...,\n",
      "         8.7374938e-06, -1.7042035e-05, -7.1552454e-06],\n",
      "       [-6.2776145e-08,  9.8931324e-08,  1.6944719e-08, ...,\n",
      "        -5.2588497e-05,  3.7468442e-05, -2.9222829e-05],\n",
      "       [-9.3292232e-07,  1.1416457e-06,  1.4440504e-07, ...,\n",
      "        -3.0358609e-05, -1.6110158e-05,  1.0093213e-06]], dtype=float32)>, <tf.Tensor: id=911851, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 3.09307652e-05,  1.60566869e-05,  6.27464578e-06, ...,\n",
      "         1.37598964e-03, -1.12047695e-04,  3.52730742e-04],\n",
      "       [ 1.83797840e-06,  2.98709574e-06,  4.29772354e-06, ...,\n",
      "        -3.95626470e-04,  5.33313614e-05, -9.82696074e-05],\n",
      "       [-5.76633738e-07, -9.73995725e-07, -4.90729462e-06, ...,\n",
      "        -4.75382083e-04,  5.87305221e-05, -6.14733508e-05],\n",
      "       ...,\n",
      "       [-6.58686119e-08,  1.84879369e-07, -6.28381827e-08, ...,\n",
      "        -2.42114838e-05,  2.00831300e-06, -5.24764073e-06],\n",
      "       [-7.46356818e-06, -1.89501097e-05, -1.18682055e-05, ...,\n",
      "         3.91257548e-04, -4.06880936e-05,  1.47009647e-04],\n",
      "       [-1.86814796e-05, -1.64551238e-05,  6.16759689e-06, ...,\n",
      "         2.43100949e-04, -8.13789520e-05,  3.72236973e-05]], dtype=float32)>, <tf.Tensor: id=911852, shape=(6144,), dtype=float32, numpy=\n",
      "array([-5.0270148e-05, -9.5190026e-06,  4.5275042e-06, ...,\n",
      "       -2.4571558e-05,  3.9550042e-04,  2.4573473e-04], dtype=float32)>, <tf.Tensor: id=911755, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.44463443e-07, -6.73387112e-05, -1.03695584e-06, ...,\n",
      "        -8.89793739e-07, -1.82503595e-06, -3.21827747e-06],\n",
      "       [-1.48265542e-06, -9.25904751e-05, -1.87435876e-06, ...,\n",
      "        -1.41914632e-06, -3.40793395e-06, -5.71202054e-06],\n",
      "       [-5.88555736e-07, -2.06093115e-04, -7.86457747e-07, ...,\n",
      "        -7.07553113e-07, -1.72141904e-06, -3.06905326e-06],\n",
      "       ...,\n",
      "       [ 5.44193063e-06,  1.09739951e-03,  5.84763166e-06, ...,\n",
      "         5.35569325e-06,  1.08905597e-05,  1.71104621e-05],\n",
      "       [-6.26645374e-07, -1.00885525e-04, -7.25665814e-07, ...,\n",
      "        -6.52815515e-07, -1.24311487e-06, -2.17691172e-06],\n",
      "       [ 1.07632036e-06, -2.19831796e-04,  1.36618655e-06, ...,\n",
      "         1.08361064e-06,  2.12857913e-06,  3.47113519e-06]], dtype=float32)>, <tf.Tensor: id=911747, shape=(110,), dtype=float32, numpy=\n",
      "array([ 5.51810172e-06,  1.11835171e-03,  5.92398465e-06, -7.41619209e-04,\n",
      "        1.54117057e-02,  3.28168244e-05,  1.09204254e-03,  7.28057103e-06,\n",
      "        6.94971186e-06,  9.53835115e-06,  5.82264503e-04, -4.60012263e-04,\n",
      "       -3.88249027e-04,  2.95432219e-05,  1.47821565e-05, -2.86958017e-03,\n",
      "        6.56836201e-04,  1.44079013e-03, -3.88409098e-05, -2.91741162e-04,\n",
      "        1.57226808e-04, -1.88751379e-04, -7.75336739e-05,  1.53834786e-04,\n",
      "        8.30758363e-05, -7.91247148e-05,  5.42584981e-04, -1.66771006e-05,\n",
      "        2.27257246e-04,  1.55440313e-04,  8.67745781e-04,  1.84246037e-05,\n",
      "        1.18228290e-05, -4.44739329e-04, -3.45463632e-04,  5.59778593e-04,\n",
      "       -3.08775518e-04,  1.49880419e-04,  6.67304994e-05,  5.85048459e-04,\n",
      "       -1.08574990e-04, -1.32160087e-04,  2.43258924e-04, -9.90214408e-04,\n",
      "        1.50559441e-04, -2.64711271e-04, -1.87444224e-04, -8.27723488e-06,\n",
      "       -5.07953984e-04, -1.95705725e-04, -1.67239108e-04,  5.79738153e-05,\n",
      "        3.82741564e-04, -1.00054385e-04,  1.33300637e-06,  9.03137479e-05,\n",
      "        1.14143375e-04, -5.62787638e-04,  4.68635808e-05,  6.80167868e-05,\n",
      "       -1.19480654e-04, -1.61355289e-04, -1.34488291e-04,  3.87592190e-05,\n",
      "        2.19769208e-05,  4.24652500e-03,  2.40291143e-03,  5.30503038e-03,\n",
      "       -1.79212994e-03, -2.14026365e-02, -5.47741493e-03,  1.10063853e-03,\n",
      "        6.97156321e-03,  9.17590689e-04, -6.66351916e-05, -2.93351826e-04,\n",
      "       -2.99175759e-03, -1.47124589e-03, -2.33013183e-03,  2.82047503e-03,\n",
      "       -1.22666801e-03, -1.21233599e-04,  4.51150350e-03, -4.38403897e-03,\n",
      "       -1.93532114e-03,  3.75065370e-04, -3.62089573e-04,  2.68649921e-04,\n",
      "       -1.34344527e-03,  5.08654339e-04,  3.01951572e-04,  8.75642036e-06,\n",
      "        6.38542224e-06,  9.86343821e-06,  6.95414928e-06,  6.00064777e-06,\n",
      "        7.35563890e-06,  6.15815725e-06,  5.72594672e-06,  6.51961591e-06,\n",
      "        7.40620089e-06,  6.48567038e-06,  7.42678139e-06,  6.09253993e-06,\n",
      "        1.18398912e-05,  6.11413680e-06,  7.08279458e-06,  5.42889165e-06,\n",
      "        1.10367237e-05,  1.73272565e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafca58>, <tf.Tensor: id=912211, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 1.30764033e-06, -1.45754484e-06, -1.16655269e-06, ...,\n",
      "        -8.32869046e-05,  3.92994552e-05,  3.22554806e-05],\n",
      "       [ 1.60512698e-07,  6.69231710e-08, -4.92236758e-08, ...,\n",
      "         6.91513487e-05,  1.12399712e-05,  5.04949185e-06],\n",
      "       [-1.28179650e-07, -1.63783028e-07,  5.86662310e-08, ...,\n",
      "         1.55464750e-05,  4.27967861e-05,  7.92661467e-06],\n",
      "       ...,\n",
      "       [-3.04891159e-07, -1.44378149e-07,  1.92405579e-07, ...,\n",
      "        -1.01048963e-05,  1.33606145e-05,  3.96639780e-06],\n",
      "       [-3.92000139e-08,  3.59945602e-08,  3.24613687e-08, ...,\n",
      "        -2.61858440e-05, -7.86441597e-05,  9.67492542e-06],\n",
      "       [-6.89595794e-08,  5.33828597e-07,  3.71438688e-07, ...,\n",
      "        -8.05818709e-05,  3.90478635e-06, -8.09405083e-06]], dtype=float32)>, <tf.Tensor: id=912212, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 8.1793896e-06,  2.8691959e-06,  1.6411939e-06, ...,\n",
      "         2.1860460e-04, -1.3481548e-05,  5.6036079e-05],\n",
      "       [-7.4026744e-08,  3.8079946e-07,  1.1038749e-06, ...,\n",
      "         9.7557742e-05, -1.4786961e-05, -5.8056849e-05],\n",
      "       [-1.4593031e-06, -7.1636157e-07, -1.2500368e-06, ...,\n",
      "        -3.4237324e-05,  1.4005800e-06,  3.5321893e-05],\n",
      "       ...,\n",
      "       [ 4.9136810e-07,  9.3580309e-07,  3.7576666e-07, ...,\n",
      "        -4.3521552e-05,  5.2801847e-06, -9.2800601e-06],\n",
      "       [ 1.8387120e-05,  1.4340547e-05,  2.5601552e-05, ...,\n",
      "        -2.8155113e-04,  1.4680212e-05, -4.3271273e-05],\n",
      "       [-6.0638081e-06, -3.7839177e-06,  8.6597811e-06, ...,\n",
      "        -2.1792822e-04,  9.7882621e-06, -4.9083712e-05]], dtype=float32)>, <tf.Tensor: id=912213, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 4.8830502e-06, -1.2023577e-06,  5.2471878e-06, ...,\n",
      "       -4.3992677e-05, -2.8500607e-04, -2.1949789e-04], dtype=float32)>, <tf.Tensor: id=912116, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.9062030e-07, -9.2959534e-05, -1.1101564e-06, ...,\n",
      "        -9.3775873e-07, -1.9218342e-06, -3.3382812e-06],\n",
      "       [-1.3332156e-06, -1.0455732e-05, -1.7552894e-06, ...,\n",
      "        -1.2648566e-06, -3.1267955e-06, -5.3702997e-06],\n",
      "       [-5.5941302e-07, -2.7180932e-04, -8.0706951e-07, ...,\n",
      "        -7.1032292e-07, -1.6748561e-06, -2.9609782e-06],\n",
      "       ...,\n",
      "       [ 5.9674094e-06,  7.1350200e-04,  6.3388225e-06, ...,\n",
      "         5.8815949e-06,  1.1400123e-05,  1.7832352e-05],\n",
      "       [-8.0132895e-07, -1.7735311e-04, -8.8477532e-07, ...,\n",
      "        -8.2785778e-07, -1.4928273e-06, -2.5097461e-06],\n",
      "       [ 1.0327392e-06,  8.8928369e-05,  1.3167178e-06, ...,\n",
      "         1.0275674e-06,  2.0082098e-06,  3.3834222e-06]], dtype=float32)>, <tf.Tensor: id=912108, shape=(110,), dtype=float32, numpy=\n",
      "array([ 6.05785317e-06,  7.28546176e-04,  6.42684017e-06,  7.95814485e-05,\n",
      "       -6.48717396e-03, -2.80029490e-04,  6.85983512e-04,  8.11269092e-06,\n",
      "        7.57581392e-06,  1.00206898e-05, -3.19538463e-04,  1.51958433e-04,\n",
      "        3.53961193e-04, -1.24558996e-04,  1.51005070e-05,  3.20533104e-03,\n",
      "        1.76314090e-04,  1.44491228e-03, -4.09478707e-05, -2.93161458e-04,\n",
      "       -1.40427932e-04, -5.29650788e-05, -7.51112602e-05, -1.81572075e-04,\n",
      "        7.21954639e-05, -2.63430993e-04,  2.27819575e-04,  2.11122009e-04,\n",
      "       -4.46821563e-04, -2.51894362e-05,  4.38140036e-04,  2.06862387e-05,\n",
      "        1.23143900e-05,  2.62357389e-05, -5.16324129e-04,  6.53561379e-04,\n",
      "        5.62161658e-05, -7.60096009e-04,  2.53900245e-04,  8.04543612e-04,\n",
      "        8.86824710e-05,  1.83996031e-04,  2.54788902e-04,  8.94289173e-04,\n",
      "        1.54629262e-04,  7.31540349e-05, -1.75100853e-04,  4.91550381e-05,\n",
      "        4.90277831e-04,  3.21192463e-04, -4.65882476e-04,  5.84512491e-05,\n",
      "        2.33522005e-04, -2.07542398e-05, -5.26007905e-04,  9.33724659e-05,\n",
      "        1.21473306e-04,  1.32258821e-04,  4.97271249e-05,  2.40737005e-04,\n",
      "        3.91301000e-05, -1.83800921e-06,  1.84983452e-04,  4.36662776e-05,\n",
      "        2.30759251e-05,  5.28255058e-03, -8.95081321e-05,  2.08465476e-03,\n",
      "       -1.30628527e-03, -6.15501031e-03, -9.16641089e-04,  1.55566365e-03,\n",
      "       -7.74016080e-04,  1.80385355e-03, -4.31518827e-04, -1.51960470e-04,\n",
      "       -2.69799563e-03,  8.64422240e-04, -4.04812954e-03,  4.24295245e-03,\n",
      "        2.73928745e-04, -5.91873075e-04, -7.89290201e-03, -7.32886977e-03,\n",
      "        4.26041242e-03,  4.86128964e-03,  2.49029393e-03, -8.36046878e-04,\n",
      "       -1.77153299e-04,  3.22482409e-03,  1.01714992e-04,  9.49489277e-06,\n",
      "        6.68316579e-06,  1.05493318e-05,  7.78747199e-06,  6.47828256e-06,\n",
      "        7.96327367e-06,  6.94393066e-06,  6.34112484e-06,  7.00500959e-06,\n",
      "        7.83446012e-06,  7.13957888e-06,  7.68562586e-06,  6.66311826e-06,\n",
      "        1.27156827e-05,  6.69487190e-06,  7.52072174e-06,  5.96862765e-06,\n",
      "        1.15612393e-05,  1.80698808e-05], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc2b0>, <tf.Tensor: id=912572, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-7.7129725e-06, -4.5790052e-06,  2.5612558e-06, ...,\n",
      "        -7.6613047e-05, -1.5975515e-05,  3.6064521e-05],\n",
      "       [-2.4871619e-07, -6.6074807e-07,  6.9108815e-08, ...,\n",
      "         6.5980596e-05,  3.5409645e-05,  1.3214111e-05],\n",
      "       [ 1.8423711e-07, -1.5171419e-07, -8.1794305e-08, ...,\n",
      "        -1.6000286e-05,  2.8937222e-05,  4.8067295e-06],\n",
      "       ...,\n",
      "       [-6.5532578e-07, -6.4345620e-08,  3.6559965e-07, ...,\n",
      "         2.8476886e-06,  2.6482837e-06, -3.0425056e-06],\n",
      "       [-8.2458534e-10, -3.1723726e-08, -4.1908841e-08, ...,\n",
      "        -1.5737358e-05, -1.9429668e-05, -6.1927465e-05],\n",
      "       [ 2.8083113e-07,  1.0305736e-06, -6.8148165e-07, ...,\n",
      "        -9.6972044e-06, -9.0619742e-06,  7.6410498e-07]], dtype=float32)>, <tf.Tensor: id=912573, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-1.73609205e-05, -1.84820383e-05, -1.44305195e-05, ...,\n",
      "        -7.79535098e-04,  1.18781121e-04, -2.08874349e-04],\n",
      "       [ 7.24859149e-07, -1.19587071e-06,  3.00526904e-06, ...,\n",
      "        -2.51894788e-04,  4.16679140e-05, -5.02674266e-05],\n",
      "       [-2.65539893e-06, -1.01966691e-06, -5.48312801e-06, ...,\n",
      "        -7.59990216e-05,  6.83533290e-06,  2.52168138e-05],\n",
      "       ...,\n",
      "       [-6.13884907e-08, -2.52065377e-07,  4.58962234e-07, ...,\n",
      "        -1.42363187e-05,  2.52628979e-06, -3.16377464e-06],\n",
      "       [-9.90122771e-06, -6.46841636e-06,  4.49312620e-06, ...,\n",
      "        -5.04317286e-04,  7.56764275e-05, -1.89068494e-04],\n",
      "       [-1.38455489e-05, -1.90701830e-05,  8.33070862e-06, ...,\n",
      "         1.08495231e-04, -9.41923918e-06,  1.02788188e-04]], dtype=float32)>, <tf.Tensor: id=912574, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 5.0774535e-05, -2.5964518e-06,  9.7434322e-06, ...,\n",
      "       -1.4307294e-05, -5.1113393e-04,  1.0929046e-04], dtype=float32)>, <tf.Tensor: id=912477, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.0258965e-06, -7.0255919e-05, -1.1943057e-06, ...,\n",
      "        -1.0587777e-06, -2.1340434e-06, -3.5979476e-06],\n",
      "       [-1.3863224e-06, -1.5374014e-04, -1.7834161e-06, ...,\n",
      "        -1.3266495e-06, -3.2919879e-06, -5.4763777e-06],\n",
      "       [-6.7573689e-07,  2.0513934e-04, -8.5744415e-07, ...,\n",
      "        -8.2488702e-07, -1.8439588e-06, -3.2060188e-06],\n",
      "       ...,\n",
      "       [ 6.4116484e-06,  5.6479510e-04,  6.5991017e-06, ...,\n",
      "         6.3218822e-06,  1.2364951e-05,  1.8735471e-05],\n",
      "       [-7.7535253e-07, -1.0596272e-05, -8.6940321e-07, ...,\n",
      "        -7.9799008e-07, -1.4974430e-06, -2.5208997e-06],\n",
      "       [ 9.8147905e-07,  2.8252701e-04,  1.3152940e-06, ...,\n",
      "         9.7899851e-07,  2.0758439e-06,  3.3514150e-06]], dtype=float32)>, <tf.Tensor: id=912469, shape=(110,), dtype=float32, numpy=\n",
      "array([ 6.5063282e-06,  5.6789158e-04,  6.6890670e-06,  6.1292882e-04,\n",
      "       -9.3390811e-03,  3.4145647e-05, -7.5770979e-04,  8.4566382e-06,\n",
      "        8.2327715e-06,  1.0221879e-05, -2.1746889e-04, -8.2530413e-04,\n",
      "       -6.4954790e-04, -1.2239839e-04,  1.5537737e-05, -1.4927449e-03,\n",
      "        1.6831376e-03,  1.9137097e-03, -3.8262948e-05,  5.9645059e-05,\n",
      "        2.4618141e-04, -2.9950798e-04, -1.8333289e-04,  3.4518715e-04,\n",
      "        1.6564275e-04, -1.5988310e-04, -1.9591556e-04,  1.9032372e-05,\n",
      "       -1.0960860e-03, -2.0864292e-04,  6.5805559e-04,  2.1632919e-05,\n",
      "        1.3335204e-05,  2.9660074e-05, -3.5354565e-04,  7.0584979e-04,\n",
      "       -2.8690585e-04, -2.8844934e-04,  2.4281425e-04, -2.1491456e-04,\n",
      "       -7.4312702e-05,  1.7731597e-04, -2.9447209e-04, -1.0949641e-03,\n",
      "       -1.1480153e-05,  8.3976949e-05, -2.3910266e-05,  1.6695610e-04,\n",
      "        6.2805790e-05, -8.1072981e-04, -4.7330192e-04, -9.9638215e-05,\n",
      "        2.4655812e-05,  1.3081898e-04, -6.6230644e-04,  8.9515037e-05,\n",
      "       -3.1790099e-05, -8.4053609e-04,  4.9740625e-05,  5.3931588e-05,\n",
      "        4.0435869e-05,  1.4496970e-04,  1.8463048e-04,  4.7735106e-05,\n",
      "        2.4178316e-05, -1.4308961e-03, -2.3322115e-03,  4.4618975e-03,\n",
      "        2.6047232e-03, -6.6425477e-05, -1.0983606e-03,  2.0733946e-03,\n",
      "       -4.0984023e-03,  9.8384358e-03,  3.4558043e-04,  3.4186314e-04,\n",
      "        1.0214788e-04, -1.8532387e-03, -6.8921405e-03,  6.7230328e-03,\n",
      "       -3.2523738e-05, -4.2274638e-04, -1.1889631e-02,  7.3074633e-03,\n",
      "        6.3366727e-03,  2.1301233e-04,  5.2144128e-04, -3.2173068e-04,\n",
      "       -4.0790375e-04,  1.5036241e-03,  7.9190498e-04,  1.0016051e-05,\n",
      "        7.2097564e-06,  1.1399106e-05,  8.3472123e-06,  6.9218008e-06,\n",
      "        8.4566418e-06,  7.4758927e-06,  6.9112675e-06,  7.5432990e-06,\n",
      "        8.3446557e-06,  7.6876640e-06,  8.1870339e-06,  7.2116627e-06,\n",
      "        1.3177665e-05,  7.0027127e-06,  8.1823755e-06,  6.4117844e-06,\n",
      "        1.2535742e-05,  1.8980125e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc3c8>, <tf.Tensor: id=912933, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-6.3938191e-06, -7.7408077e-07,  6.2169880e-07, ...,\n",
      "        -5.5671684e-05,  4.7474764e-06,  5.7103184e-06],\n",
      "       [-4.3721101e-07,  1.6855599e-07,  2.6329229e-07, ...,\n",
      "         5.1227630e-06,  1.0879150e-05,  3.5106033e-05],\n",
      "       [ 2.7151161e-08,  2.3993857e-07, -2.0783932e-07, ...,\n",
      "        -1.7272565e-06,  4.9474170e-06,  2.6156546e-05],\n",
      "       ...,\n",
      "       [-8.1598337e-08,  1.5014498e-07, -2.5860169e-08, ...,\n",
      "         2.8915792e-06,  6.9178350e-06,  2.3144237e-06],\n",
      "       [ 6.3549372e-08,  7.5610814e-08, -4.5842210e-08, ...,\n",
      "        -2.9064593e-05,  8.5763195e-06,  9.3776825e-06],\n",
      "       [ 2.5392268e-07, -1.9404453e-07,  1.1130554e-06, ...,\n",
      "        -1.4201446e-05, -5.5758890e-05,  1.7279257e-05]], dtype=float32)>, <tf.Tensor: id=912934, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-1.1876165e-05, -6.8254094e-06,  5.4541879e-06, ...,\n",
      "        -6.2483270e-04,  1.2048978e-04, -2.0588806e-04],\n",
      "       [ 8.5254180e-07,  1.0068402e-06, -4.5984194e-07, ...,\n",
      "         6.0850108e-04, -9.5229610e-05,  7.5656717e-05],\n",
      "       [ 9.2503234e-07,  8.7743308e-07, -4.7804756e-06, ...,\n",
      "        -3.0760319e-04,  5.9275662e-05, -1.0471162e-04],\n",
      "       ...,\n",
      "       [-5.2861317e-08, -6.6269671e-08,  2.6349670e-07, ...,\n",
      "        -4.0167743e-06,  2.3788196e-08, -2.6322750e-06],\n",
      "       [ 9.1171441e-06,  9.1704214e-06,  2.1812597e-05, ...,\n",
      "        -4.7497237e-05, -1.3407476e-05,  9.9021439e-05],\n",
      "       [-1.3080842e-05, -1.9568575e-05,  8.8083934e-06, ...,\n",
      "        -1.5397582e-04, -2.1816160e-05, -8.7916662e-05]], dtype=float32)>, <tf.Tensor: id=912935, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 2.1497584e-05, -5.8839714e-06, -2.3270370e-06, ...,\n",
      "       -4.0007408e-06, -5.0231552e-05, -1.5672801e-04], dtype=float32)>, <tf.Tensor: id=912838, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.0261592e-06, -2.1639754e-04, -1.2294506e-06, ...,\n",
      "        -1.0625649e-06,  3.3745600e-05, -3.7413320e-06],\n",
      "       [-1.2838437e-06, -2.4327914e-05, -1.7249095e-06, ...,\n",
      "        -1.2297809e-06,  8.1483049e-05, -5.4460693e-06],\n",
      "       [-7.5453602e-07, -6.9983571e-04, -9.1513931e-07, ...,\n",
      "        -8.7805972e-07,  6.9003465e-05, -3.4716047e-06],\n",
      "       ...,\n",
      "       [ 7.1403319e-06,  2.2234884e-03,  7.0751239e-06, ...,\n",
      "         6.8387217e-06, -1.4170876e-04,  1.9967119e-05],\n",
      "       [-8.8855847e-07, -2.7080157e-04, -9.4646430e-07, ...,\n",
      "        -8.9593141e-07,  1.0989708e-05, -2.7353396e-06],\n",
      "       [ 1.0431039e-06, -3.7559727e-04,  1.3838912e-06, ...,\n",
      "         1.0637460e-06, -3.5689765e-05,  3.5216872e-06]], dtype=float32)>, <tf.Tensor: id=912830, shape=(110,), dtype=float32, numpy=\n",
      "array([ 7.25513291e-06,  2.25994480e-03,  7.18036290e-06, -3.47042078e-06,\n",
      "        2.18542502e-03,  3.67820503e-05, -2.18062225e-04,  9.26419307e-06,\n",
      "        8.93341439e-06,  1.08044678e-05, -7.51045241e-04, -6.51658047e-04,\n",
      "       -1.59665506e-04, -1.19112774e-04,  1.63812747e-05,  9.87280277e-04,\n",
      "       -8.16897082e-04,  1.80483027e-03,  1.23698599e-04,  2.67650921e-06,\n",
      "       -1.97829155e-04, -6.53669122e-04,  1.20661418e-04, -4.92083374e-04,\n",
      "       -3.07954353e-04, -2.03949894e-04,  2.14504689e-04, -2.41702903e-04,\n",
      "       -6.93626425e-05,  1.20896802e-04,  1.37847484e-04,  2.37589848e-05,\n",
      "        1.41777900e-05,  3.31279407e-05, -1.16899191e-05, -2.75667495e-04,\n",
      "       -2.45185365e-04, -4.12279362e-04,  9.25401255e-05,  5.65102149e-04,\n",
      "        2.64889299e-04,  1.66657701e-05,  4.77802212e-04,  4.64119657e-04,\n",
      "       -7.45138323e-06,  1.08770721e-04, -1.63927747e-04,  5.18264889e-04,\n",
      "        4.29960055e-04,  4.75808702e-06,  1.82491727e-04,  5.83532419e-05,\n",
      "       -1.26826330e-04, -7.41552794e-04,  2.40869849e-06, -6.67829081e-05,\n",
      "       -2.90085190e-05,  1.67102873e-04, -4.16434981e-04,  2.10100188e-04,\n",
      "        4.25069011e-05,  1.43597863e-04,  1.92831110e-04,  5.35839063e-05,\n",
      "        2.51701276e-05, -9.26735718e-03,  7.29336811e-04,  3.28899943e-04,\n",
      "        1.28869980e-03,  1.17682964e-02, -2.83224112e-03,  2.08758516e-03,\n",
      "       -7.54811103e-04, -1.78160146e-03, -2.00400507e-04, -2.08537269e-04,\n",
      "       -3.74321389e-04, -1.51494984e-03, -6.63015898e-03, -2.36450299e-03,\n",
      "        8.69333569e-04,  5.68255200e-05, -1.14185140e-02,  4.21239156e-03,\n",
      "        1.32821559e-03,  1.97278289e-03,  3.40186246e-03,  3.92877636e-03,\n",
      "        4.96873516e-04, -6.20046456e-04,  7.03255588e-04,  1.07792075e-05,\n",
      "        7.56666213e-06,  1.26344503e-05,  9.14481825e-06,  7.61137289e-06,\n",
      "        9.40309201e-06,  8.50629567e-06,  7.62802347e-06,  8.14747273e-06,\n",
      "        8.96771235e-06,  8.50672131e-06,  8.71695374e-06,  7.73681950e-06,\n",
      "        1.48399167e-05,  7.75937315e-06,  8.91852414e-06,  6.94532855e-06,\n",
      "       -1.42937381e-04,  2.02465635e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc320>, <tf.Tensor: id=913294, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 1.5935921e-06,  4.5678421e-06,  1.3066357e-06, ...,\n",
      "        -1.1363355e-05, -3.3211461e-05,  1.4014269e-05],\n",
      "       [-5.8696224e-07,  9.8789890e-08, -1.6756810e-07, ...,\n",
      "         4.0285358e-05, -2.8794908e-05,  7.9225938e-06],\n",
      "       [-1.8138621e-07,  1.9266403e-07, -3.3458367e-08, ...,\n",
      "         7.8477351e-06,  1.9116631e-05, -7.0782612e-06],\n",
      "       ...,\n",
      "       [ 1.0345062e-06,  2.8331058e-07, -3.1329759e-07, ...,\n",
      "         2.0209180e-05, -2.1720607e-05,  3.5147380e-07],\n",
      "       [-1.3485739e-07,  1.2768688e-08,  6.4120279e-08, ...,\n",
      "        -1.9685060e-05,  5.4391970e-05,  1.9251513e-05],\n",
      "       [-1.0678516e-06, -9.3436768e-07,  1.1988834e-06, ...,\n",
      "        -3.4683133e-05, -3.8698431e-06,  2.6438165e-06]], dtype=float32)>, <tf.Tensor: id=913295, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 3.22195820e-06,  7.49889386e-06,  9.63164985e-06, ...,\n",
      "        -7.73649081e-04,  1.70702318e-04, -8.48547206e-05],\n",
      "       [-1.18521461e-06, -1.00236218e-06, -1.32592481e-06, ...,\n",
      "        -1.06287611e-04, -4.06120016e-05,  2.76354021e-05],\n",
      "       [-2.07205926e-06, -8.21202775e-07, -9.88367628e-06, ...,\n",
      "        -5.96906248e-05,  2.85681454e-05,  6.70422469e-06],\n",
      "       ...,\n",
      "       [-2.23927009e-06, -9.55537871e-07, -3.60792046e-06, ...,\n",
      "         9.16263234e-05, -8.22049424e-06,  1.33508565e-05],\n",
      "       [-1.19781935e-05, -1.75530549e-05, -5.36277503e-06, ...,\n",
      "        -3.99457087e-04,  4.14706046e-05, -5.10289101e-05],\n",
      "       [-2.67092446e-06,  3.43989586e-06, -2.07989106e-05, ...,\n",
      "        -1.87870814e-04, -8.94176355e-06,  5.99968189e-05]], dtype=float32)>, <tf.Tensor: id=913296, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 1.45116474e-06,  1.09416112e-06,  1.05502295e-05, ...,\n",
      "        9.68771274e-05, -4.27210034e-04, -1.86973266e-04], dtype=float32)>, <tf.Tensor: id=913199, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.1980890e-06,  1.3377662e-04, -1.3493540e-06, ...,\n",
      "        -1.1984621e-06, -2.4138483e-06, -4.1150288e-06],\n",
      "       [-1.5582250e-06,  1.4384493e-04, -1.9266647e-06, ...,\n",
      "        -1.4728814e-06, -3.5834632e-06, -6.0571424e-06],\n",
      "       [-7.0124872e-07,  1.2476683e-04, -8.9764148e-07, ...,\n",
      "        -8.3187024e-07, -2.0266650e-06, -3.5209850e-06],\n",
      "       ...,\n",
      "       [ 7.6767228e-06, -8.6998404e-04,  7.6017946e-06, ...,\n",
      "         7.3842125e-06,  1.4218025e-05,  2.1818378e-05],\n",
      "       [-9.0095938e-07,  1.6463130e-04, -9.6916256e-07, ...,\n",
      "        -9.0901153e-07, -1.6975555e-06, -2.8663453e-06],\n",
      "       [ 1.2301268e-06, -5.0156348e-05,  1.5422254e-06, ...,\n",
      "         1.2353660e-06,  2.4544356e-06,  3.9142160e-06]], dtype=float32)>, <tf.Tensor: id=913191, shape=(110,), dtype=float32, numpy=\n",
      "array([ 7.87594854e-06, -8.85199639e-04,  7.77865625e-06,  1.13429106e-03,\n",
      "       -2.50534806e-03,  3.92219008e-05, -8.20665387e-04,  1.03705070e-05,\n",
      "        9.96034760e-06,  1.19796405e-05, -1.02853496e-03, -7.57959846e-04,\n",
      "       -8.47992836e-04,  4.19207681e-05,  1.79335275e-05,  1.33045693e-03,\n",
      "        1.60322851e-03, -1.54203060e-03,  1.23922087e-04,  4.36990260e-04,\n",
      "        2.33709754e-04, -3.85245105e-04,  6.57002267e-04,  1.92057036e-04,\n",
      "        4.95004875e-04,  4.75524022e-04, -1.01528317e-03,  2.28251272e-04,\n",
      "       -2.05260876e-04, -2.22473376e-04,  1.87857790e-04,  2.54624701e-05,\n",
      "        1.57666946e-05,  3.76610114e-05,  3.09317809e-04, -2.44592491e-04,\n",
      "       -2.34099571e-04, -9.14245960e-04, -1.98566704e-05, -1.23189285e-03,\n",
      "        1.74680739e-04,  3.75368400e-05, -4.18542651e-04, -1.05657964e-03,\n",
      "       -3.04169516e-04, -3.16150486e-04, -1.23765611e-04, -3.01446307e-05,\n",
      "       -2.30344158e-04, -1.13226241e-03, -2.33953004e-04,  6.38241399e-05,\n",
      "       -6.82681566e-04,  2.27951576e-04, -3.04022059e-03, -5.83163273e-05,\n",
      "       -1.19562228e-05,  1.92893072e-04,  5.92231772e-05,  2.23620606e-04,\n",
      "        4.80875460e-05, -1.61429518e-04,  3.15391371e-05,  5.61545057e-05,\n",
      "        2.80240256e-05, -5.16401464e-03,  3.71900998e-04, -1.31495611e-03,\n",
      "        4.84023755e-03,  1.18924072e-02, -3.59892705e-03,  9.89859691e-04,\n",
      "        2.51550134e-03,  1.57347461e-03,  3.21532832e-04,  1.17816019e-03,\n",
      "        4.21824458e-04, -3.64125299e-04, -1.93968264e-03, -5.01351012e-03,\n",
      "        3.72214895e-03, -3.84398823e-04,  9.97128431e-04,  5.66309178e-03,\n",
      "       -2.22048885e-03,  2.80495849e-04, -5.45927556e-04,  9.05531168e-04,\n",
      "       -2.30632984e-04, -3.45460209e-03,  2.35255298e-04,  1.17711079e-05,\n",
      "        8.69230280e-06,  1.38945597e-05,  9.90431545e-06,  8.48815034e-06,\n",
      "        1.01742125e-05,  8.97459358e-06,  8.35825267e-06,  8.90796309e-06,\n",
      "        9.71471945e-06,  9.36891320e-06,  9.97561165e-06,  8.78196352e-06,\n",
      "        1.66399259e-05,  8.68776988e-06,  1.00814086e-05,  7.57054022e-06,\n",
      "        1.45412268e-05,  2.23092920e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc780>, <tf.Tensor: id=913655, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-5.70130453e-07, -4.85878616e-07, -2.57762139e-07, ...,\n",
      "        -3.87721011e-05,  1.24296357e-05,  5.96652244e-05],\n",
      "       [-4.41584746e-07, -3.19947020e-07,  3.92397084e-07, ...,\n",
      "        -3.19041646e-05,  4.14116994e-06,  8.47831780e-06],\n",
      "       [-4.94730557e-07,  1.16015446e-08, -6.38900985e-08, ...,\n",
      "         1.34977145e-05, -3.42613093e-05,  1.17086863e-07],\n",
      "       ...,\n",
      "       [ 3.55126588e-07,  1.19489727e-07, -4.55887744e-07, ...,\n",
      "         7.84500116e-06, -2.57026277e-05,  1.69303860e-06],\n",
      "       [-3.13072775e-07,  9.38447275e-09,  1.61032645e-07, ...,\n",
      "        -3.16607911e-05,  5.51611229e-05,  3.13600867e-05],\n",
      "       [ 2.41436879e-07,  1.10567498e-06, -2.02624832e-07, ...,\n",
      "         4.81477509e-05, -1.83415113e-06,  7.99429927e-06]], dtype=float32)>, <tf.Tensor: id=913656, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 2.1930059e-06,  2.0288187e-06,  6.1929863e-06, ...,\n",
      "         7.1145041e-05,  2.8465929e-05,  3.7188940e-05],\n",
      "       [-9.5784890e-07, -2.4626049e-06, -2.1746638e-07, ...,\n",
      "        -1.9463275e-04,  2.4191497e-05, -5.6420085e-05],\n",
      "       [ 6.8194794e-07,  4.4420821e-07, -1.8968606e-07, ...,\n",
      "        -3.3448789e-05, -6.9833000e-06, -5.7351906e-05],\n",
      "       ...,\n",
      "       [-1.3180991e-06, -2.4179467e-06, -7.6062423e-07, ...,\n",
      "         3.2492913e-05, -2.3804582e-06,  3.3204410e-06],\n",
      "       [ 1.6102606e-06,  4.2954689e-06,  4.7597287e-06, ...,\n",
      "         1.4818451e-05, -1.7321490e-05,  6.3076455e-05],\n",
      "       [-2.2872496e-06, -1.5031825e-05,  2.6579084e-05, ...,\n",
      "        -7.2143521e-05, -2.7100134e-05,  5.4231801e-05]], dtype=float32)>, <tf.Tensor: id=913657, shape=(6144,), dtype=float32, numpy=\n",
      "array([-6.1107899e-06,  4.2218326e-06,  1.6817305e-06, ...,\n",
      "        3.4864166e-05,  6.6581561e-07, -6.9656555e-05], dtype=float32)>, <tf.Tensor: id=913560, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.20523396e-06, -1.21305951e-04, -1.36922154e-06, ...,\n",
      "        -1.25698080e-06, -2.48494962e-06, -4.22880112e-06],\n",
      "       [-1.39322151e-06, -3.71700626e-05, -1.82585245e-06, ...,\n",
      "        -1.31887498e-06, -3.52278039e-06, -6.10313236e-06],\n",
      "       [-9.85267320e-07, -3.65199143e-04, -1.10783924e-06, ...,\n",
      "        -1.15879027e-06, -2.45282581e-06, -4.11068413e-06],\n",
      "       ...,\n",
      "       [ 9.22558866e-06,  1.48335332e-03,  8.80864991e-06, ...,\n",
      "         9.14845623e-06,  1.67594735e-05,  2.45339415e-05],\n",
      "       [-1.14518480e-06, -1.42012403e-04, -1.17553350e-06, ...,\n",
      "        -1.18883031e-06, -2.11916426e-06, -3.33112121e-06],\n",
      "       [ 1.25043925e-06, -1.18805394e-04,  1.57156319e-06, ...,\n",
      "         1.24428004e-06,  2.60066008e-06,  4.18850277e-06]], dtype=float32)>, <tf.Tensor: id=913552, shape=(110,), dtype=float32, numpy=\n",
      "array([ 9.46260752e-06,  1.50502089e-03,  9.01430212e-06,  8.32363206e-04,\n",
      "        5.33586508e-03,  4.39800897e-05,  6.58925390e-04,  1.20229388e-05,\n",
      "        1.18682492e-05,  1.34990050e-05,  4.36774513e-04, -4.87106154e-05,\n",
      "        3.42056213e-04,  4.81578281e-05,  1.97446843e-05, -1.33129559e-03,\n",
      "        1.27355830e-04, -1.65863370e-03,  1.35891431e-04, -3.19316634e-04,\n",
      "       -1.29065407e-03, -1.23057180e-05,  3.33511503e-04, -1.32660032e-04,\n",
      "        1.21394456e-04,  3.85285181e-04, -4.18730488e-04, -2.79628817e-04,\n",
      "       -4.82976466e-05, -3.47942550e-04,  3.06461181e-04,  3.12731390e-05,\n",
      "        1.79290255e-05,  4.40615149e-05,  2.11710721e-05, -8.07885895e-04,\n",
      "        3.79951496e-04,  4.25061909e-04, -3.12918448e-04, -1.26709652e-04,\n",
      "       -4.21548932e-04,  2.03283955e-04, -7.23933976e-04, -4.34026093e-04,\n",
      "       -2.91445263e-04,  3.56038690e-05, -3.96263815e-04, -3.89592024e-05,\n",
      "       -4.85235505e-04, -2.79376487e-04, -5.06515324e-04, -8.98956641e-05,\n",
      "       -1.96980458e-04,  6.16302714e-07, -2.26873017e-04, -5.48792959e-05,\n",
      "       -1.59711883e-04,  4.03352926e-04,  6.45074833e-05,  8.00965063e-05,\n",
      "        5.10937607e-05, -1.47555096e-04, -2.58853543e-04,  6.87819411e-05,\n",
      "        3.04137266e-05, -2.19331961e-03,  1.49453024e-03, -5.03966445e-03,\n",
      "       -1.61994493e-03,  3.00821313e-03, -3.98174627e-04,  6.14494202e-04,\n",
      "        2.19998020e-03, -1.05189881e-03, -7.78430782e-04,  1.74858829e-03,\n",
      "        1.85333181e-03, -1.94423180e-03,  5.40190237e-03, -5.80759160e-03,\n",
      "        4.90372721e-03, -4.04138045e-05,  1.36145006e-03, -5.89274615e-03,\n",
      "        2.29789899e-03, -1.19216274e-03,  5.02105104e-04,  6.89791166e-04,\n",
      "        6.02505577e-04, -2.00987794e-03,  3.44304892e-04,  1.37590832e-05,\n",
      "        9.91520756e-06,  1.68585266e-05,  1.26622735e-05,  1.01104888e-05,\n",
      "        1.23359177e-05,  1.16835199e-05,  1.05704939e-05,  1.08856684e-05,\n",
      "        1.19290517e-05,  1.15739194e-05,  1.14163759e-05,  1.06791322e-05,\n",
      "        1.95819357e-05,  1.04633009e-05,  1.15046041e-05,  9.37853019e-06,\n",
      "        1.71402880e-05,  2.50740195e-05], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc860>, <tf.Tensor: id=914016, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 2.50244557e-06, -2.02146134e-06, -2.58576756e-06, ...,\n",
      "         1.27404355e-05, -4.48473074e-05, -1.52936973e-05],\n",
      "       [-5.67024699e-07,  3.71498032e-07,  2.12241016e-08, ...,\n",
      "        -4.47506827e-05, -1.86877571e-06, -2.23127226e-05],\n",
      "       [ 1.83991318e-07,  1.04574873e-07, -1.19513913e-07, ...,\n",
      "         6.08786431e-05, -1.78010559e-05, -1.77620186e-05],\n",
      "       ...,\n",
      "       [-3.91859516e-07, -4.39847128e-07, -1.71540435e-07, ...,\n",
      "        -6.29241413e-07, -8.76280774e-06, -1.20971790e-05],\n",
      "       [-3.44813458e-08, -6.53901893e-08, -2.58598103e-08, ...,\n",
      "        -3.37648635e-05,  8.38114574e-05,  7.47408849e-05],\n",
      "       [ 7.07834602e-07,  9.88336069e-07, -9.27861521e-08, ...,\n",
      "         4.97772044e-06, -4.35799775e-05,  3.61911880e-05]], dtype=float32)>, <tf.Tensor: id=914017, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 1.2933432e-05,  1.5139508e-05,  1.9694446e-05, ...,\n",
      "         1.9297794e-04, -1.3741359e-05, -1.3631759e-05],\n",
      "       [ 5.4694809e-08, -7.0442002e-07, -1.3946160e-06, ...,\n",
      "         5.5327819e-05, -8.0857399e-06, -1.1455739e-04],\n",
      "       [ 1.3916186e-06, -3.5123932e-08,  1.9432189e-06, ...,\n",
      "        -6.4864071e-05,  1.5850710e-05, -1.9512405e-05],\n",
      "       ...,\n",
      "       [ 2.2555321e-07,  5.9740017e-07,  5.4702838e-07, ...,\n",
      "         3.1682910e-06,  1.5376632e-07, -5.4582642e-07],\n",
      "       [ 4.9571609e-06, -5.5489971e-07,  9.5254327e-06, ...,\n",
      "         2.9276777e-04, -3.3185122e-05,  1.2333940e-04],\n",
      "       [ 2.5215575e-06, -2.6293050e-05,  1.0471065e-05, ...,\n",
      "        -1.8553757e-04,  3.4348566e-05,  9.9800549e-05]], dtype=float32)>, <tf.Tensor: id=914018, shape=(6144,), dtype=float32, numpy=\n",
      "array([-4.1408690e-05,  1.5674710e-06, -4.7272274e-06, ...,\n",
      "        3.2409062e-06,  2.9428993e-04, -1.8693149e-04], dtype=float32)>, <tf.Tensor: id=913921, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.0805014e-06,  2.8684633e-05, -1.2507337e-06, ...,\n",
      "        -1.1150132e-06, -2.3790524e-06, -3.9908673e-06],\n",
      "       [-1.5866897e-06, -9.2847258e-05, -1.9461020e-06, ...,\n",
      "        -1.4986143e-06, -3.8587550e-06, -6.3708112e-06],\n",
      "       [-5.9697248e-07, -2.5169866e-06, -8.3285761e-07, ...,\n",
      "        -7.9802442e-07, -2.0735652e-06, -3.4980878e-06],\n",
      "       ...,\n",
      "       [ 7.6472334e-06, -5.7859637e-04,  7.5733260e-06, ...,\n",
      "         7.6470087e-06,  1.5141389e-05,  2.2134562e-05],\n",
      "       [-8.8647494e-07, -1.8955106e-06, -9.6360714e-07, ...,\n",
      "        -9.5190364e-07, -1.7770685e-06, -2.9048776e-06],\n",
      "       [ 1.2827090e-06,  6.2218832e-04,  1.6108055e-06, ...,\n",
      "         1.3056820e-06,  2.6652392e-06,  4.2360793e-06]], dtype=float32)>, <tf.Tensor: id=913913, shape=(110,), dtype=float32, numpy=\n",
      "array([ 7.78424783e-06, -5.87531831e-04,  7.70156930e-06,  3.76953860e-04,\n",
      "        9.05014016e-03,  4.17640658e-05, -5.63458714e-04,  1.06166735e-05,\n",
      "        1.02227450e-05,  1.23166174e-05, -8.87243426e-04,  1.92267889e-05,\n",
      "       -4.95814893e-04, -1.10704379e-04,  1.82033164e-05,  9.25242959e-04,\n",
      "       -3.24347755e-04, -2.57839146e-03,  1.26971194e-04, -2.13821069e-04,\n",
      "        3.93360242e-04, -8.75291589e-05, -1.80946281e-05,  3.65538814e-04,\n",
      "        2.34079736e-04, -4.27531864e-04,  3.38982572e-05,  7.20067910e-05,\n",
      "        5.59839827e-04, -6.60789257e-04, -5.59701584e-05,  2.69683569e-05,\n",
      "        1.62062406e-05,  4.11277215e-05,  1.93665343e-04,  2.89456628e-04,\n",
      "        7.19529169e-04, -1.53241112e-04,  3.09140130e-04,  1.67239938e-04,\n",
      "       -2.62455374e-04, -2.76872655e-04,  6.98522490e-05,  7.45339552e-04,\n",
      "        2.85504266e-05, -1.02654405e-04, -3.84682004e-04, -3.39624472e-04,\n",
      "        4.77665162e-04,  5.82794484e-04,  1.70077183e-04,  6.42880550e-05,\n",
      "        1.35013310e-04,  5.26120304e-04,  8.76957783e-04, -6.07975053e-05,\n",
      "        1.54273206e-04, -3.04860558e-04,  6.37328048e-05, -4.15747229e-04,\n",
      "        4.92309700e-05,  4.10064240e-06,  4.06612016e-05,  6.10701609e-05,\n",
      "        2.84018533e-05, -6.81440299e-03, -9.02246160e-04, -9.38469544e-03,\n",
      "        2.68262695e-03,  1.66068017e-03, -7.26573053e-04,  7.95732252e-04,\n",
      "        7.09589745e-04, -2.55892309e-03, -1.58535520e-04, -7.01018958e-04,\n",
      "        1.63053663e-03,  1.12997986e-04,  6.15015393e-03, -1.92368927e-04,\n",
      "        6.80171070e-04, -6.23234897e-04,  9.85422917e-03, -1.24315848e-03,\n",
      "       -8.64392146e-03, -2.57739192e-03,  1.88664044e-03,  1.55903446e-03,\n",
      "       -4.25540784e-04, -1.65056984e-03, -1.27758030e-04,  1.16086139e-05,\n",
      "        8.78586798e-06,  1.45442882e-05,  1.02578051e-05,  8.64574395e-06,\n",
      "        1.02696968e-05,  9.37616915e-06,  8.87142960e-06,  9.27330075e-06,\n",
      "        1.02373288e-05,  9.61939986e-06,  9.98070936e-06,  8.94466484e-06,\n",
      "        1.74428515e-05,  9.02674856e-06,  1.02417271e-05,  7.78196591e-06,\n",
      "        1.53948677e-05,  2.24791747e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdb0af98>, <tf.Tensor: id=914377, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-1.46268349e-06,  2.57434863e-06, -2.05182118e-06, ...,\n",
      "         3.54620238e-06, -1.64279136e-05,  8.69816267e-06],\n",
      "       [-4.61270044e-09,  4.11874019e-08, -3.18237227e-07, ...,\n",
      "         1.90720075e-05,  2.63819802e-05,  4.78309157e-05],\n",
      "       [-1.04647590e-07,  1.70509068e-07,  6.64734898e-08, ...,\n",
      "        -2.24860541e-05, -3.08194612e-05, -9.14961220e-06],\n",
      "       ...,\n",
      "       [ 4.00500085e-06,  1.57164413e-06, -1.52463269e-06, ...,\n",
      "         4.37260582e-07, -1.16806905e-05, -3.17886679e-06],\n",
      "       [-3.33452341e-08,  2.22013270e-08,  5.71657708e-08, ...,\n",
      "        -4.94509986e-05,  3.48934664e-05, -1.78327118e-05],\n",
      "       [-1.25389738e-06,  9.44991939e-07,  2.28591190e-08, ...,\n",
      "        -9.83698919e-05,  3.67464745e-05,  1.63597961e-05]], dtype=float32)>, <tf.Tensor: id=914378, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 2.4258516e-05,  2.8370072e-05,  2.7096565e-05, ...,\n",
      "         2.2825133e-04, -2.0156158e-06,  3.4811899e-05],\n",
      "       [ 7.3941145e-08,  3.0596470e-07, -1.4417044e-06, ...,\n",
      "        -9.7844269e-05, -3.3446988e-06, -7.4329284e-05],\n",
      "       [ 1.3538968e-06,  5.0141955e-07,  5.8635110e-06, ...,\n",
      "         3.1665980e-04, -1.7724751e-05,  7.7003540e-05],\n",
      "       ...,\n",
      "       [-6.5141641e-08, -5.7026256e-07,  4.8212559e-07, ...,\n",
      "         1.4624836e-05, -1.6452827e-06,  4.5709926e-06],\n",
      "       [ 5.9063495e-07,  9.1387456e-06,  2.8613311e-05, ...,\n",
      "        -1.2803650e-04, -2.5159484e-06, -8.9903158e-05],\n",
      "       [ 3.4877223e-05,  6.3278145e-05, -6.8446963e-05, ...,\n",
      "        -4.1885723e-04, -2.5447371e-05, -7.2034367e-05]], dtype=float32)>, <tf.Tensor: id=914379, shape=(6144,), dtype=float32, numpy=\n",
      "array([-6.9902344e-05,  3.4012223e-06, -9.0646836e-06, ...,\n",
      "        1.4797215e-05, -1.2949303e-04, -4.2328468e-04], dtype=float32)>, <tf.Tensor: id=914282, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.67431140e-07, -1.02128055e-04, -1.08834342e-06, ...,\n",
      "        -9.21242417e-07, -2.06494246e-06, -3.56477767e-06],\n",
      "       [-1.47233879e-06, -1.40338452e-04, -1.89128218e-06, ...,\n",
      "        -1.45323338e-06, -3.76612707e-06, -6.05920968e-06],\n",
      "       [-4.88287583e-07, -1.59963412e-04, -7.21582012e-07, ...,\n",
      "        -6.28242390e-07, -1.94345557e-06, -3.28740589e-06],\n",
      "       ...,\n",
      "       [ 6.09280687e-06,  3.36545578e-04,  6.29859187e-06, ...,\n",
      "         6.03004128e-06,  1.27169187e-05,  1.89961174e-05],\n",
      "       [-7.06104061e-07,  1.51610129e-05, -8.06830258e-07, ...,\n",
      "        -7.50006791e-07, -1.49253015e-06, -2.50675453e-06],\n",
      "       [ 1.35332380e-06,  3.32550320e-04,  1.63817595e-06, ...,\n",
      "         1.39016470e-06,  2.74752711e-06,  4.16939702e-06]], dtype=float32)>, <tf.Tensor: id=914274, shape=(110,), dtype=float32, numpy=\n",
      "array([ 6.18435934e-06,  3.39044404e-04,  6.38673464e-06,  1.02018844e-03,\n",
      "       -3.91091546e-03,  3.69460977e-05,  7.19764736e-04, -1.47659724e-04,\n",
      "       -1.48216757e-04,  1.04653582e-05,  3.74035706e-04, -1.01736012e-04,\n",
      "       -3.90129746e-04,  3.94721537e-05,  1.52551765e-05,  2.04773457e-03,\n",
      "        1.11871120e-03, -7.02193822e-04,  1.10954999e-04,  4.06864536e-04,\n",
      "        6.97035866e-04,  2.20246249e-04,  1.06205436e-04,  3.52152820e-05,\n",
      "        5.82453213e-05,  2.10695289e-04,  5.02205803e-04, -1.80199248e-04,\n",
      "       -2.49318720e-04, -3.44754371e-05, -1.15234885e-04,  2.07585508e-05,\n",
      "        1.37251127e-05,  3.56764831e-05, -2.84043083e-04,  7.45410434e-05,\n",
      "        9.13464319e-05,  4.25485341e-04,  8.07273609e-05,  6.43136911e-04,\n",
      "        4.47937171e-04, -1.59972245e-04,  2.71583180e-04, -6.88370783e-04,\n",
      "        5.61904153e-06,  3.37240577e-04, -2.67579395e-04,  1.39633106e-04,\n",
      "        3.62954481e-04,  4.42735036e-04, -1.83877666e-04, -9.77421150e-05,\n",
      "        2.18816291e-04,  2.67054238e-05,  5.91582735e-04, -7.24631318e-05,\n",
      "        1.35379247e-04, -1.08873856e-03, -4.12568246e-04, -3.06654285e-04,\n",
      "        4.46769991e-05,  1.38045012e-04,  1.76506277e-04,  4.81062743e-05,\n",
      "        2.44208841e-05,  2.10135186e-04, -3.15748970e-04, -1.60549360e-03,\n",
      "        2.65334360e-03, -2.36056163e-03,  1.02979349e-04, -3.66383698e-03,\n",
      "       -8.61713372e-04,  5.26143098e-03, -1.94573193e-04, -1.18805758e-04,\n",
      "        2.49980553e-03,  2.43580411e-03,  5.43178478e-03,  3.02726100e-03,\n",
      "       -6.07105612e-04,  1.90657171e-04,  6.73151854e-03, -1.14115747e-03,\n",
      "       -1.26167312e-02, -3.63411219e-03, -1.68059531e-04, -4.27395629e-04,\n",
      "       -6.83289138e-04, -3.25247389e-03, -4.02747304e-04,  9.58633973e-06,\n",
      "        7.16532031e-06,  1.17356240e-05,  7.78689719e-06,  6.95851077e-06,\n",
      "        8.19750949e-06,  7.05876982e-06,  6.78353081e-06,  7.16078830e-06,\n",
      "        8.11109112e-06,  7.52911228e-06,  8.33345712e-06,  7.15877422e-06,\n",
      "        1.43716215e-05,  7.19790251e-06,  8.56916722e-06,  6.11800624e-06,\n",
      "        1.28984057e-05,  1.92546140e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc7f0>, <tf.Tensor: id=914738, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 1.34687966e-07,  7.25794450e-07, -2.42026147e-07, ...,\n",
      "        -7.98095280e-05,  9.08625211e-08,  2.74118662e-07],\n",
      "       [-3.61160744e-07, -1.13014011e-07,  1.20443360e-07, ...,\n",
      "        -6.38643469e-05,  4.26382321e-05,  2.28522040e-05],\n",
      "       [ 1.17166685e-07,  1.59175030e-07, -1.86326616e-07, ...,\n",
      "        -4.08158267e-05, -1.32106306e-05, -2.06652439e-05],\n",
      "       ...,\n",
      "       [ 2.51330988e-07, -5.30728016e-07,  3.91810659e-07, ...,\n",
      "         4.11663086e-06, -6.97243104e-06, -3.46883826e-06],\n",
      "       [-2.40947998e-07,  2.65054538e-08,  1.43395724e-07, ...,\n",
      "        -7.12221954e-05,  6.07333677e-05, -2.50643006e-05],\n",
      "       [ 4.34240405e-07,  8.58868475e-07, -1.01771306e-07, ...,\n",
      "         2.11329261e-05, -4.52491895e-06,  9.03435830e-06]], dtype=float32)>, <tf.Tensor: id=914739, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 5.8845221e-06,  2.2285044e-06, -6.5902458e-07, ...,\n",
      "        -3.1088493e-05,  2.5913778e-05, -1.4257470e-05],\n",
      "       [-8.5409610e-08, -9.1308755e-07,  1.6580705e-06, ...,\n",
      "        -3.6155307e-04,  5.4615492e-05, -5.4408541e-05],\n",
      "       [ 4.5230425e-07,  1.5189786e-06,  1.8983574e-06, ...,\n",
      "         4.3875448e-04, -7.0082206e-05,  5.5803077e-05],\n",
      "       ...,\n",
      "       [ 3.6199600e-07,  2.1914190e-07, -2.6136982e-07, ...,\n",
      "         1.8670562e-05, -1.3087534e-06,  4.2003153e-06],\n",
      "       [ 1.5138942e-05,  2.1464459e-05,  3.5360565e-05, ...,\n",
      "        -4.0982754e-04,  2.1894431e-05, -4.4481436e-05],\n",
      "       [ 1.1164990e-05,  1.5183543e-05, -3.3945042e-05, ...,\n",
      "        -4.4949842e-04, -4.0735507e-05,  4.1893589e-05]], dtype=float32)>, <tf.Tensor: id=914740, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 1.2245576e-05, -2.5386710e-06, -4.2016345e-06, ...,\n",
      "        1.9007168e-05, -4.1514289e-04, -4.5578196e-04], dtype=float32)>, <tf.Tensor: id=914643, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.0602693e-06, -1.9953797e-05, -1.2333621e-06, ...,\n",
      "        -1.0973608e-06, -2.3718189e-06, -4.0298669e-06],\n",
      "       [-1.5658093e-06, -2.6182521e-05, -1.9378981e-06, ...,\n",
      "        -1.5018285e-06, -3.9670786e-06, -6.3512175e-06],\n",
      "       [-6.9748768e-07, -4.7823804e-04, -9.2386040e-07, ...,\n",
      "        -8.6800077e-07, -2.3161106e-06, -3.8125281e-06],\n",
      "       ...,\n",
      "       [ 7.6023098e-06, -7.3258986e-04,  7.3733408e-06, ...,\n",
      "         7.5468370e-06,  1.4956434e-05,  2.2242977e-05],\n",
      "       [-8.2861027e-07,  9.0179419e-06, -9.2748752e-07, ...,\n",
      "        -8.8767149e-07, -1.7297224e-06, -2.9065022e-06],\n",
      "       [ 1.3727579e-06,  1.5536214e-04,  1.6427446e-06, ...,\n",
      "         1.3994991e-06,  2.8689747e-06,  4.4132958e-06]], dtype=float32)>, <tf.Tensor: id=914635, shape=(110,), dtype=float32, numpy=\n",
      "array([ 7.72823387e-06, -7.50585459e-04,  7.48927550e-06, -2.08340003e-04,\n",
      "       -4.65153856e-03, -1.15685696e-04, -7.56745925e-04,  1.06237721e-05,\n",
      "        9.96865856e-06,  1.17251338e-05, -8.25718686e-04, -2.02182797e-04,\n",
      "        1.41660246e-04,  4.72542015e-05,  1.79264734e-05, -2.60161748e-03,\n",
      "       -1.98632828e-03, -1.03544514e-03,  1.19864737e-04, -1.90466948e-04,\n",
      "        3.66636930e-04, -1.87463258e-04, -1.73648164e-04,  3.84002458e-04,\n",
      "       -8.10414858e-05,  1.02827122e-04,  5.58677013e-04,  1.46334423e-05,\n",
      "        3.23982851e-04,  3.12948396e-04,  1.36150175e-05,  2.62104240e-05,\n",
      "       -2.96173006e-04,  4.26068800e-05,  4.20772340e-05,  5.01135131e-04,\n",
      "        2.96209590e-04, -2.63314490e-04,  2.72324949e-04, -2.46720621e-04,\n",
      "       -5.89359493e-04, -1.51262546e-04,  1.80205825e-04,  1.15450588e-03,\n",
      "        1.84929362e-04, -1.08073713e-04,  1.32149202e-04,  2.01147617e-04,\n",
      "        1.15120696e-04,  4.03178856e-04, -4.38539369e-04,  6.27674963e-05,\n",
      "        1.03733881e-04,  1.31615932e-04,  1.28859677e-03, -5.30688325e-04,\n",
      "       -3.18132865e-04,  4.16849740e-04,  6.53911993e-05,  4.61305681e-05,\n",
      "        4.51762098e-05, -8.92750995e-06,  3.60778213e-05,  6.03104418e-05,\n",
      "       -4.40644566e-04,  6.43306132e-03, -1.32509705e-03, -2.37803120e-04,\n",
      "       -2.01031170e-03, -5.67265160e-05,  6.31813426e-03,  6.10332063e-04,\n",
      "       -4.17617802e-03,  6.59499411e-03, -8.52739322e-04, -5.19341382e-04,\n",
      "       -2.12980900e-03,  2.53021158e-03,  2.40236288e-03,  2.55853380e-03,\n",
      "       -1.95400161e-03, -5.26407559e-04, -7.47784739e-04,  1.78378494e-03,\n",
      "       -5.09791728e-03, -9.58358520e-04, -1.63374818e-04, -8.60412547e-05,\n",
      "       -7.34038040e-05,  7.05106824e-04, -1.76791145e-04,  1.13872065e-05,\n",
      "        8.91311174e-06,  1.45436552e-05,  9.97765255e-06, -1.47763625e-04,\n",
      "        1.01327978e-05,  9.31867999e-06,  8.71210068e-06,  9.12313135e-06,\n",
      "        1.01981577e-05,  9.27266319e-06,  1.01610503e-05,  8.86194721e-06,\n",
      "        1.65103356e-05,  8.75831574e-06,  1.06385924e-05,  7.67016900e-06,\n",
      "        1.51937274e-05,  2.25816548e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc9e8>, <tf.Tensor: id=915099, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-5.4787752e-06, -4.4588555e-06,  5.9153291e-07, ...,\n",
      "        -2.1905607e-05,  3.7003803e-05,  5.8949456e-05],\n",
      "       [ 2.5273934e-07,  6.4593578e-08, -1.8973087e-07, ...,\n",
      "        -4.7742767e-05,  2.8166232e-05,  5.6071644e-06],\n",
      "       [-1.5787675e-08, -2.6113472e-07,  2.1599278e-07, ...,\n",
      "        -1.6547914e-05,  2.5078791e-05, -1.8603678e-05],\n",
      "       ...,\n",
      "       [-3.0826072e-08, -8.1904221e-07,  5.6097735e-09, ...,\n",
      "         2.0568623e-06,  1.9434696e-05,  8.1631852e-06],\n",
      "       [ 2.3878687e-08, -8.2825487e-08,  4.2739810e-08, ...,\n",
      "        -1.9442234e-05,  5.1288283e-05,  4.5507139e-05],\n",
      "       [ 2.5625530e-07,  7.2679268e-07,  7.8145385e-10, ...,\n",
      "        -9.7150414e-06, -2.9271623e-05, -3.6954647e-05]], dtype=float32)>, <tf.Tensor: id=915100, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 5.2560210e-07,  5.5406649e-06,  1.0865474e-05, ...,\n",
      "         9.9651748e-05, -7.5990765e-06,  1.1460646e-04],\n",
      "       [ 1.3104585e-06, -6.1591970e-07, -2.1253095e-06, ...,\n",
      "        -1.1988535e-05, -2.3083281e-05, -5.6553596e-05],\n",
      "       [ 4.4842707e-09,  1.5108437e-06,  9.2631166e-07, ...,\n",
      "         1.2836463e-04, -1.9725680e-05,  5.7014833e-05],\n",
      "       ...,\n",
      "       [ 6.7785550e-07,  6.3634610e-07,  4.7751928e-07, ...,\n",
      "        -5.3760455e-06,  1.8406902e-06, -1.2281224e-06],\n",
      "       [ 7.7215409e-06,  3.8185576e-06, -1.1758857e-06, ...,\n",
      "         2.6056252e-04, -4.7625315e-05,  1.2688780e-04],\n",
      "       [-2.1442968e-06, -3.0190722e-05,  1.1633503e-05, ...,\n",
      "         1.0164183e-03, -1.5344410e-04,  3.0127293e-04]], dtype=float32)>, <tf.Tensor: id=915101, shape=(6144,), dtype=float32, numpy=\n",
      "array([-1.7893972e-05, -2.9043999e-06,  5.8102785e-07, ...,\n",
      "       -5.4586526e-06,  2.6479538e-04,  1.0287758e-03], dtype=float32)>, <tf.Tensor: id=915004, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.4888543e-07, -9.2767390e-05, -1.1799150e-06, ...,\n",
      "        -1.0417378e-06, -2.2887152e-06, -3.8891867e-06],\n",
      "       [-1.5573951e-06,  1.5166641e-05, -1.9923675e-06, ...,\n",
      "        -1.5674333e-06, -4.0979216e-06, -6.5506056e-06],\n",
      "       [-5.8555793e-07, -2.2032906e-04, -8.8212801e-07, ...,\n",
      "        -7.9813208e-07, -2.2920503e-06, -3.8302560e-06],\n",
      "       ...,\n",
      "       [ 7.1540221e-06,  7.9705170e-04,  7.1952700e-06, ...,\n",
      "         7.2885905e-06,  1.4976833e-05,  2.1821990e-05],\n",
      "       [-8.4634814e-07, -9.2058552e-05, -9.3619803e-07, ...,\n",
      "        -9.1908805e-07, -1.7692578e-06, -2.8999750e-06],\n",
      "       [ 1.6404606e-06, -3.1186501e-04,  1.8825381e-06, ...,\n",
      "         1.6781685e-06,  3.3508222e-06,  5.0360513e-06]], dtype=float32)>, <tf.Tensor: id=914996, shape=(110,), dtype=float32, numpy=\n",
      "array([ 7.27576344e-06,  8.10305588e-04,  7.31130922e-06, -3.23316286e-04,\n",
      "       -3.70942405e-03,  4.10812427e-05, -4.75088891e-05,  1.04706724e-05,\n",
      "        9.57749944e-06,  1.17051704e-05, -3.38210637e-04, -1.26081140e-04,\n",
      "       -1.99438509e-05,  4.72884276e-05,  1.70333551e-05, -6.96602860e-04,\n",
      "        1.10772718e-03, -1.30817352e-03,  1.15532377e-04,  1.79491690e-04,\n",
      "       -2.25324882e-04,  4.88916296e-04, -2.43913673e-04,  1.51384549e-04,\n",
      "        3.74073948e-04,  2.21297872e-04,  1.17927790e-04, -2.43006652e-04,\n",
      "        6.27524394e-04,  3.27264599e-04,  1.58874827e-04,  2.53316539e-05,\n",
      "        1.63839941e-05,  4.22626399e-05, -2.52337806e-04,  7.31117616e-05,\n",
      "        8.59606589e-05,  8.80925581e-05,  1.04158738e-04,  6.41225430e-04,\n",
      "        2.03517659e-04,  3.25234229e-04, -4.77966067e-04,  9.98437026e-05,\n",
      "       -1.23960432e-04,  2.10568949e-04,  1.53828543e-04,  1.26725245e-05,\n",
      "        2.50373705e-04,  4.21015066e-05, -3.79354140e-04, -9.29049711e-05,\n",
      "       -8.89256917e-05, -8.30557256e-04,  6.55010372e-05,  1.00413527e-04,\n",
      "       -4.94316510e-06,  2.55620573e-04,  6.49345529e-05,  1.96735069e-04,\n",
      "       -1.09260356e-04, -1.05634554e-05,  3.79035118e-05,  5.63943540e-05,\n",
      "        2.88734045e-05,  5.30619826e-03, -3.81342164e-04,  4.95431898e-03,\n",
      "        1.32319843e-03, -7.38446787e-03,  6.25828700e-03, -2.44988548e-03,\n",
      "       -4.14326845e-04,  9.96329938e-04, -4.19858407e-05, -2.13508634e-03,\n",
      "       -8.89941468e-04,  4.16737515e-03, -8.18355102e-03, -5.81154460e-03,\n",
      "       -3.25073069e-03,  1.80875737e-04,  8.02998431e-04, -1.22142257e-03,\n",
      "        7.03228079e-03, -1.05157006e-03,  4.59292205e-04, -3.96516087e-04,\n",
      "        7.68980884e-04,  2.82444339e-03, -2.72910074e-05,  1.12131793e-05,\n",
      "        8.71571774e-06,  1.39215299e-05,  9.54593088e-06,  8.29033070e-06,\n",
      "        9.82278107e-06,  8.81007327e-06,  8.28811426e-06,  8.59787451e-06,\n",
      "        9.69505800e-06,  8.93870401e-06,  9.74835802e-06,  8.63196419e-06,\n",
      "        1.64054891e-05,  8.55966846e-06,  1.02763288e-05,  7.41002532e-06,\n",
      "        1.52214352e-05,  2.21628361e-05], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdb0a320>, <tf.Tensor: id=915460, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 3.9344209e-06, -4.9483532e-07, -2.4209007e-06, ...,\n",
      "        -7.7528552e-05,  6.1206578e-05,  1.2646603e-05],\n",
      "       [ 5.7941980e-07,  5.1684691e-07, -4.9253430e-08, ...,\n",
      "         1.3666038e-05, -2.2019087e-05,  5.7831044e-06],\n",
      "       [-5.1707451e-07, -1.4204726e-07,  1.4884253e-07, ...,\n",
      "         9.1855909e-06,  4.6392259e-05,  7.1696568e-06],\n",
      "       ...,\n",
      "       [ 2.0476911e-07, -3.7411226e-08,  8.3456064e-07, ...,\n",
      "        -1.7432087e-05,  2.8624647e-05,  9.7748889e-06],\n",
      "       [ 2.4488115e-08, -9.1220777e-08,  1.5327403e-08, ...,\n",
      "         2.0874262e-05, -3.5277815e-05,  8.5237498e-06],\n",
      "       [-1.9883289e-06,  2.0096505e-07,  1.2107103e-06, ...,\n",
      "         3.1741380e-05, -3.8089918e-05, -5.7156889e-05]], dtype=float32)>, <tf.Tensor: id=915461, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 1.6539412e-05, -6.1931962e-07,  8.4177209e-06, ...,\n",
      "         4.1739346e-04, -7.7061937e-05,  2.7276150e-05],\n",
      "       [-1.4586013e-06, -5.3472711e-08, -2.6238724e-06, ...,\n",
      "         4.9110118e-04, -9.1788999e-05,  6.4395950e-05],\n",
      "       [ 5.8216261e-07, -6.1223759e-08,  2.6956234e-06, ...,\n",
      "        -1.6968419e-04, -1.1215248e-05,  2.2429391e-05],\n",
      "       ...,\n",
      "       [ 4.1005413e-07,  3.8305373e-07,  1.3303036e-06, ...,\n",
      "        -5.0727449e-05,  7.8135754e-06, -1.0742904e-05],\n",
      "       [-4.6227033e-06, -1.4821621e-05, -1.0034076e-05, ...,\n",
      "         3.3131358e-04, -1.0041853e-05,  1.0597790e-04],\n",
      "       [-3.5819150e-05, -6.8406189e-05,  5.2208226e-05, ...,\n",
      "         5.0738268e-04, -7.6429620e-05,  1.3901180e-04]], dtype=float32)>, <tf.Tensor: id=915462, shape=(6144,), dtype=float32, numpy=\n",
      "array([-7.9689999e-06,  3.2967555e-06, -2.2239838e-06, ...,\n",
      "       -5.1515199e-05,  3.3604319e-04,  5.1531952e-04], dtype=float32)>, <tf.Tensor: id=915365, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.7441750e-07, -9.3902847e-05, -1.2213337e-06, ...,\n",
      "        -1.0708543e-06, -2.3925968e-06, -4.0607601e-06],\n",
      "       [-1.6717037e-06, -1.3574306e-04, -2.1228509e-06, ...,\n",
      "        -1.6610311e-06, -4.3720324e-06, -6.9619018e-06],\n",
      "       [-5.7003842e-07, -2.9044738e-04, -8.9373702e-07, ...,\n",
      "        -7.8703977e-07, -2.4058450e-06, -4.0129830e-06],\n",
      "       ...,\n",
      "       [ 7.2775260e-06,  3.9087026e-04,  7.4694553e-06, ...,\n",
      "         7.5168309e-06,  1.5349793e-05,  2.2649034e-05],\n",
      "       [-8.7107753e-07,  2.7034912e-06, -9.9090721e-07, ...,\n",
      "        -9.6257077e-07, -1.8466712e-06, -3.0814940e-06],\n",
      "       [ 1.7885268e-06, -1.0294623e-04,  2.0417651e-06, ...,\n",
      "         1.8498429e-06,  3.6316599e-06,  5.4681232e-06]], dtype=float32)>, <tf.Tensor: id=915357, shape=(110,), dtype=float32, numpy=\n",
      "array([ 7.41298345e-06,  3.89151654e-04,  7.59997283e-06,  6.84681407e-04,\n",
      "        8.11701268e-03,  4.41432967e-05, -3.14473145e-05,  1.10225310e-05,\n",
      "        9.78519711e-06,  1.20140567e-05,  9.84185492e-04,  6.29866146e-04,\n",
      "       -1.44925521e-04,  4.82358628e-05,  1.74870602e-05,  9.87505191e-04,\n",
      "       -3.01637279e-04, -1.54340034e-03,  1.16247327e-04,  2.25021387e-04,\n",
      "        2.91509292e-04,  2.98494269e-04, -2.92764980e-06,  8.48836207e-05,\n",
      "        2.78303603e-04,  1.40269796e-04,  3.23409273e-04, -2.80371489e-04,\n",
      "        4.06250008e-04, -5.53782098e-04,  1.77732800e-04,  2.62241156e-05,\n",
      "        1.73008375e-05,  4.24281534e-05,  2.73555808e-04,  3.42749117e-04,\n",
      "        7.70901388e-05,  2.96326878e-04, -3.60680075e-04,  4.41104115e-04,\n",
      "        2.30204256e-04, -3.03714041e-04,  4.73753695e-04,  1.57720270e-03,\n",
      "        4.08354535e-05,  6.44844549e-05,  3.44345026e-05,  4.57360700e-04,\n",
      "        7.07630359e-04, -1.13833579e-04,  1.65590653e-04, -9.05855195e-05,\n",
      "        7.39130337e-05,  8.88928480e-06,  1.54969364e-03, -4.91956453e-05,\n",
      "        1.51543471e-04, -2.96303217e-04,  6.58745339e-05,  4.41682205e-05,\n",
      "        4.86043282e-05, -1.64540834e-04, -1.10448469e-04,  5.61132074e-05,\n",
      "        3.05982394e-05, -1.40999036e-03,  1.19958236e-03,  3.67476675e-03,\n",
      "        2.69992248e-04, -9.34615918e-03,  1.56276161e-03, -6.94418792e-04,\n",
      "       -6.44814735e-03, -6.85247220e-03,  7.29485997e-04,  9.09074151e-05,\n",
      "       -1.39006274e-03, -3.03893117e-03, -8.85193702e-03,  7.79005094e-03,\n",
      "       -7.53572769e-03,  3.88200075e-04, -2.88732420e-03,  1.88132821e-04,\n",
      "        1.22038284e-02,  1.51155738e-03, -1.47147768e-03, -8.62794986e-04,\n",
      "        9.05593159e-04,  2.42899684e-03,  3.93100199e-04,  1.14757859e-05,\n",
      "        9.11306233e-06,  1.40533448e-05,  9.79236484e-06,  8.43895759e-06,\n",
      "        1.00953730e-05,  9.16617319e-06,  8.64165213e-06,  8.79952222e-06,\n",
      "        1.00732768e-05,  9.06244259e-06,  9.99035092e-06,  8.80966581e-06,\n",
      "        1.71296051e-05,  8.74504985e-06,  1.06964735e-05,  7.65452478e-06,\n",
      "        1.56188744e-05,  2.30280621e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc1d0>, <tf.Tensor: id=915821, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 4.55428835e-06,  2.41125946e-07, -5.52687709e-07, ...,\n",
      "         4.08560163e-06, -9.30998795e-06,  3.49978727e-05],\n",
      "       [ 7.17144388e-08,  2.30917081e-08, -2.50557015e-08, ...,\n",
      "        -7.42128759e-05,  1.32756777e-05, -4.40122058e-05],\n",
      "       [-3.03665587e-07, -1.06679813e-07, -5.58086519e-08, ...,\n",
      "         4.23876591e-05,  5.04710370e-05,  6.43685999e-05],\n",
      "       ...,\n",
      "       [ 2.34807530e-07,  7.83058567e-07, -2.37365896e-07, ...,\n",
      "        -8.41176370e-06,  1.65715846e-05,  6.04964680e-06],\n",
      "       [-2.06088998e-07,  4.86097314e-08,  2.20918693e-07, ...,\n",
      "        -1.25359875e-05, -2.92403420e-05,  3.31334340e-05],\n",
      "       [-1.36511051e-06, -1.43332130e-07,  8.27178724e-07, ...,\n",
      "         4.63875767e-05, -1.57475624e-05,  2.35225525e-05]], dtype=float32)>, <tf.Tensor: id=915822, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 1.33592921e-05,  7.07324148e-07, -4.66638585e-06, ...,\n",
      "         4.22861020e-04, -5.06719807e-05,  8.82959503e-05],\n",
      "       [-6.67461222e-07,  1.87978884e-07,  4.11982342e-07, ...,\n",
      "         9.14472621e-05,  1.57790919e-06, -1.21140736e-04],\n",
      "       [-1.87973592e-06,  5.37978124e-08, -6.11719361e-06, ...,\n",
      "        -5.19402965e-04,  5.57546882e-05, -1.19686360e-04],\n",
      "       ...,\n",
      "       [ 1.02609715e-06,  1.78254334e-06,  6.77935702e-07, ...,\n",
      "        -2.77618074e-05,  2.89903505e-06, -6.39853533e-06],\n",
      "       [-1.02145423e-05, -2.28473200e-05, -2.73135520e-05, ...,\n",
      "         1.43095400e-04,  7.36568882e-06,  4.39485448e-05],\n",
      "       [-2.00940831e-05, -3.61463099e-05,  4.70814630e-05, ...,\n",
      "         2.04896278e-05,  6.93505499e-05,  4.06410327e-05]], dtype=float32)>, <tf.Tensor: id=915823, shape=(6144,), dtype=float32, numpy=\n",
      "array([-6.8901954e-06, -3.4422071e-06,  9.0232288e-06, ...,\n",
      "       -2.8206992e-05,  1.4480944e-04,  2.3974939e-05], dtype=float32)>, <tf.Tensor: id=915726, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.5338845e-07, -5.4512919e-05, -1.1956417e-06, ...,\n",
      "        -1.0389995e-06, -2.3693269e-06, -3.9071538e-06],\n",
      "       [-1.5796336e-06, -5.7147005e-05, -2.0653388e-06, ...,\n",
      "        -1.6084607e-06, -4.2349261e-06, -6.6015673e-06],\n",
      "       [-5.8971193e-07, -1.7011660e-04, -8.9425714e-07, ...,\n",
      "        -7.5370656e-07, -2.4244168e-06, -3.8506764e-06],\n",
      "       ...,\n",
      "       [ 6.3932666e-06,  7.7147043e-04,  6.8742129e-06, ...,\n",
      "         6.5496365e-06,  1.4291744e-05,  2.0582156e-05],\n",
      "       [-7.2597641e-07, -9.3102441e-05, -8.7113705e-07, ...,\n",
      "        -8.1627911e-07, -1.6536318e-06, -2.7512397e-06],\n",
      "       [ 1.5032675e-06, -1.5080508e-04,  1.8805740e-06, ...,\n",
      "         1.6118705e-06,  3.2649368e-06,  4.8101915e-06]], dtype=float32)>, <tf.Tensor: id=915718, shape=(110,), dtype=float32, numpy=\n",
      "array([ 6.4987271e-06,  7.8647607e-04,  6.9822809e-06, -1.3135583e-04,\n",
      "        1.1835139e-02, -1.1285592e-04,  3.5312021e-04,  9.9598265e-06,\n",
      "        8.7515446e-06,  1.0953853e-05, -8.0023502e-04,  1.3765250e-04,\n",
      "        1.6787433e-04,  4.5291657e-05,  1.5457132e-05, -1.7360526e-03,\n",
      "       -3.3843631e-04,  2.1420461e-03,  1.0637682e-04,  4.1710286e-04,\n",
      "        2.0153084e-04,  1.3176312e-04, -8.4024992e-05, -6.3163979e-04,\n",
      "        3.5836617e-04, -9.1536822e-05,  2.1493809e-04,  1.8461153e-04,\n",
      "        3.3013918e-04, -2.2826521e-04,  1.4650353e-04,  2.2630666e-05,\n",
      "        1.5559974e-05,  3.8685812e-05,  4.3591962e-04, -4.1956003e-04,\n",
      "       -2.4973199e-04, -8.3309627e-05,  3.7122628e-04,  3.4371889e-04,\n",
      "       -7.7962870e-04,  1.4420101e-04, -4.0456286e-04,  6.0595106e-04,\n",
      "       -2.9308151e-04,  3.2454984e-05, -3.0724885e-04, -5.9102051e-04,\n",
      "        5.7378004e-04,  5.2227842e-05,  4.9114367e-04, -9.5252552e-05,\n",
      "       -6.2531460e-04, -2.5153806e-04,  4.5672423e-04, -2.1350238e-04,\n",
      "       -1.7253141e-04,  4.7742334e-04,  6.1435087e-05, -1.5786510e-04,\n",
      "        4.5310051e-05,  1.3299119e-04,  1.9955769e-04,  5.0081635e-05,\n",
      "        2.9593544e-05, -1.7132590e-03, -2.7619535e-05,  6.2799901e-03,\n",
      "        9.8855549e-04, -7.4590170e-03,  7.8396610e-04, -1.5594050e-03,\n",
      "       -1.7314893e-03, -1.2940740e-02,  6.6895026e-04,  5.1072211e-04,\n",
      "       -1.4817235e-03,  1.7326283e-03, -1.8797704e-03,  8.1326365e-03,\n",
      "       -1.0274787e-03,  5.4969458e-04, -1.0327891e-03, -6.7065139e-03,\n",
      "        2.3436851e-03,  2.4744789e-03, -2.0057391e-03, -2.4334018e-03,\n",
      "        1.5893237e-03,  2.1871096e-03,  1.7066451e-04,  1.0893731e-05,\n",
      "        7.9860874e-06,  1.2207459e-05,  8.4240028e-06,  7.6018550e-06,\n",
      "        8.8808802e-06,  7.8269286e-06,  7.2480252e-06,  7.6384222e-06,\n",
      "        8.7070612e-06,  7.9409256e-06,  8.9237747e-06,  7.7786863e-06,\n",
      "        1.5485986e-05,  7.7921695e-06,  9.5310888e-06,  6.6558273e-06,\n",
      "        1.4519767e-05,  2.0898140e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdb0a8d0>, <tf.Tensor: id=916182, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-1.8337226e-06,  1.9511679e-06, -1.3882191e-06, ...,\n",
      "        -1.1096937e-04,  6.0120339e-05, -7.2237262e-06],\n",
      "       [ 3.2541260e-07,  3.3349949e-08,  4.7410356e-09, ...,\n",
      "        -2.2966024e-05, -6.4361302e-06, -2.3715686e-06],\n",
      "       [-3.0478733e-07,  1.1722501e-07, -7.8900683e-08, ...,\n",
      "         3.9452414e-05,  2.3879553e-05,  3.1955071e-05],\n",
      "       ...,\n",
      "       [ 1.2295402e-06,  5.0598493e-07, -3.0080369e-07, ...,\n",
      "        -4.2390334e-06,  4.2914917e-06,  1.0019252e-05],\n",
      "       [ 1.1895745e-07,  1.7418964e-08,  5.9589517e-08, ...,\n",
      "        -1.4821764e-05,  1.2737440e-05, -1.7939656e-05],\n",
      "       [ 6.3538045e-07,  8.4534406e-07,  9.9985391e-07, ...,\n",
      "         2.1786173e-05, -6.9702091e-06,  4.8358584e-06]], dtype=float32)>, <tf.Tensor: id=916183, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-4.6676932e-06,  1.6598995e-05,  1.9157242e-05, ...,\n",
      "        -2.6281146e-04,  1.6151789e-05,  6.6363136e-06],\n",
      "       [-1.9954830e-06, -1.1676233e-06, -3.5283815e-06, ...,\n",
      "         3.9059052e-04, -3.6343205e-05,  7.0705937e-05],\n",
      "       [-1.3695806e-06, -1.5851770e-06, -3.3746162e-06, ...,\n",
      "         1.4633479e-04, -3.2409745e-05,  1.0018582e-04],\n",
      "       ...,\n",
      "       [ 2.6672024e-07,  1.0808134e-07, -9.1115857e-07, ...,\n",
      "         9.6553713e-06, -5.6011987e-07,  2.0288846e-06],\n",
      "       [ 5.7378884e-06,  1.1894563e-05,  1.3651828e-05, ...,\n",
      "        -1.3147361e-04, -7.3201832e-06,  9.4557436e-07],\n",
      "       [ 3.5790916e-05,  8.0019672e-05, -4.4276108e-05, ...,\n",
      "        -2.1160847e-04, -1.2576143e-05,  1.5667512e-05]], dtype=float32)>, <tf.Tensor: id=916184, shape=(6144,), dtype=float32, numpy=\n",
      "array([-3.2255230e-05,  1.0304830e-05,  9.9457748e-06, ...,\n",
      "        9.7721095e-06, -1.3338285e-04, -2.1538873e-04], dtype=float32)>, <tf.Tensor: id=916087, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.55199198e-07, -2.19921058e-05, -1.21201845e-06, ...,\n",
      "        -1.03513798e-06, -2.41209727e-06, -3.82906228e-06],\n",
      "       [-1.25944825e-06, -3.49813054e-05, -1.75644936e-06, ...,\n",
      "        -1.27083422e-06, -3.65118808e-06, -5.75576814e-06],\n",
      "       [-6.97513201e-07, -2.62172653e-05, -9.68174049e-07, ...,\n",
      "        -8.69815381e-07, -2.51712504e-06, -3.90365858e-06],\n",
      "       ...,\n",
      "       [ 6.33766831e-06,  7.50158681e-04,  6.82594327e-06, ...,\n",
      "         6.50494712e-06,  1.38779769e-05,  1.95772209e-05],\n",
      "       [-7.80388405e-07, -1.13578826e-04, -9.19457307e-07, ...,\n",
      "        -8.60027967e-07, -1.72608577e-06, -2.73248224e-06],\n",
      "       [ 1.19775245e-06,  8.23341252e-05,  1.53931478e-06, ...,\n",
      "         1.26313716e-06,  2.59222134e-06,  3.89482830e-06]], dtype=float32)>, <tf.Tensor: id=916079, shape=(110,), dtype=float32, numpy=\n",
      "array([ 6.44099009e-06,  7.64452852e-04,  6.93156289e-06, -7.62419659e-04,\n",
      "        3.40068201e-03,  4.14408460e-05,  2.15106556e-05,  9.69046141e-06,\n",
      "        8.86724138e-06,  1.01182914e-05, -3.94003378e-04,  6.26742622e-05,\n",
      "        1.87315673e-04,  4.31857079e-05,  1.44629703e-05, -1.05459651e-03,\n",
      "       -1.59720867e-03,  2.49102968e-03, -5.78142717e-05,  5.64421061e-04,\n",
      "       -7.99899339e-04, -1.77603633e-05, -2.24370888e-04,  1.68434010e-04,\n",
      "        2.23994939e-04, -8.34591629e-04,  4.18509124e-04,  8.99517472e-05,\n",
      "       -1.63458375e-04, -3.82894039e-04, -2.07422287e-04,  2.25057483e-05,\n",
      "        1.48383142e-05,  3.69115514e-05, -3.83292150e-04, -1.78213115e-03,\n",
      "        2.91863922e-04, -2.88499985e-04,  2.41177459e-05,  1.15832539e-04,\n",
      "        3.12492310e-04, -3.98302036e-05, -2.70843040e-04, -5.87216113e-04,\n",
      "        1.68193365e-04,  1.52626380e-04, -1.46109058e-04, -1.71565262e-05,\n",
      "        3.16536461e-04, -3.47569767e-06,  4.65351361e-04,  5.69507756e-05,\n",
      "        1.35621376e-04,  3.43121821e-04, -2.18151807e-04,  9.44496205e-05,\n",
      "        1.26073151e-04,  5.83154964e-04,  5.82536341e-05, -7.31768205e-06,\n",
      "        4.10397552e-05,  1.22251164e-04,  1.87005702e-04,  4.98967238e-05,\n",
      "        2.81951925e-05,  5.52659971e-04, -3.91459791e-04,  2.72438861e-03,\n",
      "       -4.98370314e-03,  1.40889734e-02, -3.95987416e-03, -2.20518233e-03,\n",
      "       -6.17256970e-04, -7.37463031e-03,  1.92098465e-04,  7.08939391e-04,\n",
      "        1.26649364e-04, -5.65391127e-03,  2.19706190e-03,  1.09897240e-03,\n",
      "       -1.00907497e-03,  1.75249108e-04, -3.67266452e-03, -9.36925469e-04,\n",
      "        6.76561706e-03,  4.50502150e-03, -8.52696889e-04, -3.64095392e-03,\n",
      "        1.24469341e-03, -1.17371697e-03, -1.32760790e-04,  1.09978509e-05,\n",
      "        7.74160253e-06,  1.18476455e-05,  8.71799421e-06,  7.56135478e-06,\n",
      "        8.86386351e-06,  8.05316813e-06,  7.37262235e-06,  7.61718593e-06,\n",
      "        8.51917503e-06,  7.86702094e-06,  8.47351293e-06,  7.72210842e-06,\n",
      "        1.46679831e-05,  7.85357497e-06,  9.10597646e-06,  6.60985370e-06,\n",
      "        1.40922002e-05,  1.98678208e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdb0a630>, <tf.Tensor: id=916543, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-2.09964492e-06,  7.32696662e-07, -5.57833857e-07, ...,\n",
      "        -9.17092038e-05,  2.73971036e-05,  2.42966553e-06],\n",
      "       [ 9.51694616e-08,  3.06399585e-07,  1.51117732e-08, ...,\n",
      "        -2.08212186e-05,  1.29326027e-05,  5.38762797e-05],\n",
      "       [-3.62604908e-07, -1.75959315e-07,  1.52389646e-07, ...,\n",
      "        -2.20134989e-05,  5.27444809e-06,  1.32354253e-05],\n",
      "       ...,\n",
      "       [-5.60238163e-07, -2.21203720e-07, -1.33237080e-07, ...,\n",
      "         5.23631024e-06, -8.34746697e-06, -5.65315077e-06],\n",
      "       [-3.48813280e-08, -2.32498500e-08,  1.84796377e-07, ...,\n",
      "         1.87594596e-05, -4.11156252e-05,  3.59234036e-06],\n",
      "       [ 1.22290282e-06, -6.42759119e-07,  5.67851771e-07, ...,\n",
      "         5.88053590e-05, -2.34704166e-05, -1.45449085e-05]], dtype=float32)>, <tf.Tensor: id=916544, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-1.4871498e-05, -2.8674008e-06,  8.1595053e-06, ...,\n",
      "        -2.8544044e-04,  7.8425161e-05, -4.0934769e-06],\n",
      "       [-1.4222139e-06, -3.7554755e-06, -1.1272334e-06, ...,\n",
      "        -9.7918390e-05,  1.6530588e-05,  9.7398690e-05],\n",
      "       [-1.8931993e-06, -3.5284239e-07, -2.1512881e-06, ...,\n",
      "         2.4201832e-04, -6.1112994e-05,  5.4044722e-05],\n",
      "       ...,\n",
      "       [ 7.0504711e-08,  1.3729772e-07,  7.6944082e-07, ...,\n",
      "         2.1629507e-05, -1.1863475e-06,  2.2287004e-06],\n",
      "       [-1.5878810e-06,  2.0833891e-06, -8.5080814e-07, ...,\n",
      "        -2.1359189e-04,  5.7212776e-05, -2.9617333e-05],\n",
      "       [ 2.5846972e-05,  5.1311014e-05, -1.0617412e-05, ...,\n",
      "        -4.8055733e-04,  7.9822668e-05, -2.0401545e-05]], dtype=float32)>, <tf.Tensor: id=916545, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 5.6307699e-06, -1.0579154e-06,  5.6819308e-06, ...,\n",
      "        2.1893309e-05, -2.1551119e-04, -4.8681325e-04], dtype=float32)>, <tf.Tensor: id=916448, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.75074079e-07,  4.61050804e-05, -1.11598547e-06, ...,\n",
      "        -9.51353513e-07, -2.29566876e-06, -3.58551551e-06],\n",
      "       [-1.35159394e-06, -6.53394454e-05, -1.79905885e-06, ...,\n",
      "        -1.36962331e-06, -3.88227454e-06, -5.71155397e-06],\n",
      "       [-4.55126951e-07,  9.09898590e-05, -7.57327598e-07, ...,\n",
      "        -6.03616400e-07, -2.14504462e-06, -3.27532416e-06],\n",
      "       ...,\n",
      "       [ 5.18076013e-06, -1.29299471e-03,  5.75523700e-06, ...,\n",
      "         5.33623506e-06,  1.23861055e-05,  1.72185555e-05],\n",
      "       [-6.24412223e-07,  1.00195000e-04, -7.81810741e-07, ...,\n",
      "        -6.95084736e-07, -1.52532857e-06, -2.42566853e-06],\n",
      "       [ 1.19247841e-06,  5.47072268e-04,  1.51596271e-06, ...,\n",
      "         1.26193913e-06,  2.67362338e-06,  3.84276291e-06]], dtype=float32)>, <tf.Tensor: id=916440, shape=(110,), dtype=float32, numpy=\n",
      "array([ 5.25615678e-06, -1.31969177e-03,  5.83473229e-06,  6.71954011e-04,\n",
      "       -4.57495637e-03,  3.70138550e-05, -4.61207615e-04,  8.26774885e-06,\n",
      "        7.25272230e-06,  8.75787737e-06, -5.75918762e-04,  6.93588925e-04,\n",
      "        7.87773461e-05,  3.80640304e-05,  1.28434103e-05, -2.23230850e-03,\n",
      "       -4.53282672e-04, -4.01345693e-04,  8.42290319e-05, -1.71118227e-04,\n",
      "        3.63051979e-04,  4.88864433e-04,  3.19132698e-04, -1.96826673e-04,\n",
      "        7.08923981e-05,  1.48905194e-04,  3.72309260e-05,  1.31297071e-04,\n",
      "        3.45937791e-04, -2.36363834e-04, -8.30995050e-05,  1.87617115e-05,\n",
      "        1.32020195e-05,  3.29712129e-05,  2.25533891e-04, -3.04710149e-04,\n",
      "        2.59972643e-04,  5.00037568e-04, -3.06780305e-04, -7.29817621e-05,\n",
      "       -1.39494237e-04, -5.48655480e-05, -2.48887576e-04, -1.60463256e-04,\n",
      "       -1.47396917e-04,  1.26483064e-04,  2.36331744e-05,  1.28656844e-04,\n",
      "        2.55030929e-04,  1.69600957e-04,  1.35620852e-04,  5.27158372e-05,\n",
      "       -3.27694463e-04, -5.91991644e-04,  1.70261337e-05, -6.25060857e-05,\n",
      "        1.19649259e-04,  4.90510138e-04,  5.34645951e-05,  1.46759790e-04,\n",
      "        3.73089169e-05, -4.22172379e-05,  5.31594196e-07,  4.10478460e-05,\n",
      "        2.55537998e-05,  1.31640967e-03,  7.76716479e-05,  4.50837985e-03,\n",
      "       -3.02080135e-03,  1.04905460e-02, -1.52527844e-03,  3.76514508e-05,\n",
      "       -3.91773973e-03,  4.72620537e-04,  2.97707826e-04,  5.26351156e-04,\n",
      "       -9.28030815e-04,  1.59410108e-03,  2.53601535e-03, -3.43189482e-03,\n",
      "        3.50592658e-03, -2.94367375e-04, -2.21164839e-04, -3.67094204e-03,\n",
      "        2.40178686e-03,  1.72876287e-03, -2.76026269e-03, -2.57449225e-04,\n",
      "       -1.61208620e-04, -1.71259861e-03, -1.01218908e-03,  8.92274693e-06,\n",
      "        6.70431882e-06,  1.00254656e-05,  6.80388257e-06,  6.35661308e-06,\n",
      "        7.20208573e-06,  6.41149518e-06,  6.09600738e-06,  6.32489491e-06,\n",
      "        7.10411678e-06,  6.46504395e-06,  7.35227513e-06,  6.38861911e-06,\n",
      "        1.28564225e-05,  6.49107051e-06,  7.95659707e-06,  5.41288637e-06,\n",
      "        1.25596252e-05,  1.74500219e-05], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc9e8>, <tf.Tensor: id=916904, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-6.9358712e-06, -4.4834796e-06,  1.6087159e-06, ...,\n",
      "        -9.5508958e-06, -4.6902609e-05, -5.0258775e-05],\n",
      "       [-6.2540948e-07, -3.1282300e-07,  2.9097089e-07, ...,\n",
      "        -1.1741785e-05, -1.7875271e-06,  3.1817744e-05],\n",
      "       [-9.8166913e-07, -3.9934869e-07, -9.3014691e-08, ...,\n",
      "         1.7001526e-05, -2.1415077e-05, -3.4060045e-05],\n",
      "       ...,\n",
      "       [ 1.4393236e-06, -8.5470468e-08, -2.5556642e-07, ...,\n",
      "        -6.1212204e-06, -2.4985102e-05, -2.8062135e-05],\n",
      "       [-2.5305718e-08,  4.2107306e-08,  9.7515702e-08, ...,\n",
      "         4.2639786e-05, -1.7833217e-05,  3.2730462e-05],\n",
      "       [-1.4031278e-06, -5.3073927e-07,  1.5729166e-06, ...,\n",
      "        -1.0002765e-04,  8.5415508e-05,  2.0788517e-05]], dtype=float32)>, <tf.Tensor: id=916905, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-1.9646044e-05, -2.8900893e-05, -2.6858557e-05, ...,\n",
      "        -4.8452266e-04,  5.6489600e-05, -1.9082456e-05],\n",
      "       [-1.0087288e-06, -6.4724611e-07, -2.2859349e-06, ...,\n",
      "        -6.5163033e-05,  3.0135116e-05,  5.1127510e-05],\n",
      "       [ 7.3924065e-07,  9.3079086e-07,  3.0286958e-07, ...,\n",
      "        -9.1016285e-05, -4.0971827e-07,  1.4340794e-06],\n",
      "       ...,\n",
      "       [ 2.6326944e-08, -3.8869638e-07, -4.0583410e-07, ...,\n",
      "         2.7228150e-05, -1.1681237e-06,  5.5988744e-06],\n",
      "       [ 3.5060884e-06, -1.0269750e-05,  2.4628546e-06, ...,\n",
      "        -1.4069199e-04,  1.1365533e-05,  1.4367981e-05],\n",
      "       [ 1.4243973e-05,  7.2709031e-06,  2.7133221e-05, ...,\n",
      "        -6.0120702e-04,  1.1320767e-04, -1.3587944e-04]], dtype=float32)>, <tf.Tensor: id=916906, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 5.8960111e-05,  3.9885317e-06,  8.7294404e-07, ...,\n",
      "        2.7580521e-05, -1.4175868e-04, -6.0860562e-04], dtype=float32)>, <tf.Tensor: id=916809, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.00525529e-06, -8.87956230e-06, -1.22151835e-06, ...,\n",
      "        -1.05880918e-06, -2.52050313e-06, -3.83402403e-06],\n",
      "       [-1.24948508e-06, -1.14146969e-04, -1.72722889e-06, ...,\n",
      "        -1.21296091e-06, -3.76308390e-06, -5.68609266e-06],\n",
      "       [-5.92921367e-07,  9.66517109e-05, -9.13936560e-07, ...,\n",
      "        -8.05745856e-07, -2.49190930e-06, -3.71479086e-06],\n",
      "       ...,\n",
      "       [ 6.94520349e-06, -1.62069075e-04,  7.11154507e-06, ...,\n",
      "         6.79572986e-06,  1.48291092e-05,  2.00785071e-05],\n",
      "       [-8.62586603e-07,  1.17888136e-04, -9.47434899e-07, ...,\n",
      "        -8.87616125e-07, -1.81657765e-06, -2.73867909e-06],\n",
      "       [ 1.29473801e-06,  2.67497642e-04,  1.58168882e-06, ...,\n",
      "         1.31143781e-06,  2.79337723e-06,  4.00109093e-06]], dtype=float32)>, <tf.Tensor: id=916801, shape=(110,), dtype=float32, numpy=\n",
      "array([ 7.07349272e-06, -1.60487733e-04,  7.23133689e-06, -3.05479451e-04,\n",
      "       -6.37458172e-03, -1.14644870e-04,  1.05713517e-03,  1.04502642e-05,\n",
      "        9.53370909e-06,  1.04808023e-05, -2.01929151e-03,  3.35091681e-05,\n",
      "       -1.53916510e-04,  4.38587704e-05, -1.41253448e-04, -3.10916873e-03,\n",
      "       -1.74925115e-03,  2.15168064e-03,  9.07418798e-05, -1.82000775e-04,\n",
      "       -1.18318899e-03,  5.64421411e-04,  1.48604595e-04, -4.10572451e-04,\n",
      "        1.00295474e-05,  2.03867938e-04,  1.84053919e-04,  3.46839573e-04,\n",
      "        1.99482442e-04, -3.45039618e-04,  7.19486343e-05,  2.36037286e-05,\n",
      "        1.53012934e-05,  3.78070545e-05,  7.59512477e-05,  1.96805151e-04,\n",
      "       -2.39419969e-04,  4.30427957e-04,  5.67721363e-05,  1.41580284e-04,\n",
      "       -6.20639039e-05, -2.51499150e-05, -1.91874176e-04, -3.18040838e-04,\n",
      "        1.94937893e-04, -1.44966016e-05, -3.42010317e-04,  1.98941256e-04,\n",
      "       -3.83607781e-04,  3.72329407e-04,  3.92962975e-05,  5.85108392e-05,\n",
      "        1.91806030e-04,  1.88991704e-04, -5.52280871e-05,  1.09229804e-04,\n",
      "        1.34770773e-04, -2.83130212e-04,  6.37425983e-05, -3.09100491e-04,\n",
      "        4.13614907e-05, -1.87990285e-04, -1.41967263e-04,  5.15788633e-05,\n",
      "        2.96206890e-05, -2.23163795e-03,  6.82010315e-04, -3.68650863e-03,\n",
      "       -1.49602338e-03,  4.15796461e-03, -2.05144403e-04,  2.71559088e-03,\n",
      "        3.68152186e-03,  6.34841761e-03,  7.07298284e-04,  1.04178779e-03,\n",
      "       -1.24147825e-03,  1.53781660e-03,  1.00445403e-02, -4.03786637e-03,\n",
      "        7.87507743e-04,  1.49048810e-05, -7.37358211e-03,  8.29065870e-03,\n",
      "       -1.10934377e-02,  1.13444915e-03,  4.59429633e-04, -1.65184424e-03,\n",
      "        1.95035231e-04,  2.21967255e-03, -3.96635805e-05,  1.13751994e-05,\n",
      "        8.15384738e-06,  1.25230217e-05,  9.07044614e-06,  8.19593788e-06,\n",
      "        9.43748819e-06,  8.64932827e-06,  8.03680086e-06,  8.05269701e-06,\n",
      "        8.84704696e-06,  8.25632651e-06,  8.78003539e-06, -1.47877028e-04,\n",
      "        1.55861362e-05,  8.46423245e-06,  9.28834379e-06,  6.91856849e-06,\n",
      "        1.50731985e-05,  2.03938471e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efe315710>, <tf.Tensor: id=917265, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-4.86474210e-06,  6.85742350e-07,  2.10983353e-06, ...,\n",
      "         8.34499951e-06,  2.46962099e-05,  1.90994069e-05],\n",
      "       [-3.23838520e-07,  1.72224603e-07, -1.04592466e-07, ...,\n",
      "        -4.10587199e-06, -4.76520727e-05,  1.54377449e-05],\n",
      "       [ 8.51346442e-08, -1.25806253e-07,  8.58344507e-08, ...,\n",
      "         1.60012514e-05,  2.16261287e-05, -4.16953435e-06],\n",
      "       ...,\n",
      "       [ 8.81937183e-07, -4.28025515e-07, -7.61066360e-07, ...,\n",
      "         1.41744454e-06, -1.51513095e-05, -9.31250725e-06],\n",
      "       [-1.52780885e-07, -6.58035262e-08,  1.32262812e-07, ...,\n",
      "         3.39660255e-05,  7.11465873e-07,  1.82936692e-05],\n",
      "       [-9.62396143e-07,  3.56101708e-07,  7.69922337e-07, ...,\n",
      "        -6.01641841e-05,  4.17256160e-05,  1.07733274e-06]], dtype=float32)>, <tf.Tensor: id=917266, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-1.7921880e-05, -6.0393854e-06, -6.4669875e-06, ...,\n",
      "        -7.3136040e-04,  6.6089313e-05, -2.5343432e-04],\n",
      "       [-2.7084325e-06, -3.2202611e-06, -3.3339147e-06, ...,\n",
      "        -1.3743233e-04,  6.7435026e-06,  7.2550618e-05],\n",
      "       [ 5.6621082e-07,  2.0848320e-06,  1.2863247e-06, ...,\n",
      "        -3.2119395e-05,  6.2317481e-06, -8.8364592e-05],\n",
      "       ...,\n",
      "       [-3.5521464e-07, -4.3206867e-07,  9.5199857e-07, ...,\n",
      "         9.7391603e-06, -2.7359640e-06,  5.3281056e-08],\n",
      "       [ 1.1820019e-05,  1.5090689e-05,  1.5111725e-05, ...,\n",
      "         2.4789749e-04, -3.8891623e-05,  8.6234271e-05],\n",
      "       [ 5.7847833e-06,  8.1829457e-06, -5.0168620e-07, ...,\n",
      "        -1.9265864e-05, -2.5275895e-05,  1.4998176e-05]], dtype=float32)>, <tf.Tensor: id=917267, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 2.3340370e-05,  9.4344159e-06, -3.7660632e-06, ...,\n",
      "        9.8937653e-06,  2.5135017e-04, -1.9684237e-05], dtype=float32)>, <tf.Tensor: id=917170, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.0645238e-06, -1.6041457e-05, -1.2428620e-06, ...,\n",
      "        -1.1242478e-06, -2.6329219e-06, -3.8912044e-06],\n",
      "       [-1.2624948e-06, -2.4066167e-04, -1.7475857e-06, ...,\n",
      "        -1.2054722e-06, -3.8633866e-06, -5.8004543e-06],\n",
      "       [-7.6198069e-07,  3.7509139e-04, -1.0163939e-06, ...,\n",
      "        -1.0000458e-06, -2.7269296e-06, -4.0044097e-06],\n",
      "       ...,\n",
      "       [ 7.5945190e-06, -6.9330743e-04,  7.5714424e-06, ...,\n",
      "         7.7172190e-06,  1.6167749e-05,  2.1373900e-05],\n",
      "       [-9.5346314e-07,  2.4826947e-04, -1.0177379e-06, ...,\n",
      "        -1.0125004e-06, -2.0046891e-06, -2.9535922e-06],\n",
      "       [ 1.2738938e-06,  5.9004378e-04,  1.6091942e-06, ...,\n",
      "         1.2300557e-06,  2.8859470e-06,  4.0167679e-06]], dtype=float32)>, <tf.Tensor: id=917162, shape=(110,), dtype=float32, numpy=\n",
      "array([ 7.73957254e-06, -7.14676105e-04,  7.70397037e-06,  4.08922293e-04,\n",
      "       -6.48870924e-03, -1.11769768e-04,  1.42715115e-04,  1.09369539e-05,\n",
      "        1.02401227e-05,  1.10461642e-05, -2.96021899e-05,  4.98792680e-04,\n",
      "        1.80956500e-04,  4.49220388e-05,  1.58383227e-05, -4.33792447e-04,\n",
      "        1.21205032e-03,  1.66114944e-03, -6.46024346e-05, -4.26642742e-04,\n",
      "        6.50971662e-04,  1.31037712e-04, -1.29272899e-04,  5.90491072e-06,\n",
      "       -2.70238495e-04,  5.79727581e-04, -4.01051017e-04, -5.42992493e-04,\n",
      "       -5.38699562e-04, -2.87331233e-04, -3.97251802e-04,  2.71143908e-05,\n",
      "        1.68069164e-05,  4.10311732e-05, -7.33795241e-05,  1.13394530e-03,\n",
      "        3.98205069e-04,  5.93195145e-04, -7.98032052e-05, -1.62710363e-04,\n",
      "        1.38848962e-04,  2.95959180e-04,  1.65123012e-04, -1.00002484e-03,\n",
      "        5.24069219e-05, -1.39767926e-05, -2.89922522e-04, -4.12669906e-04,\n",
      "       -8.45734467e-05,  4.06649400e-04, -2.56684172e-04, -9.61460464e-05,\n",
      "       -2.56168132e-04, -2.14375177e-04, -6.85182516e-04, -3.64289044e-05,\n",
      "       -1.76195055e-04,  3.76985525e-04,  6.44204993e-05,  1.80644693e-05,\n",
      "        4.21019686e-05,  1.30224013e-04,  1.77060123e-04,  5.32800223e-05,\n",
      "        3.02549197e-05, -5.93385007e-03, -1.37499045e-03, -4.44354257e-04,\n",
      "       -2.66494136e-03,  2.98117229e-04, -1.72359258e-04,  2.54237908e-03,\n",
      "        4.37992113e-03,  1.36480033e-02,  9.38491954e-04, -2.52555910e-04,\n",
      "       -7.82724645e-04, -5.86571987e-04,  4.38356493e-03, -6.21225359e-03,\n",
      "        4.24124207e-03,  3.36543133e-04, -4.13046405e-03,  2.21437751e-03,\n",
      "       -5.74941654e-03,  1.93813443e-03, -5.46155905e-04, -6.04744768e-04,\n",
      "        4.55920672e-04, -6.92410860e-04, -5.08009689e-04,  1.20309141e-05,\n",
      "        8.79690924e-06,  1.37984389e-05,  1.02242348e-05,  8.95759240e-06,\n",
      "        1.04781757e-05,  9.43802206e-06,  9.06015975e-06,  8.83594657e-06,\n",
      "        1.00975085e-05,  9.37807272e-06,  9.26163830e-06,  9.49652349e-06,\n",
      "        1.66299815e-05,  9.37089226e-06,  9.75641251e-06,  7.86488454e-06,\n",
      "        1.64469275e-05,  2.17248817e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc470>, <tf.Tensor: id=917626, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 5.5216951e-06,  4.5694619e-06, -3.3985796e-06, ...,\n",
      "         6.7090376e-05, -4.7487749e-05,  1.2049776e-05],\n",
      "       [ 3.4952785e-07, -1.4214709e-08, -4.6883720e-08, ...,\n",
      "        -1.4804392e-05, -7.6517499e-06, -1.2324640e-05],\n",
      "       [-5.7568991e-07, -2.6707468e-07,  8.0669061e-08, ...,\n",
      "         6.5487249e-05, -4.0863579e-05, -6.5829672e-06],\n",
      "       ...,\n",
      "       [ 2.3601331e-07, -1.6038339e-07,  5.7945812e-08, ...,\n",
      "         2.7615672e-06, -1.9512669e-05, -1.3018640e-05],\n",
      "       [-1.2112538e-07,  5.1080526e-08,  9.8615288e-08, ...,\n",
      "        -5.4511929e-05,  5.8075198e-05,  5.5402761e-05],\n",
      "       [-1.1779143e-06,  8.3900767e-07,  1.0735937e-06, ...,\n",
      "        -1.6202725e-05,  2.3376097e-05,  2.9303475e-05]], dtype=float32)>, <tf.Tensor: id=917627, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 3.7267684e-05,  1.3055523e-05,  4.6826081e-06, ...,\n",
      "         4.3269421e-04, -1.4116229e-05,  1.8368426e-04],\n",
      "       [-2.0476563e-07, -9.5604810e-07,  1.5203486e-06, ...,\n",
      "         2.0211352e-05, -1.0872731e-05,  1.7454313e-05],\n",
      "       [ 3.2257078e-06,  3.5054668e-06,  4.0335190e-06, ...,\n",
      "        -1.8091830e-04,  2.9302575e-05, -1.4741663e-05],\n",
      "       ...,\n",
      "       [-9.1867989e-08,  8.6160725e-08,  1.2124573e-07, ...,\n",
      "         1.1371225e-05, -2.5662607e-06,  1.8825713e-06],\n",
      "       [ 4.1881412e-06, -1.5116177e-05, -1.4653202e-05, ...,\n",
      "         7.1581558e-04, -6.0690400e-05,  2.4656591e-04],\n",
      "       [ 9.6050064e-07, -5.3553513e-06,  3.5825349e-05, ...,\n",
      "         2.7709264e-05,  4.4209766e-05,  1.9510630e-06]], dtype=float32)>, <tf.Tensor: id=917628, shape=(6144,), dtype=float32, numpy=\n",
      "array([-3.0562696e-05, -2.4939038e-06, -1.3565650e-05, ...,\n",
      "        1.1490743e-05,  7.2458270e-04,  2.7133727e-05], dtype=float32)>, <tf.Tensor: id=917531, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.1982190e-07,  1.0849682e-05, -1.1071513e-06, ...,\n",
      "        -9.8670762e-07, -2.3486259e-06,  5.9934200e-06],\n",
      "       [-1.5069643e-06, -7.4337389e-05, -1.9594929e-06, ...,\n",
      "        -1.4878178e-06, -4.3581467e-06,  2.7426879e-05],\n",
      "       [-5.7957311e-07,  1.5296934e-04, -8.7817102e-07, ...,\n",
      "        -7.9742870e-07, -2.5370155e-06, -2.0033269e-05],\n",
      "       ...,\n",
      "       [ 6.9848993e-06, -6.2404352e-04,  7.0044730e-06, ...,\n",
      "         7.0278638e-06,  1.5221365e-05, -1.3062889e-04],\n",
      "       [-7.8978019e-07,  1.2786576e-04, -8.9796964e-07, ...,\n",
      "        -8.4655096e-07, -1.7788464e-06,  8.1532189e-06],\n",
      "       [ 1.5051576e-06,  1.4033403e-04,  1.8080930e-06, ...,\n",
      "         1.5181390e-06,  3.2413977e-06,  5.3031421e-05]], dtype=float32)>, <tf.Tensor: id=917523, shape=(110,), dtype=float32, numpy=\n",
      "array([ 7.1229051e-06, -6.3211826e-04,  7.1296868e-06, -3.0525672e-04,\n",
      "        2.1749027e-03,  4.6354202e-05, -4.7632348e-04,  1.0640197e-05,\n",
      "        9.4236448e-06,  1.0781045e-05,  1.8428152e-03,  4.6289741e-04,\n",
      "        3.0134871e-04,  4.2941152e-05,  1.5626641e-05,  1.6083162e-03,\n",
      "        5.1970728e-04, -1.3766043e-03,  8.7443041e-05, -1.5423102e-06,\n",
      "       -4.0079300e-05, -2.0240888e-05, -1.5940174e-04,  1.5212066e-04,\n",
      "        1.4506017e-04,  3.2374333e-04,  4.1211644e-04,  8.1601218e-05,\n",
      "       -2.8262974e-04,  8.8924564e-05, -3.8192753e-04,  2.5274312e-05,\n",
      "        1.6231490e-05,  3.6765065e-05, -5.6334968e-05,  7.4413221e-04,\n",
      "        8.5547654e-05, -7.3050190e-04, -2.3366431e-04, -2.0707806e-04,\n",
      "        3.2340025e-04,  1.3626296e-04, -1.1436101e-04,  4.0067954e-04,\n",
      "       -9.9167381e-05, -2.3037472e-05, -8.5643456e-05, -5.7205663e-04,\n",
      "       -7.1725866e-04,  2.1487699e-04, -3.8869126e-04,  6.1707717e-05,\n",
      "        2.2967698e-04,  2.8437155e-04, -1.6038818e-04, -3.4166340e-05,\n",
      "        1.3294484e-04,  5.2638853e-04,  6.1664890e-05,  2.6645455e-05,\n",
      "        4.0916279e-05,  1.3634913e-04,  1.7160208e-04,  4.7986225e-05,\n",
      "        3.0201003e-05,  9.6778281e-04,  4.7556893e-04, -2.9898854e-03,\n",
      "       -1.5084052e-03, -9.6783619e-03, -9.3236368e-04,  3.0797531e-03,\n",
      "        4.5996569e-03,  4.4794837e-03,  4.0978284e-04,  1.1888703e-03,\n",
      "        2.5185928e-04, -5.3143071e-04,  1.1293071e-03, -1.0820261e-02,\n",
      "        3.2917392e-03,  5.8816524e-05,  2.1892299e-03,  8.0833659e-03,\n",
      "       -2.7973023e-03, -2.8546455e-03, -8.5148285e-04, -1.8460790e-03,\n",
      "       -8.8326931e-05,  1.1682119e-04, -4.5116048e-04,  1.1025901e-05,\n",
      "        8.4950689e-06,  1.2921799e-05,  9.0407848e-06,  8.0953996e-06,\n",
      "        9.6712956e-06,  8.2145843e-06,  8.2123033e-06,  8.2086081e-06,\n",
      "        9.4701090e-06,  8.4559324e-06,  8.9195864e-06,  8.7777935e-06,\n",
      "        1.6252859e-05,  8.7128001e-06,  9.4521165e-06,  7.1644481e-06,\n",
      "        1.5490319e-05, -1.3508744e-04], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdb0a0f0>, <tf.Tensor: id=917987, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-1.1212875e-06, -2.9336989e-06, -1.9780771e-06, ...,\n",
      "         3.1639865e-05,  2.5041596e-05,  2.0672804e-05],\n",
      "       [ 6.0785771e-07,  9.9346869e-07, -6.5490781e-08, ...,\n",
      "         4.4150180e-05, -5.0012681e-05, -2.7423199e-05],\n",
      "       [-6.5466753e-07,  2.1962903e-08,  1.8010438e-08, ...,\n",
      "         1.4433944e-06, -2.7093862e-05, -4.5343528e-05],\n",
      "       ...,\n",
      "       [ 4.5897741e-07, -3.2975458e-07, -1.3990092e-07, ...,\n",
      "         9.6198273e-06, -3.4927967e-07,  5.8628434e-06],\n",
      "       [-2.1157859e-07,  1.1971474e-07,  7.3743060e-08, ...,\n",
      "         1.2062389e-06,  2.9706549e-05,  5.8739835e-05],\n",
      "       [-4.3767656e-07, -5.6891531e-07,  9.3036647e-07, ...,\n",
      "        -1.6274837e-05,  1.0823232e-05, -1.6464870e-05]], dtype=float32)>, <tf.Tensor: id=917988, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 1.20453742e-05,  6.65843800e-06,  1.53122492e-05, ...,\n",
      "        -1.10106659e-04,  4.66655947e-05,  4.85948840e-05],\n",
      "       [-9.30266822e-07, -2.65645372e-06, -1.40785994e-06, ...,\n",
      "        -5.97001745e-05,  8.93034576e-06, -2.05331562e-05],\n",
      "       [ 1.36392885e-06,  6.89121919e-07, -4.40300084e-07, ...,\n",
      "         1.92194959e-04, -4.09507520e-05, -1.11706977e-05],\n",
      "       ...,\n",
      "       [-1.54699187e-09,  7.66528785e-07,  1.43689363e-06, ...,\n",
      "        -2.12415744e-05,  2.01281478e-06, -2.80905806e-06],\n",
      "       [ 1.70560565e-06, -5.42291582e-06, -5.93398454e-07, ...,\n",
      "         2.22114031e-04, -2.63086313e-05,  9.19797458e-05],\n",
      "       [-1.50422784e-05, -6.47841225e-05,  5.17874832e-05, ...,\n",
      "         5.83497342e-04, -1.11180496e-04,  1.80077375e-04]], dtype=float32)>, <tf.Tensor: id=917989, shape=(6144,), dtype=float32, numpy=\n",
      "array([-2.4748186e-05,  7.2338196e-07, -2.6267899e-06, ...,\n",
      "       -2.1503223e-05,  2.2455479e-04,  5.9184548e-04], dtype=float32)>, <tf.Tensor: id=917892, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.62663796e-07,  1.66777056e-04, -1.09200869e-06, ...,\n",
      "        -9.53998438e-07, -2.29105672e-06,  2.77832351e-05],\n",
      "       [-1.43418572e-06,  1.50003485e-04, -1.90519347e-06, ...,\n",
      "        -1.42984982e-06, -4.18270793e-06,  8.99441293e-05],\n",
      "       [-6.44065267e-07,  2.63704715e-04, -8.99545626e-07, ...,\n",
      "        -8.31818738e-07, -2.55846908e-06,  2.37579254e-04],\n",
      "       ...,\n",
      "       [ 6.07111497e-06, -6.38850033e-04,  6.45080809e-06, ...,\n",
      "         6.24832637e-06,  1.38292125e-05, -2.88754818e-04],\n",
      "       [-6.97783435e-07,  2.69402721e-04, -8.47514912e-07, ...,\n",
      "        -7.85253462e-07, -1.67118640e-06,  5.44775576e-05],\n",
      "       [ 1.47087690e-06,  3.60063685e-04,  1.75796924e-06, ...,\n",
      "         1.53858798e-06,  3.22435039e-06,  1.06702624e-04]], dtype=float32)>, <tf.Tensor: id=917884, shape=(110,), dtype=float32, numpy=\n",
      "array([ 6.17857722e-06, -6.49023219e-04,  6.55715212e-06,  8.33707672e-05,\n",
      "       -1.01001700e-03,  4.55758309e-05,  1.52115861e-03,  9.55748510e-06,\n",
      "        8.49420485e-06, -1.46374863e-04,  3.93803231e-04, -9.05148991e-05,\n",
      "        2.36512278e-04, -1.16287585e-04,  1.43543275e-05,  2.04146933e-03,\n",
      "       -6.07545662e-04, -1.24602288e-03,  8.04856172e-05,  4.37399081e-04,\n",
      "        8.86835041e-05,  3.13112105e-04,  7.56286900e-05,  2.61608249e-04,\n",
      "       -2.25780503e-04, -6.27574773e-05,  1.78535207e-04, -8.85601155e-04,\n",
      "        3.32555559e-04, -6.50709335e-05, -1.01129164e-03,  2.11692131e-05,\n",
      "        1.46870570e-05,  3.34817814e-05,  2.41230955e-04, -2.98022176e-04,\n",
      "       -8.00878624e-04,  4.23583202e-04, -5.70318487e-04,  3.50722585e-05,\n",
      "        5.62744390e-04, -1.97897461e-04, -1.51283180e-04,  4.03166137e-04,\n",
      "        3.32192103e-05, -2.09595979e-04,  5.14112937e-04, -4.45725484e-04,\n",
      "       -2.90966447e-04, -6.08672053e-05, -9.39512393e-05,  5.81964268e-05,\n",
      "        5.20808317e-05,  4.72666317e-04, -1.39992975e-03,  1.12633294e-04,\n",
      "       -3.15129219e-05, -4.28965635e-04,  5.75307349e-05, -1.59353134e-04,\n",
      "        3.64640400e-05, -1.89742190e-04,  2.51544316e-06, -4.25997074e-04,\n",
      "        2.87977746e-05, -6.72636204e-04, -1.63195073e-03, -8.61032866e-03,\n",
      "        8.20644107e-03, -9.22966003e-03,  4.47929429e-04, -2.78438325e-04,\n",
      "        2.39244616e-03,  6.89122244e-04,  6.37909980e-04, -6.96926960e-04,\n",
      "        5.24196355e-03, -2.56023067e-03, -1.37533969e-03, -3.79378512e-03,\n",
      "        2.93012999e-05,  6.42588566e-05,  3.54702398e-03,  4.63438826e-03,\n",
      "        2.55778851e-03, -4.35689057e-04,  1.94920576e-03,  1.08085922e-03,\n",
      "        7.84548465e-04, -1.52207271e-04, -5.74414516e-05,  1.04127821e-05,\n",
      "        7.55853262e-06,  1.13208098e-05,  8.04196134e-06,  7.25106929e-06,\n",
      "        8.53504935e-06,  7.30223928e-06,  7.04533431e-06,  7.07709432e-06,\n",
      "        8.13615952e-06,  7.41011900e-06,  8.07239030e-06,  7.73388729e-06,\n",
      "        1.45869581e-05,  7.68360587e-06,  8.76705872e-06,  6.35808101e-06,\n",
      "        1.40558732e-05, -2.93010817e-04], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc0f0>, <tf.Tensor: id=918348, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 2.0570719e-06, -1.1953455e-07, -3.3576157e-06, ...,\n",
      "        -6.2512503e-05, -8.3862933e-06,  3.0087789e-05],\n",
      "       [ 4.5852175e-08,  7.5875391e-08,  2.3280101e-07, ...,\n",
      "        -7.0192950e-06, -2.6150069e-06, -8.2028528e-06],\n",
      "       [-1.0099081e-07,  7.8367020e-09, -8.5380947e-08, ...,\n",
      "        -4.5354373e-06,  2.2442557e-05, -7.5808834e-06],\n",
      "       ...,\n",
      "       [ 2.3055125e-06,  7.7460390e-07, -1.4397632e-07, ...,\n",
      "        -7.1472514e-06,  2.7451331e-05,  1.1520820e-05],\n",
      "       [ 6.9272237e-09,  1.7133446e-07,  3.3843257e-08, ...,\n",
      "        -3.8881037e-05, -1.1255582e-05, -3.0078523e-05],\n",
      "       [-5.1916669e-07,  2.4943844e-07,  5.7096432e-08, ...,\n",
      "        -3.3956308e-05,  2.2144081e-05,  3.5611756e-05]], dtype=float32)>, <tf.Tensor: id=918349, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 2.2079212e-05,  2.7550195e-06, -2.6931521e-06, ...,\n",
      "         9.4208319e-04, -9.5930482e-05,  3.1983948e-04],\n",
      "       [-2.9817459e-06, -2.9737221e-06, -2.8064505e-08, ...,\n",
      "        -1.1724660e-04, -1.2803663e-05,  5.0317849e-06],\n",
      "       [-5.0483930e-08,  5.7453451e-07,  4.0424266e-06, ...,\n",
      "         2.0304666e-04, -9.0466847e-07,  5.1082494e-05],\n",
      "       ...,\n",
      "       [ 3.9836294e-07,  7.5681055e-07,  1.4224014e-06, ...,\n",
      "        -2.6281759e-05,  3.9795104e-06, -4.4326775e-06],\n",
      "       [-2.9492193e-07,  1.6411074e-06,  1.1000378e-05, ...,\n",
      "        -1.8138449e-04,  3.6152360e-05, -8.6936656e-05],\n",
      "       [ 1.0486443e-05, -6.6479433e-06,  2.4002595e-05, ...,\n",
      "         5.1068355e-06,  1.3431317e-05,  1.2679242e-04]], dtype=float32)>, <tf.Tensor: id=918350, shape=(6144,), dtype=float32, numpy=\n",
      "array([-2.2914239e-05,  1.1670547e-06, -3.5974522e-06, ...,\n",
      "       -2.6601751e-05, -1.8404918e-04,  3.4613586e-06], dtype=float32)>, <tf.Tensor: id=918253, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.7555816e-07,  5.3257936e-06, -1.0899244e-06, ...,\n",
      "        -9.7966381e-07, -2.2647191e-06,  1.9511896e-05],\n",
      "       [-1.3827007e-06,  2.0836966e-05, -1.8200004e-06, ...,\n",
      "        -1.3995833e-06, -3.8991657e-06,  3.2554267e-06],\n",
      "       [-5.1251124e-07, -1.3173193e-04, -8.0371211e-07, ...,\n",
      "        -7.3610545e-07, -2.3247103e-06,  1.0112976e-04],\n",
      "       ...,\n",
      "       [ 5.7944785e-06, -1.1423367e-04,  6.0146590e-06, ...,\n",
      "         5.9054491e-06,  1.2919028e-05, -1.3620630e-04],\n",
      "       [-7.2396494e-07, -6.9102534e-05, -8.6062266e-07, ...,\n",
      "        -8.1009091e-07, -1.7193332e-06,  8.4656840e-06],\n",
      "       [ 1.2385344e-06,  2.9201714e-05,  1.5715542e-06, ...,\n",
      "         1.3287894e-06,  2.7016572e-06,  4.9761522e-05]], dtype=float32)>, <tf.Tensor: id=918245, shape=(110,), dtype=float32, numpy=\n",
      "array([ 5.88643843e-06, -1.10852168e-04,  6.10409052e-06, -1.07550819e-03,\n",
      "       -1.58618530e-03,  4.56407251e-05,  8.53847450e-05,  9.22014351e-06,\n",
      "        7.99249665e-06,  9.46912132e-06,  4.44407924e-04, -3.04050598e-04,\n",
      "       -8.04674710e-05,  3.93100345e-05,  1.39334452e-05,  3.84871615e-03,\n",
      "        1.00739661e-03,  4.83750846e-05,  7.73081410e-05,  5.13055129e-05,\n",
      "        4.32637666e-04,  2.18849076e-04, -1.37125433e-04,  7.98555120e-05,\n",
      "        2.96634185e-04,  1.11519457e-05,  2.15570893e-04, -3.87059335e-05,\n",
      "        1.70005980e-04,  3.11414275e-04,  1.11513109e-04,  2.02896008e-05,\n",
      "        1.36973867e-05,  3.17364029e-05, -8.37697153e-05,  5.79777843e-05,\n",
      "        2.85698363e-04, -6.61509475e-05,  2.42300608e-04, -1.02858801e-04,\n",
      "        9.81041594e-05,  1.05266008e-04, -2.56307510e-04, -9.83521313e-05,\n",
      "        4.63488686e-05, -4.37033013e-05,  2.24089817e-04,  1.36822491e-04,\n",
      "        1.85984405e-04, -5.02231647e-04, -2.13958818e-04,  5.74424957e-05,\n",
      "        4.83437943e-05,  4.82456293e-04, -1.09924156e-04,  1.18349169e-04,\n",
      "       -3.41169471e-05, -8.83141649e-04, -2.57034117e-04,  3.58608281e-06,\n",
      "        3.34130054e-05,  1.20638353e-04,  1.54664202e-04, -4.27766296e-04,\n",
      "        2.74958857e-05,  9.08561237e-03, -3.02727887e-04, -2.07240740e-03,\n",
      "        2.87285284e-03, -8.70222319e-03,  1.69858581e-03,  1.22718525e-03,\n",
      "       -2.67839408e-03, -6.36545662e-03, -5.91269869e-04, -8.61699344e-04,\n",
      "        2.58623296e-03, -1.50970672e-03, -5.41782193e-03,  7.06650363e-03,\n",
      "       -1.19109533e-03,  7.77040623e-05,  1.93530973e-03, -4.99023404e-03,\n",
      "        3.17382091e-03, -1.41872093e-03,  9.04502813e-05,  2.21120007e-03,\n",
      "        6.31184259e-04,  7.37424707e-05,  4.30578439e-05,  9.51557195e-06,\n",
      "        7.13345298e-06, -1.45511818e-04,  7.52606820e-06,  6.85571922e-06,\n",
      "        7.86738019e-06,  6.87179863e-06, -1.49840547e-04,  6.60875139e-06,\n",
      "        7.51720791e-06,  7.05598632e-06,  7.58830402e-06,  7.15230226e-06,\n",
      "        1.37954457e-05,  7.34097148e-06,  8.47575029e-06,  5.99765144e-06,\n",
      "        1.31137813e-05, -1.37321564e-04], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdb0a940>, <tf.Tensor: id=918709, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 2.33188098e-06,  6.15489307e-06, -1.86456555e-06, ...,\n",
      "        -3.86244246e-05, -1.24088292e-05,  3.26511217e-05],\n",
      "       [ 1.24389077e-08,  1.15803303e-07, -1.67403016e-08, ...,\n",
      "        -2.20135807e-05,  2.97191127e-05,  4.04101083e-05],\n",
      "       [ 3.50309165e-07,  1.69495479e-07, -1.44619378e-07, ...,\n",
      "        -3.64724146e-05,  2.87095354e-05, -2.26366501e-05],\n",
      "       ...,\n",
      "       [ 6.24651148e-06,  2.63498032e-06, -2.22104654e-06, ...,\n",
      "        -1.21876465e-05,  2.31721024e-05,  5.13593068e-06],\n",
      "       [-1.97878649e-07, -1.89034974e-08,  9.92832909e-08, ...,\n",
      "        -3.19066821e-05, -3.33408298e-06,  1.08530849e-05],\n",
      "       [-1.66954999e-06, -2.83301290e-07,  1.15339583e-06, ...,\n",
      "         6.72119277e-05, -4.73193213e-05, -4.89160584e-06]], dtype=float32)>, <tf.Tensor: id=918710, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 3.9557657e-05,  2.0184820e-05,  1.4976462e-05, ...,\n",
      "         4.4816820e-04, -5.9600170e-05,  5.7842328e-05],\n",
      "       [-7.9243847e-07, -4.8423490e-06,  1.9476690e-06, ...,\n",
      "        -4.1244606e-05, -7.9008860e-06, -6.2561005e-05],\n",
      "       [-4.1120816e-07,  1.3192342e-06,  6.0685517e-07, ...,\n",
      "         1.2015773e-04, -1.4521065e-05, -2.6221800e-05],\n",
      "       ...,\n",
      "       [-1.8771661e-07, -5.3142338e-07,  3.8771282e-07, ...,\n",
      "        -2.5276529e-06, -1.0494555e-06,  3.0851595e-06],\n",
      "       [ 1.2066452e-05,  2.2915810e-05,  4.0568899e-05, ...,\n",
      "        -5.7038548e-04,  5.7764773e-05, -2.0824911e-04],\n",
      "       [ 2.5777022e-06, -1.9934057e-05,  2.3810737e-05, ...,\n",
      "         4.4149616e-05,  2.7140877e-05,  2.3056673e-05]], dtype=float32)>, <tf.Tensor: id=918711, shape=(6144,), dtype=float32, numpy=\n",
      "array([-6.6247871e-05, -1.7080774e-06,  1.4962268e-08, ...,\n",
      "       -2.5846696e-06, -5.7766377e-04,  4.2714422e-05], dtype=float32)>, <tf.Tensor: id=918614, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.5037448e-07, -6.0990282e-05, -1.0445640e-06, ...,\n",
      "        -9.5196577e-07, -2.1540957e-06, -3.6815384e-06],\n",
      "       [-1.2398989e-06, -9.0864487e-05, -1.7004578e-06, ...,\n",
      "        -1.3010765e-06, -3.6586291e-06, -6.1207270e-06],\n",
      "       [-5.5314831e-07,  8.0191603e-06, -8.1188637e-07, ...,\n",
      "        -7.1905947e-07, -2.3139419e-06, -3.9878146e-06],\n",
      "       ...,\n",
      "       [ 5.8726205e-06,  2.3312705e-04,  6.2058079e-06, ...,\n",
      "         6.0546854e-06,  1.3021675e-05,  1.9466972e-05],\n",
      "       [-7.7644722e-07,  1.6094731e-05, -9.0189496e-07, ...,\n",
      "        -8.6530633e-07, -1.7622129e-06, -2.9941907e-06],\n",
      "       [ 1.2103333e-06, -1.3748553e-04,  1.5467773e-06, ...,\n",
      "         1.3466093e-06,  2.6492389e-06,  4.0098275e-06]], dtype=float32)>, <tf.Tensor: id=918606, shape=(110,), dtype=float32, numpy=\n",
      "array([ 5.96944847e-06,  2.35968328e-04,  6.30107979e-06,  2.50550685e-04,\n",
      "       -4.63339128e-03,  4.47834827e-05, -2.83863454e-04,  9.56870008e-06,\n",
      "        8.29054716e-06,  9.88976808e-06,  5.84998867e-04, -1.10510300e-05,\n",
      "       -4.05006984e-04,  3.97483127e-05,  1.37970528e-05,  1.37344911e-03,\n",
      "       -3.35867284e-04,  1.68289727e-04,  7.31050895e-05,  3.19142564e-05,\n",
      "        1.73610140e-04,  3.25778179e-04, -1.78909817e-04,  2.10059836e-04,\n",
      "        1.02926082e-04, -6.44440064e-04,  2.51107103e-05,  1.31433662e-05,\n",
      "        1.48322360e-04,  2.99745356e-04, -6.60856371e-04,  2.05485303e-05,\n",
      "        1.38494534e-05,  3.10182695e-05,  5.50894365e-05,  7.49922357e-04,\n",
      "       -8.11348727e-04,  3.53113661e-04,  1.01488360e-04,  3.44697095e-04,\n",
      "        2.11132632e-04,  9.53385752e-05, -4.08204331e-04,  2.26303673e-04,\n",
      "       -1.17145071e-04,  1.03885090e-04,  3.59067089e-05, -3.39005754e-04,\n",
      "        3.35305871e-04,  8.41399306e-05,  4.30715940e-04,  5.75408776e-05,\n",
      "       -4.38962277e-04,  5.43167698e-04, -4.50606953e-04, -1.94119813e-04,\n",
      "        1.15554954e-04, -1.40145421e-04,  5.18843044e-05,  1.59447562e-04,\n",
      "        3.24952198e-05, -3.73129769e-05, -1.08591912e-05,  4.38012357e-05,\n",
      "       -2.85189482e-04,  4.10537235e-03,  2.88029108e-03,  6.04635454e-04,\n",
      "        2.19433522e-03,  1.15348445e-03, -1.24931440e-03, -1.49826764e-03,\n",
      "       -2.89791380e-03, -8.72976705e-03, -3.40903905e-04, -1.97781320e-03,\n",
      "        2.60038907e-03,  1.17458485e-03, -5.60687250e-03,  1.11289099e-02,\n",
      "        4.57911869e-04,  5.25248237e-04,  5.64382738e-03, -6.27964456e-03,\n",
      "       -1.56842684e-03, -3.03246640e-03, -7.03050115e-04,  2.53194245e-03,\n",
      "        1.09381341e-04,  5.01509989e-04,  1.37037292e-04,  1.00701009e-05,\n",
      "        7.14165571e-06,  1.09658022e-05,  7.78706544e-06,  6.91712330e-06,\n",
      "        8.13297993e-06,  6.92438471e-06,  6.88147657e-06,  6.63107767e-06,\n",
      "        7.46485057e-06,  7.25681639e-06,  7.51886000e-06,  7.55653673e-06,\n",
      "        1.38382602e-05,  7.61797264e-06,  8.36033814e-06,  6.15237423e-06,\n",
      "        1.32220348e-05,  1.97553527e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc358>, <tf.Tensor: id=919070, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 3.23183826e-06, -1.78719318e-07,  2.50532167e-07, ...,\n",
      "        -1.04853098e-04,  4.86062927e-05,  7.20910975e-06],\n",
      "       [-4.97439281e-08,  3.11676956e-07, -4.21295923e-08, ...,\n",
      "         1.25013576e-05,  9.26760731e-06, -1.48081153e-05],\n",
      "       [ 1.19895731e-07, -5.54317481e-09,  7.11797483e-08, ...,\n",
      "        -4.87823781e-05,  2.69138436e-05,  3.95985007e-05],\n",
      "       ...,\n",
      "       [ 1.88951390e-06, -3.79024272e-07,  1.81795301e-07, ...,\n",
      "        -8.07502965e-06,  3.00481079e-05,  2.56157909e-05],\n",
      "       [-1.65480927e-07,  9.86927873e-08,  4.33781473e-08, ...,\n",
      "         4.57462775e-05, -4.79686823e-05, -6.58887948e-05],\n",
      "       [-9.25906193e-07,  7.02353418e-07,  8.45165800e-07, ...,\n",
      "         3.18086495e-05, -1.04836854e-05,  1.28578595e-05]], dtype=float32)>, <tf.Tensor: id=919071, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 1.8116598e-05,  6.0400898e-06,  5.3404679e-06, ...,\n",
      "         1.8678945e-04, -3.5330831e-05, -8.1680128e-06],\n",
      "       [-1.7822823e-06, -3.4354255e-06, -5.1206973e-07, ...,\n",
      "        -9.1819667e-05, -3.8222366e-05,  1.7720300e-05],\n",
      "       [-1.7366305e-06, -8.3661280e-07, -8.2613514e-07, ...,\n",
      "        -4.2170737e-05,  2.0335876e-05,  7.6365694e-05],\n",
      "       ...,\n",
      "       [ 3.5359051e-07,  8.9994074e-07,  1.1307143e-06, ...,\n",
      "        -1.8134333e-05,  2.1983078e-06, -3.5847549e-06],\n",
      "       [ 6.1422743e-06,  1.2309542e-05,  1.6532163e-05, ...,\n",
      "        -3.9754700e-04,  5.1565294e-05, -1.4637184e-04],\n",
      "       [ 2.1127948e-05,  3.2392672e-05, -3.3224631e-05, ...,\n",
      "         2.3447974e-06,  8.6622633e-05, -5.8936279e-05]], dtype=float32)>, <tf.Tensor: id=919072, shape=(6144,), dtype=float32, numpy=\n",
      "array([-1.1089193e-05,  5.9935919e-06,  6.9103016e-06, ...,\n",
      "       -1.8241037e-05, -4.0428081e-04,  3.7229502e-06], dtype=float32)>, <tf.Tensor: id=918975, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-8.3829559e-07, -3.0501407e-05, -1.0906620e-06, ...,\n",
      "        -9.6398389e-07, -2.2208096e-06, -4.0546779e-06],\n",
      "       [-1.2616493e-06, -6.4201493e-05, -1.7561520e-06, ...,\n",
      "        -1.3572424e-06, -3.7062043e-06, -6.5231557e-06],\n",
      "       [-6.0536604e-07, -2.7765025e-05, -8.3864734e-07, ...,\n",
      "        -7.5218554e-07, -2.4607727e-06, -4.5669121e-06],\n",
      "       ...,\n",
      "       [ 4.9304899e-06,  4.0621302e-04,  5.6204308e-06, ...,\n",
      "         5.3241674e-06,  1.1702773e-05,  1.8832412e-05],\n",
      "       [-6.6817967e-07, -2.2586901e-06, -8.4804367e-07, ...,\n",
      "        -7.6212007e-07, -1.6632985e-06, -3.0280198e-06],\n",
      "       [ 1.1620665e-06,  5.2230927e-05,  1.5332540e-06, ...,\n",
      "         1.3660730e-06,  2.5204918e-06,  4.0599680e-06]], dtype=float32)>, <tf.Tensor: id=918967, shape=(110,), dtype=float32, numpy=\n",
      "array([ 5.01773684e-06,  4.12075198e-04,  5.71308465e-06, -1.54186608e-04,\n",
      "        3.57200042e-03,  4.45511469e-05,  1.35921361e-03,  8.25298139e-06,\n",
      "        7.11486473e-06,  8.86302314e-06,  1.18891196e-03, -4.64859331e-05,\n",
      "       -4.13985166e-04,  3.51229573e-05,  1.24873959e-05,  2.34023901e-07,\n",
      "        1.10044144e-03, -2.76999985e-04,  6.83398102e-05, -2.97008024e-04,\n",
      "        5.44497103e-04, -1.38668620e-05,  3.07894195e-04,  3.62300343e-04,\n",
      "       -1.65511738e-04, -7.92310457e-04, -1.20568271e-04,  2.36403677e-04,\n",
      "        1.88731312e-04,  4.94871056e-04,  3.27591988e-04,  1.74712295e-05,\n",
      "        1.27098792e-05,  2.79470751e-05,  2.09901686e-04,  7.31723849e-05,\n",
      "        2.62680551e-04,  3.86512489e-04, -8.28881239e-05,  1.34116795e-04,\n",
      "        1.24450948e-04, -8.76468694e-05, -1.03951199e-04,  1.63829944e-04,\n",
      "        2.17802790e-05, -7.93495710e-05,  1.07585634e-04, -4.92172956e-04,\n",
      "       -6.48561807e-04, -1.45272512e-04,  5.39628556e-04,  5.37440137e-05,\n",
      "        3.04051850e-04,  6.54225296e-05,  1.33473775e-04,  1.07373766e-04,\n",
      "        9.70214387e-05,  2.35702333e-04,  4.53008652e-05,  1.37773182e-04,\n",
      "        2.95156824e-05,  1.06710919e-04,  1.35539391e-04,  3.85888889e-05,\n",
      "        2.56374897e-05, -1.44525652e-03, -1.79927301e-04,  8.75562138e-04,\n",
      "        1.27746677e-03,  5.24948072e-03, -1.14793082e-04,  1.07879844e-03,\n",
      "       -3.39672365e-03, -4.45746910e-03,  5.75911137e-04,  2.93676509e-04,\n",
      "       -3.77141219e-03,  3.26107023e-03, -1.07310778e-02,  1.11981295e-02,\n",
      "       -2.07848311e-03,  1.94929991e-04, -7.80853094e-04, -9.21035744e-03,\n",
      "        3.74524575e-03, -2.72535463e-03,  3.27401882e-04,  1.42820587e-03,\n",
      "        5.33996208e-05, -1.37383211e-03,  5.88101917e-04,  8.97526843e-06,\n",
      "        6.36781260e-06,  9.37771438e-06,  6.63242372e-06,  6.01162992e-06,\n",
      "        7.02328180e-06,  5.90097898e-06,  5.85882617e-06,  5.66497283e-06,\n",
      "        6.50900165e-06,  6.18037711e-06,  6.70364079e-06,  6.63241599e-06,\n",
      "        1.22949277e-05,  6.51633991e-06,  7.57351791e-06,  5.41590589e-06,\n",
      "        1.18930939e-05,  1.91247746e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdb0a358>, <tf.Tensor: id=919431, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[ 1.4287278e-06,  1.1226808e-06,  2.8462738e-08, ...,\n",
      "        -3.8442406e-05,  2.4122435e-06,  7.1018338e-05],\n",
      "       [-3.8866298e-07,  2.7186917e-07,  1.9769374e-07, ...,\n",
      "         3.5858677e-05,  9.9381787e-06,  1.5594722e-05],\n",
      "       [-1.9743817e-08, -8.2715941e-09, -9.4739256e-08, ...,\n",
      "        -4.9278846e-05,  4.0559338e-05,  5.7683668e-05],\n",
      "       ...,\n",
      "       [ 4.2224201e-06,  8.2515214e-07, -1.3571689e-06, ...,\n",
      "         1.8665951e-05,  1.0039324e-05,  1.5276672e-05],\n",
      "       [-4.8276945e-08, -2.9934815e-09,  4.8334901e-08, ...,\n",
      "        -1.4523836e-06, -1.2015834e-05, -1.5546184e-05],\n",
      "       [-1.5217132e-06,  5.1592065e-07,  7.5257515e-07, ...,\n",
      "         2.7915088e-05, -4.4697965e-05, -3.2715609e-05]], dtype=float32)>, <tf.Tensor: id=919432, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-2.14782040e-06, -9.17102352e-06, -7.59939724e-07, ...,\n",
      "        -3.45633831e-04,  7.86919991e-05, -9.97573952e-05],\n",
      "       [-2.63620268e-06, -3.74850833e-06,  7.50987397e-07, ...,\n",
      "        -7.95527594e-05, -5.22454866e-06, -4.21831210e-05],\n",
      "       [-4.12535655e-07,  9.44197751e-08, -3.41763325e-06, ...,\n",
      "        -1.62627344e-04,  1.53483816e-05,  7.56161171e-07],\n",
      "       ...,\n",
      "       [ 2.35197902e-07,  4.78292634e-07, -9.90761578e-07, ...,\n",
      "         2.11938113e-05, -1.74106526e-06,  3.46195793e-06],\n",
      "       [ 1.65572686e-07, -5.08348194e-06,  1.58862179e-06, ...,\n",
      "        -1.96567824e-04,  1.20914674e-05, -1.08204906e-04],\n",
      "       [ 7.57078487e-06, -3.83190581e-06, -2.21278988e-05, ...,\n",
      "         8.80347274e-04, -1.49902888e-04,  3.08148126e-04]], dtype=float32)>, <tf.Tensor: id=919433, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 2.4572906e-05,  9.5335063e-06,  1.8528216e-06, ...,\n",
      "        2.1481839e-05, -2.0018310e-04,  8.9287997e-04], dtype=float32)>, <tf.Tensor: id=919336, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.4591962e-07,  9.7541742e-05, -1.1660145e-06, ...,\n",
      "        -1.0801054e-06, -2.4038573e-06, -4.3310383e-06],\n",
      "       [-1.3453894e-06, -6.3584303e-05, -1.7772495e-06, ...,\n",
      "        -1.3899541e-06, -3.7169216e-06, -6.6732932e-06],\n",
      "       [-5.4148728e-07,  1.1719577e-04, -8.1501241e-07, ...,\n",
      "        -8.0973382e-07, -2.4266219e-06, -4.6572659e-06],\n",
      "       ...,\n",
      "       [ 5.6476174e-06, -9.2426897e-05,  5.9200765e-06, ...,\n",
      "         6.0004413e-06,  1.2623843e-05,  2.0491276e-05],\n",
      "       [-6.9918667e-07,  3.6191002e-05, -8.3896384e-07, ...,\n",
      "        -8.1560444e-07, -1.6529420e-06, -3.0837552e-06],\n",
      "       [ 1.0801866e-06,  1.8574782e-04,  1.4972400e-06, ...,\n",
      "         1.2419528e-06,  2.3827160e-06,  3.9495808e-06]], dtype=float32)>, <tf.Tensor: id=919328, shape=(110,), dtype=float32, numpy=\n",
      "array([ 5.73600664e-06, -8.84381370e-05,  6.00722933e-06, -2.74657272e-04,\n",
      "        3.43026198e-03,  4.47682469e-05, -8.33002501e-04,  9.18823571e-06,\n",
      "        7.87924091e-06,  9.46842374e-06,  6.67918532e-04,  1.10294670e-04,\n",
      "       -2.16935645e-04,  3.92813672e-05,  1.38699515e-05, -2.72948120e-04,\n",
      "       -3.24765133e-04,  8.36791762e-04, -8.67582858e-05,  5.43256756e-04,\n",
      "        4.31734457e-04, -4.04275459e-04, -1.89500992e-04,  4.83042590e-04,\n",
      "       -1.10093570e-04, -1.92596999e-04, -6.92355330e-04,  4.27155683e-05,\n",
      "        2.71255907e-04, -3.10947886e-04, -1.28073720e-04,  2.01027269e-05,\n",
      "        1.39089143e-05,  3.03936522e-05, -4.33981710e-04, -9.16593650e-04,\n",
      "        3.39621583e-05,  2.72413075e-04, -2.20996990e-05,  2.14489573e-05,\n",
      "        3.41248349e-04, -7.11722096e-05,  3.68940702e-04, -3.00593238e-04,\n",
      "        5.87047398e-05, -1.97862915e-04,  4.98355483e-04, -2.41545597e-04,\n",
      "       -2.72717443e-04, -5.01427712e-05,  6.04002271e-04,  5.70129669e-05,\n",
      "        5.48370008e-05, -7.62151903e-04,  5.95841993e-05, -1.88797392e-04,\n",
      "       -4.88346559e-05,  2.38791574e-04,  5.13422165e-05, -4.59375093e-04,\n",
      "        2.97019433e-05, -1.94976586e-04, -2.50282301e-05,  4.66115707e-05,\n",
      "        2.77942763e-05, -3.92678799e-03, -8.32091144e-04,  7.98417535e-03,\n",
      "        9.60713078e-05,  8.55912734e-03,  1.47293706e-03, -1.22262340e-03,\n",
      "        2.00187016e-04, -3.94730014e-04, -2.01620991e-04, -4.15707269e-04,\n",
      "       -3.63040040e-03,  2.22557387e-03, -5.98223833e-03, -4.03397222e-04,\n",
      "       -3.10892682e-03, -4.81749477e-04,  7.67026446e-04, -5.43114496e-03,\n",
      "        2.04361952e-03, -3.16674449e-03,  1.21346721e-03,  3.18098883e-03,\n",
      "       -2.19766167e-04,  3.60869919e-04, -1.37176830e-05,  9.39186521e-06,\n",
      "        7.03614478e-06,  1.10419587e-05,  7.65409641e-06,  6.82003110e-06,\n",
      "        7.64969991e-06,  7.10734275e-06,  7.02361240e-06,  6.68352550e-06,\n",
      "       -1.48770603e-04, -1.49206375e-04,  7.51388689e-06,  7.25047039e-06,\n",
      "        1.36359868e-05,  7.16374825e-06,  8.51885488e-06,  6.09239305e-06,\n",
      "        1.28117590e-05,  2.07839894e-05], dtype=float32)>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdb0af60>, <tf.Tensor: id=919792, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-3.71483884e-06, -3.15255147e-06,  2.62256322e-06, ...,\n",
      "         1.59173069e-05, -1.92328407e-05, -2.98543237e-05],\n",
      "       [-1.20727543e-06, -3.34444508e-07,  3.48126008e-07, ...,\n",
      "         6.47762936e-05, -8.63289461e-06, -1.20562345e-05],\n",
      "       [ 3.14985670e-07, -3.92516739e-08,  7.32983141e-09, ...,\n",
      "        -4.67955470e-05,  1.08446684e-05, -1.33166641e-05],\n",
      "       ...,\n",
      "       [ 1.89275147e-06,  1.19670881e-06, -1.00265663e-06, ...,\n",
      "        -6.16513398e-06, -1.40243974e-05, -1.59270458e-05],\n",
      "       [-6.27527825e-08,  8.70638033e-08,  2.89620736e-08, ...,\n",
      "         9.38427888e-07,  2.15831624e-05, -4.53977409e-05],\n",
      "       [ 6.41095141e-07,  1.71443173e-06,  5.65115066e-08, ...,\n",
      "        -9.20488164e-05,  1.44293663e-04,  3.86730935e-05]], dtype=float32)>, <tf.Tensor: id=919793, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-2.95812533e-05, -2.48144697e-05, -1.13728674e-05, ...,\n",
      "        -9.30296141e-04,  1.21608369e-04, -3.37986305e-04],\n",
      "       [-1.20307266e-06, -2.31015065e-06,  1.22371262e-06, ...,\n",
      "        -2.12723084e-04,  2.87741677e-05, -9.52945848e-05],\n",
      "       [ 1.10021278e-06, -2.84288888e-07,  1.99228430e-06, ...,\n",
      "         1.46630355e-05,  1.53663823e-05, -5.62870355e-06],\n",
      "       ...,\n",
      "       [ 2.73456806e-07,  1.90960805e-07,  7.80248342e-07, ...,\n",
      "        -5.59564796e-06,  7.51800968e-08, -1.24351129e-06],\n",
      "       [ 1.62731703e-06,  1.37339212e-05,  2.10837698e-05, ...,\n",
      "        -9.29874659e-05, -1.31440393e-05,  8.04570391e-07],\n",
      "       [ 1.05230538e-05,  2.59194621e-05, -2.21342143e-05, ...,\n",
      "        -4.71197738e-04,  2.75267412e-05, -4.19109347e-05]], dtype=float32)>, <tf.Tensor: id=919794, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 7.5725307e-05,  4.0812101e-06, -2.2574721e-07, ...,\n",
      "       -5.5854562e-06, -9.5269221e-05, -4.7639044e-04], dtype=float32)>, <tf.Tensor: id=919697, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.4445988e-07, -3.8039070e-06, -1.1796340e-06, ...,\n",
      "         4.8119193e-05, -2.3886798e-06, -4.5018896e-06],\n",
      "       [-1.2091181e-06, -7.9729361e-06, -1.6374781e-06, ...,\n",
      "         8.4630556e-05, -3.5003795e-06, -6.4167489e-06],\n",
      "       [-5.8631827e-07, -1.5964734e-04, -8.1842848e-07, ...,\n",
      "         1.0750995e-04, -2.3071002e-06, -4.7231320e-06],\n",
      "       ...,\n",
      "       [ 5.5980481e-06, -8.7202532e-04,  5.9310505e-06, ...,\n",
      "        -1.4822104e-04,  1.2377703e-05,  2.0641250e-05],\n",
      "       [-6.6147607e-07,  1.8110900e-04, -7.9186975e-07, ...,\n",
      "         3.3297471e-05, -1.5167077e-06, -2.9502994e-06],\n",
      "       [ 1.1138105e-06, -4.3275973e-04,  1.4983498e-06, ...,\n",
      "         1.5303533e-05,  2.5105517e-06,  3.9660395e-06]], dtype=float32)>, <tf.Tensor: id=919689, shape=(110,), dtype=float32, numpy=\n",
      "array([ 5.69249914e-06, -8.77254060e-04,  6.02332420e-06,  9.95105365e-05,\n",
      "       -5.37142763e-03,  4.36325354e-05,  1.52699911e-04,  9.15284909e-06,\n",
      "        8.08265304e-06,  9.30570241e-06, -4.82180330e-04,  6.95986673e-05,\n",
      "       -1.82637916e-04,  3.77637189e-05, -1.42686549e-04, -3.00260191e-03,\n",
      "        4.24163503e-04, -5.63900452e-04, -8.87922579e-05,  4.12414578e-04,\n",
      "       -2.30000413e-04,  2.30540580e-04, -1.73286549e-04,  6.46556087e-04,\n",
      "       -9.88693428e-05, -4.45353762e-05, -3.43477499e-04,  2.67510914e-04,\n",
      "       -8.11423175e-04,  2.47280899e-04,  2.29044381e-05,  1.96779438e-05,\n",
      "        1.33606900e-05,  2.98830473e-05,  1.73054956e-04,  6.72015522e-05,\n",
      "       -4.13369271e-04, -8.52557307e-04, -7.10693712e-07,  3.47681053e-05,\n",
      "        7.53410859e-06, -5.42501977e-04, -6.22494495e-04, -7.08346488e-04,\n",
      "       -4.01167083e-04, -3.44027241e-04,  3.69472727e-05,  3.81584628e-04,\n",
      "        7.03507903e-05, -4.89852508e-04, -6.35290489e-05,  5.48047683e-05,\n",
      "       -8.58118656e-05, -2.51681835e-04,  5.47250966e-04, -3.14015488e-05,\n",
      "       -2.08522688e-04,  1.11366033e-04,  5.16299369e-05, -1.32669520e-04,\n",
      "        2.81757439e-05, -3.46006746e-05, -3.05452631e-05,  4.54830224e-05,\n",
      "        2.81241009e-05, -5.29547175e-03,  1.67320902e-03,  2.13249028e-03,\n",
      "        7.99296249e-04,  9.89461085e-04,  8.33398663e-04,  2.75749422e-04,\n",
      "        3.51784704e-03,  6.95511326e-03,  4.65018238e-04, -1.86816382e-03,\n",
      "       -3.30142933e-03,  1.11039542e-03,  8.91799573e-03, -4.91018780e-03,\n",
      "       -6.64261752e-04, -2.00888426e-05, -5.03951078e-03,  2.54017720e-03,\n",
      "        1.36899867e-03,  2.17817980e-03,  2.36614910e-03, -9.01445630e-04,\n",
      "        1.70778541e-04, -6.66946289e-04, -4.05836821e-04,  9.50950198e-06,\n",
      "        6.86503108e-06,  1.09802295e-05,  7.57323323e-06,  6.84045426e-06,\n",
      "        7.67286110e-06,  6.94311348e-06,  6.90863908e-06,  6.46163335e-06,\n",
      "        7.43887949e-06,  7.14454609e-06,  7.54559187e-06,  7.42412021e-06,\n",
      "        1.36488507e-05,  7.35595995e-06,  8.30625959e-06, -1.50228720e-04,\n",
      "        1.25693241e-05,  2.09549471e-05], dtype=float32)>]\n",
      "Epoch 1 Batch 100 Loss 2.0690\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafcb38>, <tf.Tensor: id=920154, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-8.9270115e-06, -3.4838565e-06,  5.6338154e-06, ...,\n",
      "        -1.5811451e-05, -5.6480665e-05, -2.4272105e-05],\n",
      "       [-1.2808327e-07,  4.2975230e-07, -5.2311616e-08, ...,\n",
      "        -5.7022182e-05,  2.9607771e-05,  4.2234287e-06],\n",
      "       [-4.7581025e-07, -2.4620283e-07, -2.9024392e-08, ...,\n",
      "        -6.4111748e-07, -2.0059118e-05, -1.0994399e-05],\n",
      "       ...,\n",
      "       [ 2.7966259e-06,  1.3408137e-06, -1.4043452e-06, ...,\n",
      "         2.8671249e-07, -1.4216143e-05, -2.0047351e-05],\n",
      "       [-2.4877497e-07, -1.3144843e-07,  9.4427278e-08, ...,\n",
      "         9.2588059e-05, -7.2925417e-05,  6.8201596e-05],\n",
      "       [-3.2205682e-07, -4.4409231e-07,  5.6456730e-07, ...,\n",
      "        -7.8980611e-06,  9.6509590e-05,  5.0511415e-05]], dtype=float32)>, <tf.Tensor: id=920155, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-5.1288287e-05, -3.6203448e-05, -2.6522082e-05, ...,\n",
      "        -1.1686133e-03,  1.2599611e-04, -2.1881876e-04],\n",
      "       [ 2.1534265e-06, -8.2972150e-08,  2.8039515e-06, ...,\n",
      "        -2.3580043e-04,  1.7766764e-05, -1.5045152e-06],\n",
      "       [ 5.1518117e-07, -3.9856005e-07, -3.4857871e-06, ...,\n",
      "        -2.7950975e-04,  2.8700837e-05, -4.9045480e-05],\n",
      "       ...,\n",
      "       [ 7.0594496e-08, -2.8357383e-07, -2.4503973e-07, ...,\n",
      "         2.3910154e-05, -2.5804909e-06,  2.5800241e-06],\n",
      "       [ 2.9010710e-06, -6.0192538e-06, -1.0943392e-05, ...,\n",
      "         2.3631034e-04, -3.7532103e-05,  1.7823304e-04],\n",
      "       [-2.5862386e-05, -3.3890461e-05,  6.0480170e-06, ...,\n",
      "        -1.0057662e-03,  1.2841952e-04, -2.1960432e-07]], dtype=float32)>, <tf.Tensor: id=920156, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 1.0412797e-04, -9.2938217e-06,  3.5285295e-06, ...,\n",
      "        2.4242165e-05,  2.3947333e-04, -1.0180215e-03], dtype=float32)>, <tf.Tensor: id=920059, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.13214662e-06, -6.89468579e-05, -1.33290109e-06, ...,\n",
      "        -1.29712316e-06, -2.73503701e-06, -5.03789443e-06],\n",
      "       [-1.17452521e-06, -4.22880694e-05, -1.69648456e-06, ...,\n",
      "        -1.26997281e-06, -3.62990841e-06, -6.84139650e-06],\n",
      "       [-7.31033879e-07, -1.71532301e-04, -9.70150495e-07, ...,\n",
      "        -1.01894352e-06, -2.62813160e-06, -5.48099797e-06],\n",
      "       ...,\n",
      "       [ 7.57390353e-06,  3.16556136e-04,  7.40833275e-06, ...,\n",
      "         7.93092659e-06,  1.54314548e-05,  2.49883506e-05],\n",
      "       [-9.30203100e-07,  9.65109211e-05, -9.94825996e-07, ...,\n",
      "        -1.03944149e-06, -1.91430968e-06, -3.59558499e-06],\n",
      "       [ 1.14556804e-06,  1.18581345e-04,  1.55383884e-06, ...,\n",
      "         1.26548196e-06,  2.48253264e-06,  4.23995698e-06]], dtype=float32)>, <tf.Tensor: id=920051, shape=(110,), dtype=float32, numpy=\n",
      "array([ 7.71432133e-06,  3.18111939e-04,  7.53433642e-06,  8.01878748e-04,\n",
      "        1.38290646e-03, -1.06801235e-04, -3.07096401e-04,  1.13984506e-05,\n",
      "        1.06606485e-05,  1.14482746e-05,  7.47605518e-04, -1.44728459e-03,\n",
      "       -1.13435264e-03,  4.47916573e-05,  1.68304050e-05, -1.50885968e-03,\n",
      "        1.07458665e-03, -2.49907753e-05, -7.04241334e-04,  8.24620656e-04,\n",
      "       -5.85109345e-04, -3.76269280e-04, -1.34055445e-04, -8.30675344e-06,\n",
      "       -7.42045988e-04,  4.03736543e-04, -1.82948716e-04, -1.33180321e-04,\n",
      "       -2.33604733e-04, -1.75505440e-04,  2.34800682e-04, -4.43135796e-04,\n",
      "        1.66238551e-05, -4.31189255e-04, -2.64432485e-04,  7.88010715e-04,\n",
      "       -3.02255998e-04,  4.71250067e-04,  7.32060071e-05, -2.30054909e-04,\n",
      "       -4.22283483e-04,  1.25467981e-04,  3.71077404e-05,  1.50850008e-03,\n",
      "       -3.54735530e-04,  1.88954982e-05,  2.43560309e-04,  1.72665386e-04,\n",
      "       -1.20705690e-05,  6.42918822e-05, -1.75764333e-04,  6.07506299e-05,\n",
      "       -1.93757514e-05, -5.89645293e-04,  6.05159148e-04, -5.33527691e-06,\n",
      "        1.25732768e-04,  2.64526261e-05,  6.35397446e-05, -8.82995882e-05,\n",
      "        3.08404997e-05,  1.46455932e-04,  1.43393292e-04,  6.09288400e-05,\n",
      "        3.33960270e-05, -4.97085229e-03,  2.11612205e-03, -5.02501847e-03,\n",
      "        9.72224050e-04, -2.26049405e-03,  2.75855360e-04, -1.99819123e-03,\n",
      "        5.96182374e-03,  5.21988235e-03, -2.52935970e-05, -2.14180996e-04,\n",
      "       -5.78910392e-03,  4.03198274e-03,  1.47165973e-02, -8.33652727e-03,\n",
      "        3.31841467e-04,  4.62195487e-04, -3.65240662e-03,  5.01803961e-03,\n",
      "       -3.99742974e-03,  1.49415212e-03,  1.44022796e-03, -4.19421634e-03,\n",
      "       -7.97657354e-04,  3.84462473e-05, -1.38918913e-04,  1.17162172e-05,\n",
      "        8.77759521e-06,  1.44887990e-05, -1.45817001e-04, -1.47147599e-04,\n",
      "        1.01543419e-05,  1.01305304e-05,  9.92585774e-06,  8.81428969e-06,\n",
      "        9.89941782e-06,  9.59969293e-06,  9.50973845e-06,  9.81348330e-06,\n",
      "       -1.39673386e-04,  9.63671482e-06,  1.02974745e-05,  8.07427932e-06,\n",
      "        1.56926671e-05,  2.53827184e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdb0a2b0>, <tf.Tensor: id=920515, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-2.6782541e-06, -2.9224445e-06, -4.3400263e-07, ...,\n",
      "        -6.6730441e-05, -2.0485648e-05, -5.2056756e-05],\n",
      "       [-1.3094714e-06, -2.1840721e-07,  6.0534541e-07, ...,\n",
      "        -1.3481648e-06, -5.8701748e-06,  2.7050364e-05],\n",
      "       [-4.6695200e-07, -2.0929598e-08,  2.7275950e-08, ...,\n",
      "        -1.5694186e-05,  6.8184067e-06, -3.3669121e-06],\n",
      "       ...,\n",
      "       [ 7.5708493e-07, -1.2657119e-06,  5.2059562e-07, ...,\n",
      "         1.6603848e-05, -2.2201151e-05, -8.8623965e-06],\n",
      "       [-2.0775597e-07,  8.4096863e-09,  1.1130529e-08, ...,\n",
      "         3.4765850e-05, -6.1021936e-05, -4.1006293e-05],\n",
      "       [ 4.1533386e-07,  8.1833508e-07, -6.8451743e-07, ...,\n",
      "        -3.6065125e-05,  2.4790330e-05,  3.8243721e-05]], dtype=float32)>, <tf.Tensor: id=920516, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[-2.5419097e-06,  5.7023208e-06,  8.8986490e-06, ...,\n",
      "        -2.9282592e-04,  4.4184177e-05, -2.4187229e-04],\n",
      "       [-1.0134547e-06,  8.8281359e-07, -5.3868316e-06, ...,\n",
      "         5.5033359e-04, -7.2329058e-05,  1.9490418e-04],\n",
      "       [ 6.9396421e-08,  1.3966049e-06,  1.5674136e-06, ...,\n",
      "        -1.4525738e-04,  3.9999297e-05,  1.6897767e-06],\n",
      "       ...,\n",
      "       [ 2.3804543e-07,  5.0615017e-07,  2.0262050e-08, ...,\n",
      "        -3.5564619e-06, -9.5410257e-07,  1.5705980e-06],\n",
      "       [-2.2234486e-05, -2.1703545e-05, -1.5820651e-05, ...,\n",
      "         5.9301709e-04, -6.6461027e-05,  2.7424755e-04],\n",
      "       [-8.6822083e-06, -1.7825039e-05,  1.0136295e-05, ...,\n",
      "        -2.1400678e-04,  3.8155482e-05, -1.2774801e-05]], dtype=float32)>, <tf.Tensor: id=920517, shape=(6144,), dtype=float32, numpy=\n",
      "array([ 1.3069522e-05,  4.5755064e-06, -1.1912513e-06, ...,\n",
      "       -3.6178135e-06,  6.0202443e-04, -2.1560838e-04], dtype=float32)>, <tf.Tensor: id=920420, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-9.04028695e-07,  5.95711062e-05, -1.77679522e-05, ...,\n",
      "        -1.05484310e-06, -2.35283937e-06, -4.55769077e-06],\n",
      "       [-1.40443763e-06,  1.48561958e-06,  9.94420589e-06, ...,\n",
      "        -1.49359016e-06, -3.93441269e-06, -7.46094429e-06],\n",
      "       [-5.48911487e-07,  1.53314671e-04, -1.65001984e-05, ...,\n",
      "        -8.01045076e-07, -2.45273714e-06, -5.37657115e-06],\n",
      "       ...,\n",
      "       [ 6.04556499e-06, -5.57513093e-04, -1.46580584e-04, ...,\n",
      "         6.50429229e-06,  1.34964475e-05,  2.28679892e-05],\n",
      "       [-6.24055247e-07, -3.76689386e-05, -7.31120053e-06, ...,\n",
      "        -7.51316009e-07, -1.48941649e-06, -3.03452589e-06],\n",
      "       [ 1.21735820e-06,  4.83589509e-04,  2.53420112e-05, ...,\n",
      "         1.43665056e-06,  2.75117509e-06,  4.31912576e-06]], dtype=float32)>, <tf.Tensor: id=920412, shape=(110,), dtype=float32, numpy=\n",
      "array([ 6.14550345e-06, -5.65814960e-04, -1.49525746e-04, -7.22329016e-04,\n",
      "        4.17023338e-03,  4.97019537e-05,  1.03748149e-04,  9.94632592e-06,\n",
      "        8.75795649e-06,  1.06920370e-05,  1.25636160e-03, -1.28317814e-04,\n",
      "        2.19933441e-04,  3.94458730e-05,  1.51968688e-05, -1.70538755e-04,\n",
      "       -1.80148199e-04,  1.40200625e-03,  7.16673021e-05, -2.12471743e-04,\n",
      "       -1.59178635e-05, -2.91134638e-04, -2.17839435e-04,  1.84483128e-04,\n",
      "       -1.25138817e-04,  5.49842429e-04,  2.43107701e-04,  2.30785939e-04,\n",
      "       -1.27192252e-04,  2.76258303e-04,  1.20383316e-04,  2.13394815e-05,\n",
      "        1.44595351e-05,  3.12874545e-05, -9.82225611e-05,  4.80184972e-05,\n",
      "        1.88120903e-04, -6.10677089e-05,  2.16303888e-04,  4.82776341e-05,\n",
      "        4.01490252e-06,  1.41361859e-04,  4.19070362e-04,  5.47507836e-04,\n",
      "       -2.49685636e-05,  1.85398574e-04,  6.26426598e-04,  1.07247464e-03,\n",
      "        4.43994068e-04, -1.22801022e-04, -1.42424196e-05,  5.90102245e-05,\n",
      "        4.41618904e-04,  9.90983681e-04,  2.22819726e-04, -3.24977358e-04,\n",
      "       -1.99287359e-04,  2.82820489e-04,  5.53363388e-05,  5.63086796e-05,\n",
      "        3.02678745e-05,  1.37513256e-04,  1.26647254e-04,  4.83332551e-05,\n",
      "        3.20652580e-05, -4.80666198e-03,  1.99658520e-04, -3.60512198e-03,\n",
      "       -4.22705757e-03, -1.71607616e-03, -1.63254165e-03, -3.45570873e-03,\n",
      "        1.41086313e-03, -2.30090460e-03, -4.35432506e-04,  1.53456512e-03,\n",
      "        1.78673002e-03, -1.20725378e-03,  6.13880437e-03, -3.94007983e-03,\n",
      "       -7.91141763e-04,  6.36824581e-04, -2.27062753e-03,  7.49641098e-03,\n",
      "       -6.63931773e-04,  6.24222623e-04,  3.38876387e-03, -2.20291968e-03,\n",
      "       -8.28255608e-04, -1.19069777e-03,  4.78285481e-04,  1.03086140e-05,\n",
      "        7.49130459e-06,  1.16417395e-05,  8.15310432e-06,  7.43875717e-06,\n",
      "        8.20965033e-06,  7.39507868e-06,  7.64942979e-06,  7.12477095e-06,\n",
      "       -1.47725528e-04, -1.48532257e-04,  8.24645485e-06,  8.02431441e-06,\n",
      "        1.51138538e-05,  7.88571560e-06,  8.79229538e-06,  6.60931346e-06,\n",
      "        1.37072157e-05,  2.32097827e-05], dtype=float32)>]\n",
      "[<tensorflow.python.framework.ops.IndexedSlices object at 0x7f4efdafc390>, <tf.Tensor: id=920876, shape=(256, 3072), dtype=float32, numpy=\n",
      "array([[-5.15475813e-06, -2.12734972e-06, -4.12893115e-07, ...,\n",
      "        -1.36710398e-04,  3.71869328e-06,  1.33528401e-05],\n",
      "       [-1.47869321e-06,  2.29530443e-07,  2.79018025e-07, ...,\n",
      "         9.74903651e-06,  2.22665058e-05,  2.43510694e-05],\n",
      "       [-3.89717655e-07, -1.36302035e-07, -3.15540575e-07, ...,\n",
      "        -1.60855489e-05,  2.17746820e-05, -2.76188457e-05],\n",
      "       ...,\n",
      "       [ 1.22394385e-06, -5.52183863e-08, -2.54194177e-08, ...,\n",
      "         1.27775575e-05, -1.87366295e-05, -3.67800044e-06],\n",
      "       [-8.06829874e-08,  1.65661831e-07,  9.83325688e-09, ...,\n",
      "         8.78988139e-05, -6.75499614e-05,  2.59976860e-05],\n",
      "       [ 3.38303096e-07,  2.53797992e-07,  1.21453908e-07, ...,\n",
      "         5.97692924e-05, -6.93980037e-05, -6.47456318e-06]], dtype=float32)>, <tf.Tensor: id=920877, shape=(1024, 3072), dtype=float32, numpy=\n",
      "array([[ 1.1983578e-05,  6.5608765e-07, -1.4082102e-05, ...,\n",
      "         1.3387877e-04, -4.8942242e-05,  8.8387242e-06],\n",
      "       [-3.2407181e-06, -4.3466912e-06,  2.1747817e-06, ...,\n",
      "        -4.2578086e-04,  4.1576339e-05, -1.3353521e-05],\n",
      "       [ 2.0356147e-06,  1.3525598e-06,  3.1382783e-06, ...,\n",
      "         1.4521522e-04, -8.9698206e-06, -3.1902928e-05],\n",
      "       ...,\n",
      "       [-1.9024859e-06, -7.2040638e-07, -6.6938787e-06, ...,\n",
      "         1.4437396e-04, -1.1429507e-05,  2.4157222e-05],\n",
      "       [ 1.6170055e-05,  2.5906022e-05,  2.7201453e-05, ...,\n",
      "        -4.0819403e-04,  3.5453915e-05,  7.0665556e-06],\n",
      "       [ 1.0735885e-06, -1.2991260e-05,  8.9904188e-06, ...,\n",
      "         1.8418841e-04, -7.5726115e-05, -1.1886954e-05]], dtype=float32)>, <tf.Tensor: id=920878, shape=(6144,), dtype=float32, numpy=\n",
      "array([-1.5938866e-05, -1.3012943e-06, -1.1461530e-05, ...,\n",
      "        1.5793460e-04, -4.5137291e-04,  1.9300389e-04], dtype=float32)>, <tf.Tensor: id=920781, shape=(1024, 110), dtype=float32, numpy=\n",
      "array([[-1.3390134e-06, -4.7503418e-05, -1.4799866e-06, ...,\n",
      "        -1.4377291e-06, -2.8309753e-06, -5.6359140e-06],\n",
      "       [-1.5878254e-06,  7.0485243e-05, -2.0333973e-06, ...,\n",
      "        -1.6381553e-06, -4.1681465e-06, -8.1193002e-06],\n",
      "       [-1.0560906e-06,  3.2082445e-04, -1.2567734e-06, ...,\n",
      "        -1.3702219e-06, -3.2954185e-06, -7.0018232e-06],\n",
      "       ...,\n",
      "       [ 1.0307606e-05, -1.6722738e-04,  9.7781613e-06, ...,\n",
      "         1.0504362e-05,  1.9095427e-05,  3.2226068e-05],\n",
      "       [-1.0309456e-06, -7.0306833e-06, -1.1354837e-06, ...,\n",
      "        -1.1232967e-06, -2.0530456e-06, -4.2512343e-06],\n",
      "       [ 1.5119231e-06,  3.9762491e-04,  1.9251984e-06, ...,\n",
      "         1.6721297e-06,  3.0362444e-06,  4.7681106e-06]], dtype=float32)>, <tf.Tensor: id=920773, shape=(110,), dtype=float32, numpy=\n",
      "array([ 1.08209470e-05, -1.70710002e-04,  1.01932883e-05, -1.05736544e-04,\n",
      "       -1.88150210e-03, -9.48278030e-05, -7.88424339e-04,  1.47875844e-05,\n",
      "        1.44217720e-05,  1.49006055e-05, -3.80567508e-04, -2.15180597e-04,\n",
      "       -2.08901707e-04,  5.40452165e-05,  2.14529118e-05, -9.42097045e-04,\n",
      "        6.87180902e-04, -1.23705395e-05, -3.73995426e-04, -3.87789361e-04,\n",
      "       -2.49213888e-04,  5.85186179e-04,  4.90462116e-04,  1.64510348e-04,\n",
      "        2.17603767e-04,  3.59882833e-04,  1.32506539e-04,  6.47162902e-04,\n",
      "       -4.45248123e-04,  1.63836579e-04, -2.16665994e-05, -1.23618098e-04,\n",
      "        2.21156406e-05, -1.05893581e-04, -6.97225405e-05, -2.25107046e-03,\n",
      "        1.60939177e-04, -4.05946863e-04, -1.12008757e-03, -1.85243506e-03,\n",
      "       -1.00415840e-04, -2.86091934e-04,  5.99234307e-04, -1.87539193e-03,\n",
      "       -9.51463589e-05,  2.49537406e-04, -1.16031122e-04,  7.43210665e-04,\n",
      "       -1.50088221e-03, -9.34429001e-04, -4.62821510e-04,  7.11972971e-05,\n",
      "       -5.48529148e-04, -1.38929056e-03, -4.40906675e-04,  4.47360435e-05,\n",
      "       -3.11742770e-04, -2.52003956e-04, -7.94258303e-05,  2.68802280e-04,\n",
      "        3.61283528e-05,  1.53484565e-04,  1.50173903e-04,  7.05958155e-05,\n",
      "        4.37441304e-05,  8.53149686e-03, -1.39749818e-03, -5.03241841e-04,\n",
      "       -1.90336170e-04,  5.21497475e-03,  3.45450686e-03, -1.99965923e-03,\n",
      "        4.56202775e-03,  1.12674583e-03,  2.42303795e-04,  9.96409217e-04,\n",
      "        5.29107684e-03, -2.27412721e-03, -5.86909708e-03, -1.34697533e-04,\n",
      "        3.21955304e-03,  3.11824784e-04,  2.02529971e-03,  7.80721253e-04,\n",
      "       -1.17294649e-02,  2.38390127e-03,  1.06122973e-03,  5.49798890e-04,\n",
      "       -1.74147647e-03,  1.29470835e-04,  6.60768055e-05,  1.62113738e-05,\n",
      "        1.19706192e-05,  1.92369189e-05,  1.35851942e-05,  1.26431369e-05,\n",
      "        1.42290883e-05,  1.21666308e-05,  1.27579369e-05,  1.12553671e-05,\n",
      "        1.31119650e-05,  1.30568797e-05,  1.36060180e-05,  1.39790482e-05,\n",
      "        2.10458511e-05,  1.24944954e-05,  1.37601673e-05,  1.09880302e-05,\n",
      "        1.98533726e-05,  3.34923170e-05], dtype=float32)>]\n"
     ]
    }
   ],
   "source": [
    "# Training step\n",
    "EPOCHS = 1\n",
    "\n",
    "for epoch in range(EPOCHS):\n",
    "    start = time.time()\n",
    "    \n",
    "    # initializing the hidden state at the start of every epoch\n",
    "    # initally hidden is None\n",
    "    hidden = model.reset_states()\n",
    "    \n",
    "    for (batch_n, (inp, target)) in enumerate(dataset):\n",
    "          with tf.GradientTape() as tape:\n",
    "              # feeding the hidden state back into the model\n",
    "              # This is the interesting step\n",
    "              predictions = model(inp)\n",
    "              loss = tf.losses.sparse_softmax_cross_entropy(target, predictions)\n",
    "              \n",
    "          grads = tape.gradient(loss, model.trainable_variables)\n",
    "          optimizer.apply_gradients(zip(grads, model.trainable_variables))\n",
    "\n",
    "          if batch_n % 100 == 0:\n",
    "              template = 'Epoch {} Batch {} Loss {:.4f}'\n",
    "              print(template.format(epoch+1, batch_n, loss))\n",
    "\n",
    "    # saving (checkpoint) the model every 5 epochs\n",
    "    if (epoch + 1) % 5 == 0:\n",
    "      model.save_weights(checkpoint_prefix.format(epoch=epoch))\n",
    "\n",
    "    print ('Epoch {} Loss {:.4f}'.format(epoch+1, loss))\n",
    "    print ('Time taken for 1 epoch {} sec\\n'.format(time.time() - start))\n",
    "\n",
    "model.save_weights(checkpoint_prefix.format(epoch=epoch))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "collapsed_sections": [],
   "name": "text_generation.ipynb",
   "private_outputs": true,
   "provenance": [],
   "toc_visible": true,
   "version": "0.3.2"
  },
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}