neural-network-regression.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "source": [
    "# [Module 31 Creating sample regression data](https://youtu.be/tpCFfeUEGs8?t=15459)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "\n",
    "X = np.array([-7, -4, -1, 2, 5, 8, 11, 14])  # making data to view and fit\n",
    "y = np.array([3, 6, 9, 12, 15, 18, 21, 24])  # create labels\n",
    "\n",
    "# visualize it\n",
    "plt.scatter(X, y)\n",
    "\n",
    "# we want it to learn that\n",
    "# print(X == Y - 10)\n",
    "\n",
    "# house price demo tensor\n",
    "house_info = tf.constant([\"bedroom\", \"bathroom\", \"garage\"])\n",
    "house_price = tf.constant([937_000])\n",
    "\n",
    "# X and Y are the same shape, so we want to use X[i] to predict Y[i]\n"
   ]
  },
  {
   "cell_type": "markdown",
   "source": [
    "# [Module 32 Steps in modelling with Tensorflow](https://youtu.be/tpCFfeUEGs8?t=16119)\n",
    "\n",
    "### Steps in modelling with tensorflow\n",
    "1. Creating the model: define the input and output layers, as well as the hidden layers if it's a deep learning model\n",
    "2. Compile the model:\n",
    "3. Fitting the model: let the model try to find patterns between X and y (features and labels)\n",
    "4. Evaluate the model\n",
    "\n",
    "### Compiling the model\n",
    "1. Define the loss function: function that tells our model how wrong it is\n",
    "2. Define the optimizer: function that tells our model how to improve the patterns it's learning\n",
    "3. Evaluation metrics: what we use to interpret performance of the model\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "outputs": [],
   "source": [
    "# turn x and y into tensors\n",
    "X = tf.constant(X)\n",
    "y = tf.constant(y)\n",
    "input_shape = X[0].shape\n",
    "# input_shape\n",
    "# set random seed:\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/5\n",
      "1/1 [==============================] - 0s 138ms/step - loss: 19.4446 - mae: 19.4446\n",
      "Epoch 2/5\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 19.1634 - mae: 19.1634\n",
      "Epoch 3/5\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 18.8821 - mae: 18.8821\n",
      "Epoch 4/5\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 18.6009 - mae: 18.6009\n",
      "Epoch 5/5\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 18.3196 - mae: 18.3196\n"
     ]
    },
    {
     "data": {
      "text/plain": "<keras.callbacks.History at 0x1740c64d0>"
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# tf.keras.optimizers.legacy.Adam()  # need this to work on M1 CPU. wait maybe not\n",
    "# set random seed\n",
    "tf.random.set_seed(42)\n",
    "\n",
    "X = tf.reshape(X, (-1, 1)) # I don't know why this is needed??\n",
    "\n",
    "# 1 create the model\n",
    "# use sequential API\n",
    "model = tf.keras.models.Sequential(\n",
    "    # tf.keras.layers.Input(shape=(1,)),\n",
    "    tf.keras.layers.Dense(1)  # 1 because we have 1 number as input and one as output\n",
    ")\n",
    "# equivalent way to do step 1:\n",
    "model2 = tf.keras.Sequential()\n",
    "model2.add(tf.keras.layers.Dense(1))\n",
    "\n",
    "# 2 compile the model\n",
    "model.compile(loss=tf.keras.losses.MAE,  # mean absolute error is loss fn\n",
    "              optimizer=tf.keras.optimizers.SGD(),  # stochastic gradient descent\n",
    "              metrics=[\"mae\"])\n",
    "\n",
    "# 3 fitting the model\n",
    "model.fit(X, y, epochs=5)\n",
    "\n",
    "# 4 evaluate the model"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1/1 [==============================] - 0s 30ms/step\n"
     ]
    },
    {
     "data": {
      "text/plain": "array([[29.53802]], dtype=float32)"
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# try and make a prediction using our model\n",
    "model.predict([17])  # not that good, can be improved w more epochs"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# [Module 33 Improving a model part 1](https://youtu.be/tpCFfeUEGs8?t=17333)\n",
    "(No code associated with this module)\n",
    "\n",
    "Improving step 1: create a model\n",
    "1. We might add more layers (neurons)\n",
    "2. We might change activation function within each layer\n",
    "\n",
    "Improving step 2: compile the model\n",
    "1. change optimization function\n",
    "2. change learning rate of optimization function\n",
    "\n",
    "Improving step 3: fit a model\n",
    "1. Fit model for more epochs\n",
    "2. Give more training data"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# [Module 34 Improving a model part 2](https://youtu.be/tpCFfeUEGs8?t=17696)\n",
    "Better version of module 32's model"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/100\n",
      "1/1 [==============================] - 0s 195ms/step - loss: 13.7619 - mae: 13.7619\n",
      "Epoch 2/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 13.2854 - mae: 13.2854\n",
      "Epoch 3/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 12.8858 - mae: 12.8858\n",
      "Epoch 4/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 12.4689 - mae: 12.4689\n",
      "Epoch 5/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 12.0380 - mae: 12.0380\n",
      "Epoch 6/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 11.5601 - mae: 11.5601\n",
      "Epoch 7/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 11.0134 - mae: 11.0134\n",
      "Epoch 8/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 10.3592 - mae: 10.3592\n",
      "Epoch 9/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 9.5750 - mae: 9.5750\n",
      "Epoch 10/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 8.5773 - mae: 8.5773\n",
      "Epoch 11/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 7.3262 - mae: 7.3262\n",
      "Epoch 12/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 5.8486 - mae: 5.8486\n",
      "Epoch 13/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 4.0542 - mae: 4.0542\n",
      "Epoch 14/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.9337 - mae: 3.9337\n",
      "Epoch 15/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 4.0109 - mae: 4.0109\n",
      "Epoch 16/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.9392 - mae: 3.9392\n",
      "Epoch 17/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 3.9643 - mae: 3.9643\n",
      "Epoch 18/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.9491 - mae: 3.9491\n",
      "Epoch 19/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 3.9145 - mae: 3.9145\n",
      "Epoch 20/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.9612 - mae: 3.9612\n",
      "Epoch 21/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 3.8628 - mae: 3.8628\n",
      "Epoch 22/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.9738 - mae: 3.9738\n",
      "Epoch 23/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.8095 - mae: 3.8095\n",
      "Epoch 24/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.9867 - mae: 3.9867\n",
      "Epoch 25/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.7805 - mae: 3.7805\n",
      "Epoch 26/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 3.9632 - mae: 3.9632\n",
      "Epoch 27/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 3.7957 - mae: 3.7957\n",
      "Epoch 28/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.9093 - mae: 3.9093\n",
      "Epoch 29/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.8125 - mae: 3.8125\n",
      "Epoch 30/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.8537 - mae: 3.8537\n",
      "Epoch 31/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 3.8296 - mae: 3.8296\n",
      "Epoch 32/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.7971 - mae: 3.7971\n",
      "Epoch 33/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.8474 - mae: 3.8474\n",
      "Epoch 34/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.7395 - mae: 3.7395\n",
      "Epoch 35/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.8654 - mae: 3.8654\n",
      "Epoch 36/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.6789 - mae: 3.6789\n",
      "Epoch 37/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.8868 - mae: 3.8868\n",
      "Epoch 38/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 3.6843 - mae: 3.6843\n",
      "Epoch 39/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 3.8364 - mae: 3.8364\n",
      "Epoch 40/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 3.7426 - mae: 3.7426\n",
      "Epoch 41/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.7382 - mae: 3.7382\n",
      "Epoch 42/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.7649 - mae: 3.7649\n",
      "Epoch 43/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 3.6724 - mae: 3.6724\n",
      "Epoch 44/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.7887 - mae: 3.7887\n",
      "Epoch 45/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 3.6064 - mae: 3.6064\n",
      "Epoch 46/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 3.8124 - mae: 3.8124\n",
      "Epoch 47/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 3.6153 - mae: 3.6153\n",
      "Epoch 48/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 3.7470 - mae: 3.7470\n",
      "Epoch 49/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 3.6404 - mae: 3.6404\n",
      "Epoch 50/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.6805 - mae: 3.6805\n",
      "Epoch 51/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 3.7028 - mae: 3.7028\n",
      "Epoch 52/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.5830 - mae: 3.5830\n",
      "Epoch 53/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.7299 - mae: 3.7299\n",
      "Epoch 54/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 3.5385 - mae: 3.5385\n",
      "Epoch 55/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.7164 - mae: 3.7164\n",
      "Epoch 56/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 3.5649 - mae: 3.5649\n",
      "Epoch 57/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.6435 - mae: 3.6435\n",
      "Epoch 58/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 3.5916 - mae: 3.5916\n",
      "Epoch 59/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.5683 - mae: 3.5683\n",
      "Epoch 60/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.6212 - mae: 3.6212\n",
      "Epoch 61/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.4909 - mae: 3.4909\n",
      "Epoch 62/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.6525 - mae: 3.6525\n",
      "Epoch 63/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 3.4657 - mae: 3.4657\n",
      "Epoch 64/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.6414 - mae: 3.6414\n",
      "Epoch 65/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.5291 - mae: 3.5291\n",
      "Epoch 66/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.5146 - mae: 3.5146\n",
      "Epoch 67/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.5603 - mae: 3.5603\n",
      "Epoch 68/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.4326 - mae: 3.4326\n",
      "Epoch 69/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.5927 - mae: 3.5927\n",
      "Epoch 70/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.4089 - mae: 3.4089\n",
      "Epoch 71/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.5529 - mae: 3.5529\n",
      "Epoch 72/100\n",
      "1/1 [==============================] - 0s 4ms/step - loss: 3.4411 - mae: 3.4411\n",
      "Epoch 73/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.4684 - mae: 3.4684\n",
      "Epoch 74/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 3.4752 - mae: 3.4752\n",
      "Epoch 75/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.3815 - mae: 3.3815\n",
      "Epoch 76/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 3.5108 - mae: 3.5108\n",
      "Epoch 77/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.3398 - mae: 3.3398\n",
      "Epoch 78/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.4697 - mae: 3.4697\n",
      "Epoch 79/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 3.4004 - mae: 3.4004\n",
      "Epoch 80/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.3799 - mae: 3.3799\n",
      "Epoch 81/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.4360 - mae: 3.4360\n",
      "Epoch 82/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 3.2869 - mae: 3.2869\n",
      "Epoch 83/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.4735 - mae: 3.4735\n",
      "Epoch 84/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.2949 - mae: 3.2949\n",
      "Epoch 85/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 3.3957 - mae: 3.3957\n",
      "Epoch 86/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.3319 - mae: 3.3319\n",
      "Epoch 87/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 3.2999 - mae: 3.2999\n",
      "Epoch 88/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.3701 - mae: 3.3701\n",
      "Epoch 89/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 3.2058 - mae: 3.2058\n",
      "Epoch 90/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 3.4431 - mae: 3.4431\n",
      "Epoch 91/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.2633 - mae: 3.2633\n",
      "Epoch 92/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.2820 - mae: 3.2820\n",
      "Epoch 93/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.3001 - mae: 3.3001\n",
      "Epoch 94/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.1781 - mae: 3.1781\n",
      "Epoch 95/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.3394 - mae: 3.3394\n",
      "Epoch 96/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 3.1677 - mae: 3.1677\n",
      "Epoch 97/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.2725 - mae: 3.2725\n",
      "Epoch 98/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 3.2069 - mae: 3.2069\n",
      "Epoch 99/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 3.1663 - mae: 3.1663\n",
      "Epoch 100/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 3.2491 - mae: 3.2491\n",
      "1/1 [==============================] - 0s 35ms/step\n"
     ]
    },
    {
     "data": {
      "text/plain": "array([[31.208849]], dtype=float32)"
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "\n",
    "# 1 create the model\n",
    "model = tf.keras.models.Sequential([\n",
    "    tf.keras.layers.Dense(100, activation=\"relu\"),  # 3 hidden layers. 100 neurons per hidden layer\n",
    "    tf.keras.layers.Dense(100, activation=\"relu\"),\n",
    "    tf.keras.layers.Dense(100, activation=\"relu\"),\n",
    "    tf.keras.layers.Dense(1)  # 1 because we have 1 number as input and one as output\n",
    "])\n",
    "\n",
    "# 2 compile the model\n",
    "model.compile(loss=tf.keras.losses.mae,  # mean absolute error is loss fn\n",
    "              optimizer=tf.keras.optimizers.SGD(),  # could use adam instead of SGD with Adam(learning_rate=0.0001). idk why we'd chose such a low rate\n",
    "              metrics=[\"mae\"])\n",
    "\n",
    "# 3 fitting the model\n",
    "model.fit(X, y, epochs=100)\n",
    "\n",
    "# make prediction\n",
    "model.predict([17])"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# [Module 35 Improving a model part 3](https://youtu.be/tpCFfeUEGs8?t=18262)\n",
    "Changing the same model again. It's better to start with smaller changes.  It's also better to start with less complicated models which are fast to run and then tweak to make more complex (like adding more layers or epochs).\n",
    "\n",
    "Not every tweak to make it more complex helps. We might overfit by accident\n",
    "\n",
    "### Common ways to improve deep learning model:\n",
    "1. Adding layers\n",
    "2. Increase number of hidden units (neurons per layer)\n",
    "3. Change activation function (like \"relu\")\n",
    "4. Change optimization function (like SGD() or Adam())\n",
    "5. Change learning rate (tends to be the most important hyperparameter)\n",
    "6. Train on more data\n",
    "7. Fit for longer"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/100\n",
      "1/1 [==============================] - 0s 191ms/step - loss: 13.2601 - mae: 13.2601\n",
      "Epoch 2/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 13.2243 - mae: 13.2243\n",
      "Epoch 3/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 13.1884 - mae: 13.1884\n",
      "Epoch 4/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 13.1504 - mae: 13.1504\n",
      "Epoch 5/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 13.1124 - mae: 13.1124\n",
      "Epoch 6/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 13.0761 - mae: 13.0761\n",
      "Epoch 7/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 13.0402 - mae: 13.0402\n",
      "Epoch 8/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 13.0043 - mae: 13.0043\n",
      "Epoch 9/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 12.9682 - mae: 12.9682\n",
      "Epoch 10/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 12.9320 - mae: 12.9320\n",
      "Epoch 11/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 12.8955 - mae: 12.8955\n",
      "Epoch 12/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 12.8588 - mae: 12.8588\n",
      "Epoch 13/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 12.8219 - mae: 12.8219\n",
      "Epoch 14/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 12.7843 - mae: 12.7843\n",
      "Epoch 15/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 12.7461 - mae: 12.7461\n",
      "Epoch 16/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 12.7075 - mae: 12.7075\n",
      "Epoch 17/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 12.6687 - mae: 12.6687\n",
      "Epoch 18/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 12.6302 - mae: 12.6302\n",
      "Epoch 19/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 12.5919 - mae: 12.5919\n",
      "Epoch 20/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 12.5532 - mae: 12.5532\n",
      "Epoch 21/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 12.5141 - mae: 12.5141\n",
      "Epoch 22/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 12.4746 - mae: 12.4746\n",
      "Epoch 23/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 12.4350 - mae: 12.4350\n",
      "Epoch 24/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 12.3950 - mae: 12.3950\n",
      "Epoch 25/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 12.3546 - mae: 12.3546\n",
      "Epoch 26/100\n",
      "1/1 [==============================] - 0s 4ms/step - loss: 12.3133 - mae: 12.3133\n",
      "Epoch 27/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 12.2699 - mae: 12.2699\n",
      "Epoch 28/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 12.2260 - mae: 12.2260\n",
      "Epoch 29/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 12.1814 - mae: 12.1814\n",
      "Epoch 30/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 12.1365 - mae: 12.1365\n",
      "Epoch 31/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 12.0911 - mae: 12.0911\n",
      "Epoch 32/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 12.0450 - mae: 12.0450\n",
      "Epoch 33/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 11.9987 - mae: 11.9987\n",
      "Epoch 34/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 11.9518 - mae: 11.9518\n",
      "Epoch 35/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 11.9050 - mae: 11.9050\n",
      "Epoch 36/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 11.8575 - mae: 11.8575\n",
      "Epoch 37/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 11.8096 - mae: 11.8096\n",
      "Epoch 38/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 11.7614 - mae: 11.7614\n",
      "Epoch 39/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 11.7128 - mae: 11.7128\n",
      "Epoch 40/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 11.6634 - mae: 11.6634\n",
      "Epoch 41/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 11.6133 - mae: 11.6133\n",
      "Epoch 42/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 11.5624 - mae: 11.5624\n",
      "Epoch 43/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 11.5111 - mae: 11.5111\n",
      "Epoch 44/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 11.4597 - mae: 11.4597\n",
      "Epoch 45/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 11.4078 - mae: 11.4078\n",
      "Epoch 46/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 11.3566 - mae: 11.3566\n",
      "Epoch 47/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 11.3047 - mae: 11.3047\n",
      "Epoch 48/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 11.2518 - mae: 11.2518\n",
      "Epoch 49/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 11.1979 - mae: 11.1979\n",
      "Epoch 50/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 11.1430 - mae: 11.1430\n",
      "Epoch 51/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 11.0871 - mae: 11.0871\n",
      "Epoch 52/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 11.0302 - mae: 11.0302\n",
      "Epoch 53/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 10.9720 - mae: 10.9720\n",
      "Epoch 54/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 10.9127 - mae: 10.9127\n",
      "Epoch 55/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 10.8523 - mae: 10.8523\n",
      "Epoch 56/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 10.7909 - mae: 10.7909\n",
      "Epoch 57/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 10.7285 - mae: 10.7285\n",
      "Epoch 58/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 10.6647 - mae: 10.6647\n",
      "Epoch 59/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 10.5994 - mae: 10.5994\n",
      "Epoch 60/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 10.5328 - mae: 10.5328\n",
      "Epoch 61/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 10.4646 - mae: 10.4646\n",
      "Epoch 62/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 10.3947 - mae: 10.3947\n",
      "Epoch 63/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 10.3232 - mae: 10.3232\n",
      "Epoch 64/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 10.2500 - mae: 10.2500\n",
      "Epoch 65/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 10.1748 - mae: 10.1748\n",
      "Epoch 66/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 10.0978 - mae: 10.0978\n",
      "Epoch 67/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 10.0189 - mae: 10.0189\n",
      "Epoch 68/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 9.9376 - mae: 9.9376\n",
      "Epoch 69/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 9.8542 - mae: 9.8542\n",
      "Epoch 70/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 9.7687 - mae: 9.7687\n",
      "Epoch 71/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 9.6811 - mae: 9.6811\n",
      "Epoch 72/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 9.5911 - mae: 9.5911\n",
      "Epoch 73/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 9.4988 - mae: 9.4988\n",
      "Epoch 74/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 9.4042 - mae: 9.4042\n",
      "Epoch 75/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 9.3071 - mae: 9.3071\n",
      "Epoch 76/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 9.2074 - mae: 9.2074\n",
      "Epoch 77/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 9.1046 - mae: 9.1046\n",
      "Epoch 78/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 8.9990 - mae: 8.9990\n",
      "Epoch 79/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 8.8903 - mae: 8.8903\n",
      "Epoch 80/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 8.7782 - mae: 8.7782\n",
      "Epoch 81/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 8.6628 - mae: 8.6628\n",
      "Epoch 82/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 8.5443 - mae: 8.5443\n",
      "Epoch 83/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 8.4219 - mae: 8.4219\n",
      "Epoch 84/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 8.2959 - mae: 8.2959\n",
      "Epoch 85/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 8.1659 - mae: 8.1659\n",
      "Epoch 86/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 8.0303 - mae: 8.0303\n",
      "Epoch 87/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 7.8894 - mae: 7.8894\n",
      "Epoch 88/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 7.7441 - mae: 7.7441\n",
      "Epoch 89/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 7.5938 - mae: 7.5938\n",
      "Epoch 90/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 7.4389 - mae: 7.4389\n",
      "Epoch 91/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 7.2786 - mae: 7.2786\n",
      "Epoch 92/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 7.1144 - mae: 7.1144\n",
      "Epoch 93/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 6.9647 - mae: 6.9647\n",
      "Epoch 94/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 6.8098 - mae: 6.8098\n",
      "Epoch 95/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 6.6498 - mae: 6.6498\n",
      "Epoch 96/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 6.4841 - mae: 6.4841\n",
      "Epoch 97/100\n",
      "1/1 [==============================] - 0s 3ms/step - loss: 6.3125 - mae: 6.3125\n",
      "Epoch 98/100\n",
      "1/1 [==============================] - 0s 4ms/step - loss: 6.1343 - mae: 6.1343\n",
      "Epoch 99/100\n",
      "1/1 [==============================] - 0s 2ms/step - loss: 5.9496 - mae: 5.9496\n",
      "Epoch 100/100\n",
      "1/1 [==============================] - 0s 1ms/step - loss: 5.7579 - mae: 5.7579\n",
      "1/1 [==============================] - 0s 33ms/step\n"
     ]
    },
    {
     "data": {
      "text/plain": "array([[24.329931]], dtype=float32)"
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 1 create the model\n",
    "model = tf.keras.models.Sequential([\n",
    "    tf.keras.layers.Dense(100, activation=\"relu\"),  # 3 hidden layers. 100 neurons per hidden layer\n",
    "    tf.keras.layers.Dense(100, activation=\"relu\"),\n",
    "    tf.keras.layers.Dense(100, activation=\"relu\"),\n",
    "    tf.keras.layers.Dense(1)  # 1 because we have 1 number as input and one as output\n",
    "])\n",
    "\n",
    "# 2 compile the model\n",
    "model.compile(loss=tf.keras.losses.mae,  # mean absolute error is loss fn\n",
    "              optimizer=tf.keras.optimizers.SGD(learning_rate=0.001),  # could use adam instead of SGD with Adam(learning_rate=0.0001). idk why we'd chose such a low rate\n",
    "              metrics=[\"mae\"])\n",
    "\n",
    "# 3 fitting the model\n",
    "model.fit(X, y, epochs=100)\n",
    "\n",
    "# make prediction\n",
    "model.predict([17])\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# [Module 36 evaluating a model part 1: visualizations](https://youtu.be/tpCFfeUEGs8?t=19015)\n",
    "\n",
    "### Typical workflow when building neural networks:\n",
    "1. Build model\n",
    "2. Fit it\n",
    "3. Evaluate it\n",
    "4. Tweak it\n",
    "5. Repeat\n",
    "\n",
    "### What should we visualize when evaluating:\n",
    "- The data we're working with\n",
    "- The model itself (what does the model look like)\n",
    "- The training of a model (how does the model perform while it learns)\n",
    "- The predictions of the model, and how do they line up with the real answer"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "outputs": [
    {
     "data": {
      "text/plain": "<matplotlib.collections.PathCollection at 0x166094460>"
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# make larger dataset\n",
    "X = tf.range(-100, 100, 4)\n",
    "y = X + 10\n",
    "\n",
    "plt.scatter(X, y)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "\n",
    "# [Module 37 evaluating a model part 2: the datasets](https://youtu.be/tpCFfeUEGs8?t=19460)\n",
    "\n",
    "### The sets of data and how we split it up\n",
    "1. Training dataset 70-80%\n",
    "2. Validation dataset (where we tweak hyperparameters) 10-15%  (sometimes optional)\n",
    "3. Test set"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# we'll skip the validation set for now\n",
    "\n",
    "# train test split:\n",
    "X_train, X_test = X[:40], X[40:]\n",
    "y_train, y_test = y[:40], y[40:]\n",
    "\n",
    "# visualizing the data\n",
    "plt.scatter(X_train, y_train, c=\"blue\", label=\"Training data\")  # plot training data in blue\n",
    "plt.scatter(X_test, y_test, c=\"green\", label=\"Testing data\") # plot training data in green\n",
    "plt.legend()\n",
    "plt.show()"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# [Module 38 evaluating a model part 3: model summary](https://youtu.be/tpCFfeUEGs8?t=20122)\n",
    "let's build a neural network for our data"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "outputs": [],
   "source": [
    "tf.random.set_seed(42)\n",
    "# 1 create a model\n",
    "model = tf.keras.Sequential([\n",
    "    tf.keras.layers.Dense(10, input_shape=[1], name=\"input_layer\"),\n",
    "    # tf.keras.layers.Dense(100, input_shape=[1], name=\"uh\"),\n",
    "    tf.keras.layers.Dense(1, input_shape=[1], name=\"output_layer\")  # input shape is 1 because X is one input to give one output\n",
    "], name=\"module_38_model\")\n",
    "# 2 compile\n",
    "model.compile(loss=tf.keras.losses.mae,\n",
    "              optimizer=tf.keras.optimizers.SGD(),\n",
    "              metrics=[\"mae\"])"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"module_38_model\"\n",
      "_________________________________________________________________\n",
      " Layer (type)                Output Shape              Param #   \n",
      "=================================================================\n",
      " input_layer (Dense)         (None, 10)                20        \n",
      "                                                                 \n",
      " output_layer (Dense)        (None, 1)                 11        \n",
      "                                                                 \n",
      "=================================================================\n",
      "Total params: 31\n",
      "Trainable params: 31\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "# visualize model before we fit it. we need to determine input_shape\n",
    "model.build()\n",
    "model.summary()"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "### model.summary() explanation\n",
    "\n",
    "- total params: total number of parameters in the model. this one is 2 because it's learning relationship between X and y\n",
    "- trainable parameters:\n",
    "- non-trainable params aren't updated during training. Used in already learned patterns/parameters from already learned models you import in tansfer learning"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/100\n",
      "2/2 [==============================] - 0s 2ms/step - loss: 17.5051 - mae: 17.5051\n",
      "Epoch 2/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 27.6223 - mae: 27.6223\n",
      "Epoch 3/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 31.2000 - mae: 31.2000\n",
      "Epoch 4/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 16.8919 - mae: 16.8919\n",
      "Epoch 5/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 23.0515 - mae: 23.0515\n",
      "Epoch 6/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 10.9046 - mae: 10.9046\n",
      "Epoch 7/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 11.9422 - mae: 11.9422\n",
      "Epoch 8/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 13.0135 - mae: 13.0135\n",
      "Epoch 9/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 34.3622 - mae: 34.3622\n",
      "Epoch 10/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 23.7757 - mae: 23.7757\n",
      "Epoch 11/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 11.8615 - mae: 11.8615\n",
      "Epoch 12/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 23.8416 - mae: 23.8416\n",
      "Epoch 13/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 20.7837 - mae: 20.7837\n",
      "Epoch 14/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 23.1549 - mae: 23.1549\n",
      "Epoch 15/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 15.0742 - mae: 15.0742\n",
      "Epoch 16/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 11.2737 - mae: 11.2737\n",
      "Epoch 17/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 22.9148 - mae: 22.9148\n",
      "Epoch 18/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 11.3152 - mae: 11.3152\n",
      "Epoch 19/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 13.8600 - mae: 13.8600\n",
      "Epoch 20/100\n",
      "2/2 [==============================] - 0s 2ms/step - loss: 11.1663 - mae: 11.1663\n",
      "Epoch 21/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 17.1716 - mae: 17.1716\n",
      "Epoch 22/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 15.4105 - mae: 15.4105\n",
      "Epoch 23/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 9.2505 - mae: 9.2505\n",
      "Epoch 24/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 17.3088 - mae: 17.3088\n",
      "Epoch 25/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 15.9708 - mae: 15.9708\n",
      "Epoch 26/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 21.0029 - mae: 21.0029\n",
      "Epoch 27/100\n",
      "2/2 [==============================] - 0s 2ms/step - loss: 25.8029 - mae: 25.8029\n",
      "Epoch 28/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 18.3225 - mae: 18.3225\n",
      "Epoch 29/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 9.2413 - mae: 9.2413\n",
      "Epoch 30/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 28.9263 - mae: 28.9263\n",
      "Epoch 31/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 52.3736 - mae: 52.3736\n",
      "Epoch 32/100\n",
      "2/2 [==============================] - 0s 2ms/step - loss: 11.8662 - mae: 11.8662\n",
      "Epoch 33/100\n",
      "2/2 [==============================] - 0s 2ms/step - loss: 15.4159 - mae: 15.4159\n",
      "Epoch 34/100\n",
      "2/2 [==============================] - 0s 2ms/step - loss: 12.5412 - mae: 12.5412\n",
      "Epoch 35/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 9.1728 - mae: 9.1728\n",
      "Epoch 36/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 16.3763 - mae: 16.3763\n",
      "Epoch 37/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 11.1140 - mae: 11.1140\n",
      "Epoch 38/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 18.2458 - mae: 18.2458\n",
      "Epoch 39/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 19.1744 - mae: 19.1744\n",
      "Epoch 40/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 20.5645 - mae: 20.5645\n",
      "Epoch 41/100\n",
      "2/2 [==============================] - 0s 2ms/step - loss: 14.7459 - mae: 14.7459\n",
      "Epoch 42/100\n",
      "2/2 [==============================] - 0s 2ms/step - loss: 12.1573 - mae: 12.1573\n",
      "Epoch 43/100\n",
      "2/2 [==============================] - 0s 2ms/step - loss: 10.6768 - mae: 10.6768\n",
      "Epoch 44/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 32.1084 - mae: 32.1084\n",
      "Epoch 45/100\n",
      "2/2 [==============================] - 0s 2ms/step - loss: 12.4261 - mae: 12.4261\n",
      "Epoch 46/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 17.3932 - mae: 17.3932\n",
      "Epoch 47/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 15.8017 - mae: 15.8017\n",
      "Epoch 48/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 8.4036 - mae: 8.4036\n",
      "Epoch 49/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 14.2579 - mae: 14.2579\n",
      "Epoch 50/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 15.1884 - mae: 15.1884\n",
      "Epoch 51/100\n",
      "2/2 [==============================] - 0s 2ms/step - loss: 13.7005 - mae: 13.7005\n",
      "Epoch 52/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 18.1140 - mae: 18.1140\n",
      "Epoch 53/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 22.7822 - mae: 22.7822\n",
      "Epoch 54/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 26.7741 - mae: 26.7741\n",
      "Epoch 55/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 26.4047 - mae: 26.4047\n",
      "Epoch 56/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 11.3511 - mae: 11.3511\n",
      "Epoch 57/100\n",
      "2/2 [==============================] - 0s 2ms/step - loss: 13.3706 - mae: 13.3706\n",
      "Epoch 58/100\n",
      "2/2 [==============================] - 0s 2ms/step - loss: 9.9555 - mae: 9.9555\n",
      "Epoch 59/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 13.9218 - mae: 13.9218\n",
      "Epoch 60/100\n",
      "2/2 [==============================] - 0s 973us/step - loss: 9.9972 - mae: 9.9972\n",
      "Epoch 61/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 14.9759 - mae: 14.9759\n",
      "Epoch 62/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 11.9311 - mae: 11.9311\n",
      "Epoch 63/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 10.2920 - mae: 10.2920\n",
      "Epoch 64/100\n",
      "2/2 [==============================] - 0s 2ms/step - loss: 23.8846 - mae: 23.8846\n",
      "Epoch 65/100\n",
      "2/2 [==============================] - 0s 2ms/step - loss: 10.4684 - mae: 10.4684\n",
      "Epoch 66/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 20.9299 - mae: 20.9299\n",
      "Epoch 67/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 10.4453 - mae: 10.4453\n",
      "Epoch 68/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 14.1742 - mae: 14.1742\n",
      "Epoch 69/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 10.5150 - mae: 10.5150\n",
      "Epoch 70/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 12.5858 - mae: 12.5858\n",
      "Epoch 71/100\n",
      "2/2 [==============================] - 0s 2ms/step - loss: 13.0143 - mae: 13.0143\n",
      "Epoch 72/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 19.3500 - mae: 19.3500\n",
      "Epoch 73/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 11.1303 - mae: 11.1303\n",
      "Epoch 74/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 21.4324 - mae: 21.4324\n",
      "Epoch 75/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 9.4596 - mae: 9.4596\n",
      "Epoch 76/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 12.2819 - mae: 12.2819\n",
      "Epoch 77/100\n",
      "2/2 [==============================] - 0s 959us/step - loss: 16.3099 - mae: 16.3099\n",
      "Epoch 78/100\n",
      "2/2 [==============================] - 0s 2ms/step - loss: 9.0113 - mae: 9.0113\n",
      "Epoch 79/100\n",
      "2/2 [==============================] - 0s 2ms/step - loss: 23.5516 - mae: 23.5516\n",
      "Epoch 80/100\n",
      "2/2 [==============================] - 0s 2ms/step - loss: 26.2447 - mae: 26.2447\n",
      "Epoch 81/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 11.4365 - mae: 11.4365\n",
      "Epoch 82/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 12.4492 - mae: 12.4492\n",
      "Epoch 83/100\n",
      "2/2 [==============================] - 0s 2ms/step - loss: 17.1182 - mae: 17.1182\n",
      "Epoch 84/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 7.2643 - mae: 7.2643\n",
      "Epoch 85/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 37.1417 - mae: 37.1417\n",
      "Epoch 86/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 21.1697 - mae: 21.1697\n",
      "Epoch 87/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 11.0191 - mae: 11.0191\n",
      "Epoch 88/100\n",
      "2/2 [==============================] - 0s 2ms/step - loss: 24.8966 - mae: 24.8966\n",
      "Epoch 89/100\n",
      "2/2 [==============================] - 0s 2ms/step - loss: 9.6612 - mae: 9.6612\n",
      "Epoch 90/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 18.8951 - mae: 18.8951\n",
      "Epoch 91/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 10.5124 - mae: 10.5124\n",
      "Epoch 92/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 18.1314 - mae: 18.1314\n",
      "Epoch 93/100\n",
      "2/2 [==============================] - 0s 956us/step - loss: 6.5523 - mae: 6.5523\n",
      "Epoch 94/100\n",
      "2/2 [==============================] - 0s 2ms/step - loss: 11.1691 - mae: 11.1691\n",
      "Epoch 95/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 24.5608 - mae: 24.5608\n",
      "Epoch 96/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 10.7780 - mae: 10.7780\n",
      "Epoch 97/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 15.4852 - mae: 15.4852\n",
      "Epoch 98/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 9.0458 - mae: 9.0458\n",
      "Epoch 99/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 10.8647 - mae: 10.8647\n",
      "Epoch 100/100\n",
      "2/2 [==============================] - 0s 1ms/step - loss: 27.0731 - mae: 27.0731\n"
     ]
    },
    {
     "data": {
      "text/plain": "<keras.callbacks.History at 0x1788fb670>"
     },
     "execution_count": 61,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 3 fit model\n",
    "model.fit(X_train, y_train, epochs=100) # verbose=0 to leave out progress bars"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# [Module 39 evaluating a model part 4: visualizing layers](https://youtu.be/tpCFfeUEGs8?t=21159)\n",
    "\n",
    "This doesn't work at all. I don't know why. I wonder if it's a problem with my M1 computer. installing pydot and graphviz didn't help"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "3\n",
      "You must install pydot (`pip install pydot`) and install graphviz (see instructions at https://graphviz.gitlab.io/download/) for plot_model to work.\n"
     ]
    }
   ],
   "source": [
    "# from tensorflow.keras.utils import plot_model\n",
    "# plot_model(model=model)\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    " # [Module 40 evaluating a model part 5: visualizing predictions](https://youtu.be/tpCFfeUEGs8?t=21596)\n",
    "\n",
    "Plot predictions against ground truth labels. Usually y_test vs y_pred"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1/1 [==============================] - 0s 26ms/step\n",
      "tf.Tensor([ 70  74  78  82  86  90  94  98 102 106], shape=(10,), dtype=int32)\n"
     ]
    }
   ],
   "source": [
    "# make predictions\n",
    "predications = model.predict(X_test)\n",
    "print(y_test)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "outputs": [],
   "source": [
    "# make plotting function we can use going forward\n",
    "def plot_predictions(train_data, train_labels, test_data, test_labels, predictions):\n",
    "    plt.figure(figsize=(10, 7))\n",
    "    plt.scatter(train_data, train_labels, c=\"blue\", label=\"Training Data\")\n",
    "    plt.scatter(test_data, test_labels, c=\"green\", label=\"Testing Data\")\n",
    "    plt.scatter(test_data, predictions, c=\"red\", label=\"Predictions\")\n",
    "    plt.legend()\n",
    "    plt.show()\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "outputs": [],
   "source": [
    "plot_predictions(X_train, y_train, X_test, y_test, predications)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    " # [Module 41 evaluating a model part 6: regression evaluation metrics](https://youtu.be/tpCFfeUEGs8?t=21596)\n",
    "\n",
    "### Evaluating metrics depends on the type of problem we're dealing with\n",
    "Main ones for regression:\n",
    "1. MAE: on average, how wrong are my model's predictions. Generally appropriate for most things\n",
    "2. MSE: mean square errors. Appropriate when larger errors are more significant than smaller ones.\n",
    "3. Huber: (tf.keras.losses.Huber()). Combination of MAE and MSE. Less sensitive to outliers than MSE"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# [Module 42 evaluating a model part 7: MAE as regression evaluation metric](https://youtu.be/tpCFfeUEGs8?t=22639)\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1/1 [==============================] - 0s 10ms/step - loss: 10.0433 - mae: 10.0433\n",
      "[10.043340682983398, 10.043340682983398]\n",
      "(10,) (10, 1)\n"
     ]
    },
    {
     "data": {
      "text/plain": "<tf.Tensor: shape=(), dtype=float32, numpy=108.140625>"
     },
     "execution_count": 69,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# evaluate the model on the test set\n",
    "print(model.evaluate(X_test, y_test)) # gives loss, evalueation metric\n",
    "tf.metrics.mean_absolute_error(y_test, y_pred=predications) # gives us accuracy for each prediction if we have mismatched shapes\n",
    "\n",
    "print(y_test.shape, predications.shape)\n",
    "\n",
    "# get rid of extra dimension on predictions with squeeze\n",
    "y_pred = tf.squeeze(predications)\n",
    "\n",
    "tf.metrics.mean_absolute_error(y_test, y_pred) # now we have the overall\n",
    "\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# [Module 43 evaluating a model part 8: MSE as regression evaluation metric](https://youtu.be/tpCFfeUEGs8?t=22990)\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "outputs": [],
   "source": [
    "print(tf.metrics.mean_squared_error(y_true=y_test, y_pred=y_pred)) # errors tend to be larger on MSE than MAE (obviously)\n",
    "\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "outputs": [],
   "source": [
    "# functions for MAE and MSE\n",
    "def mae(y_true, y_pred):\n",
    "    return tf.metrics.mean_absolute_error(y_true=y_true, y_pred=tf.squeeze(y_pred)).numpy()\n",
    "\n",
    "def mse(y_true, y_pred):\n",
    "    return tf.metrics.mean_squared_error(y_true=y_true, y_pred=tf.squeeze(y_pred)).numpy()"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# [Module 44: Modelling experiments part 1 (start simple)](https://youtu.be/tpCFfeUEGs8?t=23189)\n",
    "\n",
    "### Way to try to improve model\n",
    "1. Get more data\n",
    "2. Make model more complex (more hidden layers)\n",
    "3. Train for longer: give model more of a chance to find patterns in data\n",
    "\n",
    "### Let's do 3 experiments\n",
    "1. 1 layer trained for 100 epochs\n",
    "2. 2 layers trained for 100 epochs\n",
    "3. 2 layers trainer for 500 epochs"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "outputs": [
    {
     "data": {
      "text/plain": "<keras.callbacks.History at 0x1699d75b0>"
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# model1\n",
    "tf.random.set_seed(42)\n",
    "# print(X_train)\n",
    "# print(y_train)\n",
    "# build model\n",
    "model1 = tf.keras.models.Sequential([\n",
    "    tf.keras.Input(shape=(1,)),\n",
    "    tf.keras.layers.Dense(1)\n",
    "])\n",
    "\n",
    "# compile model\n",
    "model1.compile(loss=tf.keras.losses.mae,\n",
    "               optimizer=tf.keras.optimizers.SGD(),\n",
    "               metrics=[\"mae\"])\n",
    "\n",
    "# fit the model\n",
    "model1.fit(X_train, y_train, epochs=100, verbose=0)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1/1 [==============================] - 0s 22ms/step\n"
     ]
    },
    {
     "data": {
      "text/plain": "<Figure size 1000x700 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "8.599912\n",
      "80.46549\n"
     ]
    }
   ],
   "source": [
    "# make and plot predictions:\n",
    "y_pred1 = model1.predict(X_test)\n",
    "plot_predictions(X_train, y_train, X_test, y_test, y_pred1)\n",
    "\n",
    "# evaluation metrics\n",
    "mae_1 = mae(y_test, y_pred1)\n",
    "mse_1 = mse(y_test, y_pred1)\n",
    "print(mae_1)\n",
    "print(mse_1)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# [Module 45: Modelling experiments part 2 (more complex)](https://youtu.be/tpCFfeUEGs8?t=24019)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "outputs": [
    {
     "data": {
      "text/plain": "<keras.callbacks.History at 0x16a7846a0>"
     },
     "execution_count": 46,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# model2\n",
    "tf.random.set_seed(42)\n",
    "# print(X_train)\n",
    "# print(y_train)\n",
    "# build model\n",
    "model2 = tf.keras.models.Sequential([\n",
    "    tf.keras.Input(shape=(1,)),\n",
    "    tf.keras.layers.Dense(10),\n",
    "    tf.keras.layers.Dense(1)\n",
    "])\n",
    "\n",
    "# compile model\n",
    "model2.compile(loss=tf.keras.losses.mae,\n",
    "               optimizer=tf.keras.optimizers.SGD(),\n",
    "               metrics=[\"mae\"])\n",
    "\n",
    "# fit the model\n",
    "model2.fit(X_train, y_train, epochs=100, verbose=0)\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1/1 [==============================] - 0s 24ms/step\n"
     ]
    },
    {
     "data": {
      "text/plain": "<Figure size 1000x700 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2.9850974\n",
      "11.6633835\n"
     ]
    }
   ],
   "source": [
    "y_pred2 = model2.predict(X_test)\n",
    "plot_predictions(X_train, y_train, X_test, y_test, y_pred2)\n",
    "\n",
    "# evaluation metrics\n",
    "mae_2 = mae(y_test, y_pred2)\n",
    "mse_2 = mse(y_test, y_pred2)\n",
    "print(mae_2)\n",
    "print(mse_2)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "outputs": [
    {
     "data": {
      "text/plain": "<keras.callbacks.History at 0x16a1c96c0>"
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# model3. probably overfitting\n",
    "tf.random.set_seed(42)\n",
    "# print(X_train)\n",
    "# print(y_train)\n",
    "# build model\n",
    "model3 = tf.keras.models.Sequential([\n",
    "    tf.keras.Input(shape=(1,)),\n",
    "    tf.keras.layers.Dense(10),\n",
    "    tf.keras.layers.Dense(1)\n",
    "])\n",
    "\n",
    "# compile model\n",
    "model3.compile(loss=tf.keras.losses.mae,\n",
    "               optimizer=tf.keras.optimizers.SGD(),\n",
    "               metrics=[\"mae\"])\n",
    "\n",
    "# fit the model\n",
    "model3.fit(X_train, y_train, epochs=500, verbose=0)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1/1 [==============================] - 0s 25ms/step\n"
     ]
    },
    {
     "data": {
      "text/plain": "<Figure size 1000x700 with 1 Axes>",
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA0EAAAJGCAYAAACdj47VAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjcuMCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy88F64QAAAACXBIWXMAAA9hAAAPYQGoP6dpAABj7ElEQVR4nO3de3xU1b3///fOABEkkwgCGTIDwRvgrSpaflBHw1cqqNXoGLViq7QWq0VLRFvlnNbr6YF6a9Bja60aPK1WMd211KNYQNDRorUoWm2gosHEYYBWJQEvASb798d2pplMwkySuc/r+XjkgXvtNZOVcdfmzVrrswzLsiwBAAAAQIEoyvQAAAAAACCdCEEAAAAACgohCAAAAEBBIQQBAAAAKCiEIAAAAAAFhRAEAAAAoKAQggAAAAAUlAGZHkB/dXR0aMuWLSopKZFhGJkeDgAAAIAMsSxLO3fu1OjRo1VU1PN8T86HoC1btsjj8WR6GAAAAACyREtLi9xud4/3cz4ElZSUSLJ/UKfTmeHRAAAAAMiUtrY2eTyeSEboSc6HoPASOKfTSQgCAAAAEHebDIURAAAAABQUQhAAAACAgkIIAgAAAFBQcn5PUKJCoZD27NmT6WEgBwwcOFAOhyPTwwAAAECK5H0IsixLW7du1Y4dOzI9FOSQsrIylZeXc/YUAABAHsr7EBQOQCNHjtSQIUP4pRb7ZFmWPv30U23fvl2S5HK5MjwiAAAAJFteh6BQKBQJQMOHD8/0cJAjBg8eLEnavn27Ro4cydI4AACAPJPXhRHCe4CGDBmS4ZEg14SfGfaRAQAA5J+8DkFhLIFDb/HMAAAA5K+CCEEAAAAAEEYIKhCVlZWqq6tLuP+aNWtkGAZV9QAAAJB3CEFZxjCMfX7ddNNNfXrfV199VZdddlnC/adOnapgMKjS0tI+fb9EhcOWYRgqKipSaWmpjj32WP3whz9UMBjs9fsZhqEnn3wy+QMFAABA3sjr6nDJEgpJfr8UDEoul+T1SqkqGNb5F//HH39cN9xwgzZu3BhpGzp0aOSfLctSKBTSgAHx/zWOGDGiV+MYNGiQysvLe/Wa/ti4caOcTqfa2tr02muv6bbbbtODDz6oNWvW6KijjkrbOAAAAJD/mAmKwzSlykpp2jRp1iz7z8pKuz0VysvLI1+lpaUyDCNyvWHDBpWUlOiZZ57RpEmTVFxcrBdffFHvvvuuqqurNWrUKA0dOlQnnHCCVq5cGfW+XZfDGYahBx54QOecc46GDBmiQw89VMuWLYvc77ocbsmSJSorK9Ozzz6riRMnaujQoZo5c2ZUaNu7d6++//3vq6ysTMOHD9d1112nSy65RGeffXbcn3vkyJEqLy/XYYcdpq9//et66aWXNGLECF1xxRWRPq+++qq++tWv6sADD1RpaalOPvlkvfbaa1E/oySdc845Mgwjcp3I5wMAAIDCQQjaB9OUamqkDz6Ibg8E7PZUBaF4rr/+ei1atEiNjY06+uijtWvXLp1++ulatWqVXn/9dc2cOVNnnnmmmpub9/k+N998s84//3y9+eabOv3003XRRRfpo48+6rH/p59+qjvuuEO//vWv9cILL6i5uVnXXntt5P5Pf/pTPfLII6qvr9dLL72ktra2Pi9NGzx4sC6//HK99NJLkYNLd+7cqUsuuUQvvviiXn75ZR166KE6/fTTtXPnTkl2SJKk+vp6BYPByHVfPx8AAADkKSvHtba2WpKs1tbWmHufffaZ9fe//9367LPPev2+e/daltttWVL3X4ZhWR6P3S9V6uvrrdLS0sj16tWrLUnWk08+Gfe1RxxxhHXPPfdErseOHWv97Gc/i1xLsn70ox9Frnft2mVJsp555pmo7/Xxxx9HxiLJ2rRpU+Q19957rzVq1KjI9ahRo6zbb789cr13715rzJgxVnV1dY/j7Pp9OnvmmWcsSdYrr7zS7WtDoZBVUlJi/fGPf4z6uX7/+9/3+P3Cun4+XfXn2QEAAEBm7CsbdMZMUA/8/tgZoM4sS2ppsful2/HHHx91vWvXLl177bWaOHGiysrKNHToUDU2Nsad6Tj66KMj/7z//vvL6XRGZl26M2TIEB188MGRa5fLFenf2tqqbdu26ctf/nLkvsPh0KRJk3r1s3VmWZakf5/Zs23bNs2ZM0eHHnqoSktL5XQ6tWvXrrg/Z18/HwAAAOQnCiP0INHCZH0oYNZv+++/f9T1tddeqxUrVuiOO+7QIYccosGDB6umpka7d+/e5/sMHDgw6towDHV0dPSqfziopEJjY6Okf+/1ueSSS/Thhx9q8eLFGjt2rIqLizVlypS4P2dfPx8AAADsW6gjJH+zX8GdQblKXPKO8cpRlKIKYklECOqBy5Xcfqn00ksvafbs2TrnnHMk2TMfmzdvTusYSktLNWrUKL366qs66aSTJEmhUEivvfaajjnmmF6/32effab7779fJ510UqSy3UsvvaSf//znOv300yVJLS0t+te//hX1uoEDByoUCkW1ZcPnAwAAkG/MRlPzls/TB23/Xj7ldrq1eOZi+Sb6Mjiy+FgO1wOvV3K7pS9WYsUwDMnjsftl2qGHHirTNLV+/Xq98cYbmjVr1j5ndFLlqquu0sKFC/WHP/xBGzdu1Lx58/Txxx9HlrPty/bt27V161a98847euyxx/SVr3xF//rXv/SLX/wi0ufQQw/Vr3/9azU2NuqVV17RRRddpMGDB0e9T2VlpVatWqWtW7fq448/jrwuGz4fAACAfGE2mqpZWhMVgCQp0BZQzdIamY0ZqiCWIEJQDxwOafFi+5+7/g4fvq6rS915Qb1x11136YADDtDUqVN15plnasaMGTruuOPSPo7rrrtOF154oS6++GJNmTJFQ4cO1YwZM7TffvvFfe348eM1evRoTZo0SYsWLdL06dP11ltv6fDDD4/0efDBB/Xxxx/ruOOO0ze/+U19//vf18iRI6Pe584779SKFSvk8Xh07LHHSsqezwcAACAfhDpCmrd8nizFbosIt9Uur1WoIxRzP1sYVio3daRBW1ubSktL1draKqfTGXXv888/V1NTk8aNG5fQL+LdMU1p3rzoIgkejx2AfNk9y5dxHR0dmjhxos4//3zdeuutmR5OryTj2QEAAMhHazav0bSHp8Xtt/qS1aqqrEr9gDrZVzbojD1Bcfh8UnW1XQUuGLT3AHm92TEDlG3ef/99/elPf9LJJ5+s9vZ2/c///I+ampo0a9asTA8NAAAASRLcmVhlsET7ZQIhKAEOh1RVlelRZL+ioiItWbJE1157rSzL0pFHHqmVK1dq4sSJmR4aAAAAksRVklhlsET7ZQIhCEnj8Xj00ksvZXoYAAAASCHvGK/cTrcCbYFu9wUZMuR2uuUdkwUVxHpAYQQAAAAACXMUObR4pl1BzFB0BbHwdd3Muqw+L4gQBAAAAKBXfBN9aji/QRXOiqh2t9OthvMbsv6cIJbDAQAAAAUs1BGSv9mv4M6gXCUuecd4E5rF8U30qXp8dZ9em2mEIAAAAKBAmY2m5i2fF3Xoqdvp1uKZixOazXEUOdJeBjsZWA4HAAAAFCCz0VTN0pqoACRJgbaAapbWyGw0MzSy1CMEAQAAAAUm1BHSvOXzuq3uFm6rXV6rUEco3UNLC0JQgbvpppt0zDHHZHoYAAAASCN/sz9mBqgzS5Za2lrkb/ancVTpQwjKMoZh7PPrpptu6td7P/nkk1Ft1157rVatWtW/QSfgpptuivwMAwYM0IEHHqiTTjpJdXV1am9v79V7rVmzRoZhaMeOHakZLAAAQJ4L7gwmtV+u6XMIeuGFF3TmmWdq9OjR3f5ybVmWbrjhBrlcLg0ePFjTp0/XO++8E9Xno48+0kUXXSSn06mysjJdeuml2rVrV1+HlDKhjpDWbF6j3/7tt1qzeU1KpwWDwWDkq66uTk6nM6rt2muvTer3Gzp0qIYPH57U9+zJEUccoWAwqObmZq1evVrnnXeeFi5cqKlTp2rnzp1pGQMAAAAkV4krqf1yTZ9D0CeffKIvfelLuvfee7u9f9ttt+nuu+/Wfffdp1deeUX777+/ZsyYoc8//zzS56KLLtLbb7+tFStW6KmnntILL7ygyy67rK9DSgmz0VTl4kpNe3iaZpmzNO3haapcXJmyjWLl5eWRr9LSUhmGEdX22GOPaeLEidpvv/00YcIE/fznP4+8dvfu3bryyivlcrm03377aezYsVq4cKEkqbKyUpJ0zjnnyDCMyHXX5XCzZ8/W2WefrTvuuEMul0vDhw/X3LlztWfPnkifYDCoM844Q4MHD9a4ceP06KOPqrKyUnV1dfv82QYMGKDy8nKNHj1aRx11lK666io9//zzeuutt/TTn/400u/Xv/61jj/+eJWUlKi8vFyzZs3S9u3bJUmbN2/WtGnTJEkHHHCADMPQ7NmzJUnLly/XiSeeqLKyMg0fPlxf+9rX9O677/blXwMAAEBe847xyu10xxx2GmbIkMfpkXeMN80jS48+h6DTTjtN//Vf/6Vzzjkn5p5lWaqrq9OPfvQjVVdX6+ijj9b//u//asuWLZEZo8bGRi1fvlwPPPCAJk+erBNPPFH33HOPHnvsMW3ZsqXPP1AyZVvFjEceeUQ33HCDfvKTn6ixsVH//d//rR//+Md6+OGHJUl33323li1bpqVLl2rjxo165JFHImHn1VdflSTV19crGAxGrruzevVqvfvuu1q9erUefvhhLVmyREuWLIncv/jii7VlyxatWbNGv/vd73T//fdHQkpvTZgwQaeddppM89+f5Z49e3TrrbfqjTfe0JNPPqnNmzdHgo7H49Hvfvc7SdLGjRsVDAa1eLF9YvEnn3yi+fPn669//atWrVqloqIinXPOOero6OjT2AAAAPKVo8ihxTPt36G6BqHwdd3Mupw486cvUnJOUFNTk7Zu3arp06dH2kpLSzV58mStXbtWX//617V27VqVlZXp+OOPj/SZPn26ioqK9Morr3QbriSpvb09ag9JW1tbKn6EuBUzDBmqXV6r6vHVaXs4brzxRt15553y+eya7ePGjdPf//53/fKXv9Qll1yi5uZmHXrooTrxxBNlGIbGjh0bee2IESMkSWVlZSovL9/n9znggAP0P//zP3I4HJowYYLOOOMMrVq1SnPmzNGGDRu0cuVKvfrqq5F/dw888IAOPfTQPv9cEyZM0J/+9KfI9be//e3IPx900EG6++67dcIJJ2jXrl0aOnSohg0bJkkaOXKkysrKIn3PPffcqPd96KGHNGLECP3973/XkUce2efxAQAA5CPfRJ8azm/o9pygupl1CZ0TlKtSEoK2bt0qSRo1alRU+6hRoyL3tm7dqpEjR0YPZsAADRs2LNKnOwsXLtTNN9+c5BHH6k3FjHQcEPXJJ5/o3Xff1aWXXqo5c+ZE2vfu3avS0lJJ9lK2r371qxo/frxmzpypr33tazr11FN7/b2OOOIIORz/DnYul0t/+9vfJNmzLwMGDNBxxx0XuX/IIYfogAMO6OuPJsuyZBj//huIdevW6aabbtIbb7yhjz/+ODKT09zcrMMPP7zH93nnnXd0ww036JVXXtG//vWvqNcRggAAQL4LdYTkb/YruDMoV4lL3jHeuH9Z75voU/X46l6/LtelJASl0oIFCzR//vzIdVtbmzweT9K/T7ZVzAgXjPjVr36lyZMnR90LB5bjjjtOTU1NeuaZZ7Ry5Uqdf/75mj59uhoaGnr1vQYOHBh1bRhGSpeUNTY2aty4cZLssDdjxgzNmDFDjzzyiEaMGKHm5mbNmDFDu3fv3uf7nHnmmRo7dqx+9atfafTo0ero6NCRRx4Z93UAAAC5zmw0u53RWTxzcdwZHUeRIy1/qZ9NUhKCwsuttm3bJpfr3xUltm3bFtmEX15eHrOPZO/evfroo4/2uVyruLhYxcXFyR90F9lWMWPUqFEaPXq03nvvPV100UU99nM6nbrgggt0wQUXqKamRjNnztRHH32kYcOGaeDAgQqF+lfZbvz48dq7d69ef/11TZo0SZK0adMmffzxx316vw0bNmj58uVasGBB5PrDDz/UokWLIuH2r3/9a9RrBg0aJElRP8uHH36ojRs36le/+pW8XnsD34svvtinMQEAAOSS8D72rts4wvvYG85vyOulbX2RknOCxo0bp/Ly8qjzZ9ra2vTKK69oypQpkqQpU6Zox44dWrduXaTPc889p46OjpiZjkzIxooZN998sxYuXKi7775b//jHP/S3v/1N9fX1uuuuuyRJd911l377299qw4YN+sc//qEnnnhC5eXlkX0zlZWVWrVqlbZu3drn0DJhwgRNnz5dl112mf7yl7/o9ddf12WXXabBgwdHLWnrzt69e7V161Zt2bJFf/vb33TPPffo5JNP1jHHHKMf/OAHkqQxY8Zo0KBBuueee/Tee+9p2bJluvXWW6PeZ+zYsTIMQ0899ZT++c9/ateuXTrggAM0fPhw3X///dq0aZOee+65qBlDAACAfBRvH7sk1S6vTekRL7mozyFo165dWr9+vdavXy/JLoawfv16NTc3yzAM1dbW6r/+67+0bNky/e1vf9PFF1+s0aNH6+yzz5YkTZw4UTNnztScOXP0l7/8RS+99JKuvPJKff3rX9fo0aOT8bP1SzZWzPjOd76jBx54QPX19TrqqKN08skna8mSJZGlZCUlJbrtttt0/PHH64QTTtDmzZv19NNPq6jI/td85513asWKFfJ4PDr22GP7PI7//d//1ahRo3TSSSfpnHPO0Zw5c1RSUqL99ttvn697++235XK5NGbMGFVVVWnp0qVasGCB/H6/hg4dKsku4LBkyRI98cQTOvzww7Vo0SLdcccdUe9TUVGhm2++Wddff71GjRqlK6+8UkVFRXrssce0bt06HXnkkbr66qt1++239/lnBAAAyAW92ceOfzMsy4qNjQlYs2ZN5LyWzi655BItWbJElmXpxhtv1P33368dO3boxBNP1M9//nMddthhkb4fffSRrrzySv3xj39UUVGRzj33XN19992RX4gT0dbWptLSUrW2tsrpdEbd+/zzz9XU1KRx48bF/QW9J92tr/Q4PXlfMaM3PvjgA3k8Hq1cuVKnnHJKpoeTFMl4dgAAAFLtt3/7rWaZs+L2e9T3qC486sI0jCiz9pUNOutzCMoWqQ5BUt8qbeSz5557Trt27dJRRx2lYDCoH/7whwoEAvrHP/4RU1QhVxGCAABALlizeY2mPRw7MdHV6ktWF0Txg0RDUM5Vh8uEQqyYsS979uzRf/zHf+i9995TSUmJpk6dqkceeSRvAhAAAECuCO9jD7QFut0XZMiQ2+lO6z72XEAIQq+FS1gDAAAgs8L72GuW1siQERWEMrWPPRekpDocAAAAgPTwTfSp4fwGVTgrotrdTjflsXvATBAAAACQJfq6F9030afq8dXsY08QIQgAAADIAt1VJXY73Vo8c3FCsznsY08cy+EAAACADDMbTdUsrYk58yfQFlDN0hqZjWaGRpafCEEAAABABoU6Qpq3fF631d3CbbXLaxXqCKV7aHmLEAQAAABkkL/ZHzMD1JklSy1tLfI3+9M4qvxGCCpws2fP1tlnnx25rqqqUm1tbb/eMxnvAQAAUCiCO4NJ7Yf4CEFZavbs2TIMQ4ZhaNCgQTrkkEN0yy23aO/evSn9vqZp6tZbb02o75o1a2QYhnbs2NHn9wAAACh0rhJXUvshPqrDJSIUkvx+KRiUXC7J65UcqS83OHPmTNXX16u9vV1PP/205s6dq4EDB2rBggVR/Xbv3q1BgwYl5XsOGzYsK94DAACgUHjHeOV2uhVoC3S7L8iQIbfTLe8YbwZGl5+YCYrHNKXKSmnaNGnWLPvPykq7PcWKi4tVXl6usWPH6oorrtD06dO1bNmyyBK2n/zkJxo9erTGjx8vSWppadH555+vsrIyDRs2TNXV1dq8eXPk/UKhkObPn6+ysjINHz5cP/zhD2VZ0f9D67qUrb29Xdddd508Ho+Ki4t1yCGH6MEHH9TmzZs1bdo0SdIBBxwgwzA0e/bsbt/j448/1sUXX6wDDjhAQ4YM0WmnnaZ33nkncn/JkiUqKyvTs88+q4kTJ2ro0KGaOXOmgsF/T/muWbNGX/7yl7X//vurrKxMX/nKV/T+++8n6ZMGAADIHEeRQ4tnLpZkB57Owtd1M+s48yeJCEH7YppSTY30QZeNaoGA3Z6GINTZ4MGDtXv3bknSqlWrtHHjRq1YsUJPPfWU9uzZoxkzZqikpER+v18vvfRSJEyEX3PnnXdqyZIleuihh/Tiiy/qo48+0u9///t9fs+LL75Yv/3tb3X33XersbFRv/zlLzV06FB5PB797ne/kyRt3LhRwWBQixcv7vY9Zs+erb/+9a9atmyZ1q5dK8uydPrpp2vPnj2RPp9++qnuuOMO/frXv9YLL7yg5uZmXXvttZKkvXv36uyzz9bJJ5+sN998U2vXrtVll10mwzC6/X4AAAC5xjfRp4bzG1ThrIhqdzvdaji/IaFzgpA4lsP1JBSS5s2TrNgpSVmWZBhSba1UXZ3ypXGWZWnVqlV69tlnddVVV+mf//yn9t9/fz3wwAORZXC/+c1v1NHRoQceeCASDurr61VWVqY1a9bo1FNPVV1dnRYsWCCfz/4f0X333adnn322x+/7j3/8Q0uXLtWKFSs0ffp0SdJBBx0UuR9e9jZy5EiVlZV1+x7vvPOOli1bppdeeklTp06VJD3yyCPyeDx68skndd5550mS9uzZo/vuu08HH3ywJOnKK6/ULbfcIklqa2tTa2urvva1r0XuT5w4sfcfJAAAQJqEOkLyN/sV3BmUq8Ql7xhv3Jkc30SfqsdX9/p16D1CUE/8/tgZoM4sS2ppsftVVaVkCE899ZSGDh2qPXv2qKOjQ7NmzdJNN92kuXPn6qijjoraB/TGG29o06ZNKikpiXqPzz//XO+++65aW1sVDAY1efLkyL0BAwbo+OOPj1kSF7Z+/Xo5HA6dfPLJff4ZGhsbNWDAgKjvO3z4cI0fP16NjY2RtiFDhkQCjiS5XC5t375dkh22Zs+erRkzZuirX/2qpk+frvPPP18uF5sDAQBA9jEbTc1bPi+q7LXb6dbimYvjzug4ihyqqqxK8QjBcrieBBMsQZhovz6YNm2a1q9fr3feeUefffaZHn74Ye2///6SFPkzbNeuXZo0aZLWr18f9fWPf/xDs2bN6tP3Hzx4cL9/hkQNHDgw6towjKhwVl9fr7Vr12rq1Kl6/PHHddhhh+nll19O2/gAAAASYTaaqllaE3PuT6AtoJqlNTIb07udAt0jBPUk0VmGFM5G7L///jrkkEM0ZswYDRiw70m74447Tu+8845GjhypQw45JOqrtLRUpaWlcrlceuWVVyKv2bt3r9atW9fjex511FHq6OjQ888/3+398ExUKNTz6cUTJ07U3r17o77vhx9+qI0bN+rwww/f58/U1bHHHqsFCxboz3/+s4488kg9+uijvXo9AABAKoU6Qpq3fF63Fd7CbbXLaxXq6Pl3J6QHIagnXq/kdtt7f7pjGJLHY/fLAhdddJEOPPBAVVdXy+/3q6mpSWvWrNH3v/99ffDFsr558+Zp0aJFevLJJ7VhwwZ973vfiznjp7PKykpdcskl+va3v60nn3wy8p5Lly6VJI0dO1aGYeipp57SP//5T+3atSvmPQ499FBVV1drzpw5evHFF/XGG2/oG9/4hioqKlRdXZ3Qz9bU1KQFCxZo7dq1ev/99/WnP/1J77zzDvuCAABAVvE3+2NmgDqzZKmlrUX+Zn8aR4XuEIJ64nBI4WpnXYNQ+LquLi3nBSViyJAheuGFFzRmzBj5fD5NnDhRl156qT7//HM5nU5J0jXXXKNvfvObuuSSSzRlyhSVlJTonHPO2ef7/uIXv1BNTY2+973vacKECZozZ44++eQTSVJFRYVuvvlmXX/99Ro1apSuvPLKbt+jvr5ekyZN0te+9jVNmTJFlmXp6aefjlkCt6+fbcOGDTr33HN12GGH6bLLLtPcuXP13e9+txefEAAAQGoFdya2TSLRfkgdw+ppV3yOaGtrU2lpqVpbWyO/7Id9/vnnampq0rhx47Tffvv17RuYpl0lrnORBI/HDkA+ShXmq6Q8OwAAoKCs2bxG0x6eFrff6ktWU/wgRfaVDTqjOlw8Pp9dBtvvt4sguFz2ErgsmQECAABAdvCO8crtdCvQFuh2X5AhQ26nW94x2bGdopARghLhcKSsDDYAAADyg6PIocUzF6tmaY0MGVFByJC9naJuZh3n/mQB9gQBAAAASeKb6FPD+Q2qcFZEtbudbjWc3xD3nCCkBzNBAAAAQA9CHSH5m/0K7gzKVeKSd4w37kyOb6JP1eOre/06pE9BhKAcr/2ADOCZAQAAZqOpecvnRZW9djvdWjxzcdwZHUeRg+IHWSyvl8OFSzB/+umnGR4Jck34mUm0jDcAAMgvZqOpmqU1Mef+BNoCqllaI7PRzNDIkAx5PRPkcDhUVlam7du3S7LPmzF6OvwUkD0D9Omnn2r79u0qKyuTgyqAAAAUnFBHSPOWz+u2wpslS4YM1S6vVfX4apa45ai8DkGSVF5eLkmRIAQkoqysLPLsAACAwuJv9sfMAHVmyVJLW4v8zX6WvOWovA9BhmHI5XJp5MiR2rNnT6aHgxwwcOBAZoAAAChgwZ3BpPZD9sn7EBTmcDj4xRYAAABxuUpcSe2H7JPXhREAAACA3vKO8crtdEcOOO3KkCGP0yPvGG+aR4ZkIQQBAAAAnTiKHFo8c7EkxQSh8HXdzDqKIuQwQhAAAADQhW+iTw3nN6jCWRHV7na61XB+Q9xzgpDdDCvHT4Vsa2tTaWmpWltb5XQ6Mz0cAAAAZJlQR0j+Zr+CO4NylbjkHeNNeBanP69F+iWaDQqmMAIAAAAKj9loat7yeVElr91OtxbPXJzQbI6jyEEZ7DzEcjgAAADkJbPRVM3SmpgzfwJtAdUsrZHZaGZoZMg0QhAAAADyTqgjpHnL58lS7M6PcFvt8lqFOkLpHhqyACEIAAAAecff7I+ZAerMkqWWthb5m/1pHBWyBSEIAAAAeSe4M5jUfsgvhCAAAADkHVeJK6n9kF8IQQAAAMg73jFeuZ3umMNOwwwZ8jg98o7xpnlkyAaEIAAAAOQdR5FDi2culqSYIBS+rptZx5k/BYoQBAAAgLzkm+hTw/kNqnBWRLW7nW41nN+Q0DlByE+GZVmxdQNzSKKnwgIAACC3hTpC8jf7FdwZlKvEJe8Yb0IzOX19HXJPotlgQBrHBAAAAPSJ2Whq3vJ5UWWv3U63Fs9cHHdGx1HkUFVlVYpHiFzCcjgAAABkNbPRVM3SmphzfwJtAdUsrZHZaGZoZMhVhCAAAABkrVBHSPOWz5Ol2B0c4bba5bUKdYTSPTTkMEIQAAAAspa/2R8zA9SZJUstbS3yN/vTOCrkOkIQAAAAslZwZzCp/QCJEAQAAIAs5ipxJbUfIBGCAAAAkMW8Y7xyO90xB56GGTLkcXrkHeNN88iQy1IagiorK2UYRszX3LlzJUlVVVUx9y6//PJUDgkAAAA5xFHk0OKZiyUpJgiFr+tm1nHuD3olpSHo1VdfVTAYjHytWLFCknTeeedF+syZMyeqz2233ZbKIQEAACDH+Cb61HB+gyqcFVHtbqdbDec3xD0nCOgqpYeljhgxIup60aJFOvjgg3XyySdH2oYMGaLy8vJUDgMAAABZJNQRkr/Zr+DOoFwlLnnHeOPO5Pgm+lQ9vrrXrwO6k9IQ1Nnu3bv1m9/8RvPnz5dh/Hsq85FHHtFvfvMblZeX68wzz9SPf/xjDRkypMf3aW9vV3t7e+S6ra0tpeMGAABA8piNpuYtnxdV9trtdGvxzMVxZ3QcRQ5VVValeIQoBGkLQU8++aR27Nih2bNnR9pmzZqlsWPHavTo0XrzzTd13XXXaePGjTLNnk/9XbhwoW6++eY0jBgAAADJZDaaqllaE3PwaaAtoJqlNSxtQ9oYlmXFHr+bAjNmzNCgQYP0xz/+scc+zz33nE455RRt2rRJBx98cLd9upsJ8ng8am1tldPpTPq4AQAA0H+hjpAqF1f2ePCpIUNup1tN85pY4oY+a2trU2lpadxskJYS2e+//75Wrlyp73znO/vsN3nyZEnSpk2beuxTXFwsp9MZ9QUAAIDs5m/29xiAJMmSpZa2Fvmb/WkcFQpVWkJQfX29Ro4cqTPOOGOf/davXy9Jcrk47AoAACCfBHcGk9oP6I+U7wnq6OhQfX29LrnkEg0Y8O9v9+677+rRRx/V6aefruHDh+vNN9/U1VdfrZNOOklHH310qocFAACANHKVJPaX3In2A/oj5SFo5cqVam5u1re//e2o9kGDBmnlypWqq6vTJ598Io/Ho3PPPVc/+tGPUj0kAAAApJl3jFdup1uBtkBMYQTp33uCvGO8GRgdCk3aCiOkSqKbnwAAAJBZ4epwkqKCkCH7+BSqw6G/sqowAgAAAOCb6FPD+Q2qcFZEtbudbgIQ0oqZIAAAAPRaqCMkf7NfwZ1BuUpc8o7xJlzauj+vBfYl0WyQtsNSAQAAkB/MRlPzls+LKnntdrq1eObihGZzHEUOVVVWpXCEwL6xHA4AAAAJC+/r6XrmT6AtoJqlNTIbzQyNDEgcIQgAAAAJCXWENG/5vG6ru4XbapfXKtQRSvfQgF4hBAEAACAh/mZ/zAxQZ5YstbS1yN/sT+OogN4jBAEAACAhwZ3BpPYDMoUQBAAAgIS4SlxJ7QdkCiEIAAAACfGO8crtdEcON+3KkCGP0yPvGG+aRwb0DiEIAAAACXEUObR45mJJiglC4eu6mXWc+YOsRwgCAABAwnwTfWo4v0EVzoqodrfTrYbzGxI6JwjINMOyrNgahzkk0VNhAQAAECsUkvx+KRiUXC7J65UcCUzkhDpC8jf7FdwZlKvEJe8YLzNAyLhEs8GANI4JAAAAWcQ0pXnzpA86Vb12u6XFiyVfnAkdR5FDVZVVKR0fkCoshwMAAChApinV1EQHIEkKBOx208zMuIB0IAQBAAAUmFDIngHqblNEuK221u4H5CNCEAAAQIHx+2NngDqzLKmlxe4H5CNCEAAAQIEJBpPbD8g1hCAAAIAC43Iltx+QawhBAAAABcbrtavAGUb39w1D8njsfkA+IgQBAAAUGIfDLoMtxQah8HVdXWLnBQG5iBAEAABQgHw+qaFBqqiIbne77fZ45wQBuYzDUgEAAHJcKGRXcgsG7X08Xm9iszg+n1Rd3bfXArmMEAQAAJDDTNM+86dzyWu3217ulshsjsMhVVWlbHhAVmI5HAAAQI4yTammJvbMn0DAbjfNzIwLyHaEIAAAgBwUCtkzQJYVey/cVltr9wMQjRAEAACQg/z+2BmgzixLammx+wGIRggCAADIQcFgcvsBhYQQBAAAkINcruT2AwoJIQgAACAHeb12Fbiuh52GGYbk8dj9AEQjBAEAAOQgh8Mugy3FBqHwdV0dZ/4A3SEEAQAA5CifT2pokCoqotvdbrs9kXOCgELEYakAAABZIhSyq7kFg/ZeHq83/kyOzydVV/f+dUAhIwQBAABkAdO0z/3pXPba7baXvMWb0XE4pKqqlA4PyCsshwMAAMgw05RqamLP/QkE7HbTzMy4gHxFCAIAAMigUMieAbKs2Hvhttpaux+A5CAEAQAAZJDfHzsD1JllSS0tdj8AyUEIAgAAyKBgMLn9AMRHCAIAAMgglyu5/QDER3U4AACADPJ67SpwgUD3+4IMw77v9aZ/bEBcfanrngWYCQIAAMggh8Mugy3Zgaez8HVdXU78XolCY5pSZaU0bZo0a5b9Z2VlTpQzJAQBAABkmM8nNTRIFRXR7W633R7vnCAg7XK8rrthWd1NvOaOtrY2lZaWqrW1VU6nM9PDAQAA6PMKoRxdWYRCEwrZMz49lTUMr+Fsakr7A5xoNmBPEAAAQBKZpn3uT+ffD91ue8lbvBkdh0Oqqkrp8ID+601d9yx9oFkOBwAAkCQ5vkIISEwe1HUnBAEAACRBKGTPAHW30SDcVltr9wOySigkrVkj/fa39p/xHtI8qOtOCAIAAEiC3qwQArJGXyq8heu6dy1nGGYYkseT1XXdCUEAAABJkAcrhFBo+rp+Mw/quhOCAAAAkiAPVgihkPR3/WaO13WnRDYAAEAShKsGBwLd/16ZwarBQKw1a+ylb/GsXr3vCm9ZVtedEtkAAABpFF4hVFNjB57OQShHVggh1/UmkCRr/WaO1nVnORwAAECS5PgKIeSy3hY4KPD1myyHAwAA6EZ/Vvlk2Qoh5LtwgYOuv9aHpyC7S+B5un4z0WyQ0pmgm266SYZhRH1NmDAhcv/zzz/X3LlzNXz4cA0dOlTnnnuutm3blsohAQAAxNWXqsGdhVcIXXih/WcO/Q6JXNPXAgd5UOGtP1K+HO6II45QMBiMfL344ouRe1dffbX++Mc/6oknntDzzz+vLVu2yMc8MQAAyKC+Vg0GMqI/B1QV8PrNlBdGGDBggMrLy2PaW1tb9eCDD+rRRx/V//t//0+SVF9fr4kTJ+rll1/W//f//X+pHhoAAECUeH+pbhj2X6pXV+ftX5Aj1/S3wIHPZz/QBbZ+M+Uh6J133tHo0aO13377acqUKVq4cKHGjBmjdevWac+ePZo+fXqk74QJEzRmzBitXbu2xxDU3t6u9vb2yHVbW1uqfwQAAFAgevOX6jlYEAu5oLcbypJR4CBHK7z1R0qXw02ePFlLlizR8uXL9Ytf/EJNTU3yer3auXOntm7dqkGDBqmsrCzqNaNGjdLWrVt7fM+FCxeqtLQ08uXxeFL5IwAAgAKSrKrBQJ/0ZTOa12svX+u6ryfMMCSPx+6HiJSGoNNOO03nnXeejj76aM2YMUNPP/20duzYoaVLl/b5PRcsWKDW1tbIV0tLSxJHDAAAClmBVw1GJvV1M1qBFzjoq7SeE1RWVqbDDjtMmzZtUnl5uXbv3q0dO3ZE9dm2bVu3e4jCiouL5XQ6o74AAACSgb9UR0b0tcJbWAEXOOirtIagXbt26d1335XL5dKkSZM0cOBArVq1KnJ/48aNam5u1pQpU9I5LAAAAEn8pToypD8V3sJ8PmnzZmn1aunRR+0/m5oIQD1IaWGEa6+9VmeeeabGjh2rLVu26MYbb5TD4dCFF16o0tJSXXrppZo/f76GDRsmp9Opq666SlOmTKEyHAAAyJjwX6rPmxf9e6nbbQcgfqdE0iVrM1oBFjjoq5SGoA8++EAXXnihPvzwQ40YMUInnniiXn75ZY0YMUKS9LOf/UxFRUU699xz1d7erhkzZujnP/95KocEAAAKTG+LbUkFWzUYydSbB4/NaGlnWFZ3iw9zR1tbm0pLS9Xa2sr+IAAAEMU0u5/RWbyYGR2kUG8fvFDIrgIXCHS/L8gw7Nc3NZHE40g0G6R1TxAAAEC69LXYFtAvfXnw2IyWdoQgAACQd/pbbAvok/48eFR4SytCEAAAyDvJKLYF9Fp/HzwqvKVNSgsjAAAAZEKyim0BvSpwkIwHjwpvaUEIAgAAeYdiW0iK3hY44MHLGSyHAwAAecfrtX9X7brHPMwwJI/H7gd0qy8FDnjwcgYhCAAA5B2KbaFf+lrggAcvZxCCAABAXqLYFvqsPwUOePByAnuCAABATujN/vQwn0+qru7961Dg+lvggAcv6xGCAABA1uvt/vTOKLaFXifoZBQ44MHLaiyHAwAAWa0v+9OBCNOUKiuladOkWbPsPysr9/3gUOAg7xGCAABA1urr/nRAUt8TNAUO8h4hCAAAZK3+7E9HgetvgqbAQV5jTxAAAMha/d2fjgLWmwTd094dChzkLUIQAADIWsnYn4480psCB8lK0BQ4yEsshwMAAFmL/emI6G2BAxI09oEQBAAAshb70yGpbwUOSNDYB0IQAADIauxPL3B9LXBAgsY+GJbV3ROVO9ra2lRaWqrW1lY5nc5MDwcAAOxDb8+sTNZrkcPWrLGXvsWzenX3e3e6O2nX47EDEAk67ySaDSiMAAAA0qK730Xdbvsv6xP5XZT96QWqvwUOqPCGbhCCAABAyoW3dHRdfxLe0sGytgLTm2m9ZBQ4IEGjC/YEAQCAlOrvmZXIM72t8kaBA6QAIQgAAKRUb86sRJ7rS5U3ChwgBQhBAAAgpZJ1ZiVyXH+mBCkRiCRjTxAAAEgpzqyEpN5NCXa3f4cCB0giQhAAAEip8JaOQKD7SQDDsO+zpSPH9LZmeTKmBClwgCRhORwAAEgptnTkod4WN5CYEkRWIQQBAICUY0tHHulLcQOJKm/IKoZldTcxnTsSPRUWAAAkT29XQvX3dcgSoZA949PT3p7w2sampu7/xYYDlBS9NjIcjEjE6KdEswEzQQAAoFf6shIqLLyl48IL7T8JQDmmv/XOmRJElqAwAgAASFj4L/K7riMJr4Ti99g8l4ziBlR5QxYgBAEAgITEO+bFMOxjXqqr+X02byWruAFV3pBhLIcDAAAJ6e9KKOQBihsgTxCCAABAQpKxEgo5jnrnyBOEIAAAkBCOeYEkihsgL1AiGwAAJCRcHTkQ6H5fULzqyMgz1DtHFko0G1AYAQAAJCS8Eqqmxg483R3zwkqoAkJxA+QwlsMBAICEsRIKQD5gJggAgALV19VMHPMCINcRggAAKECmaZ/507nktdttL3dLZDaHlVAAchnL4QAAKDCmae/r6XrmTyBgt5tmZsYFAOlCCAIAoICEQvYMUHfV3cJttbV2PwDIV4QgAAAKiN8fOwPUmWVJLS12PwDIV4QgAAAKSDCY3H4AkIsIQQAAFBCXK7n9ACAXEYIAACggXq9dBS58uGlXhiF5PHY/AMhXhCAAAAqIw2GXwZZig1D4uq6OM38A5DdCEAAABcbnkxoapIqK6Ha3225P5JwgAMhlHJYKAECOC4Xsam7BoL2Xx+uNP5Pj80nV1b1/HQDkA0IQAAA5zDTtc386l712u+0lb/FmdBwOqaoqpcMDgKzEcjgAAHKUaUo1NbHn/gQCdrtpZmZcAJDtUhqCFi5cqBNOOEElJSUaOXKkzj77bG3cuDGqT1VVlQzDiPq6/PLLUzksAAByXihkzwBZVuy9cFttrd0PABAtpSHo+eef19y5c/Xyyy9rxYoV2rNnj0499VR98sknUf3mzJmjYDAY+brttttSOSwAAHKe3x87A9SZZUktLXY/AEC0lO4JWr58edT1kiVLNHLkSK1bt04nnXRSpH3IkCEqLy9P5VAAAMgrwWBy+wFAIUnrnqDW1lZJ0rBhw6LaH3nkER144IE68sgjtWDBAn366ac9vkd7e7va2tqivgAAKDQuV3L7AUAhSVt1uI6ODtXW1uorX/mKjjzyyEj7rFmzNHbsWI0ePVpvvvmmrrvuOm3cuFFmD7s5Fy5cqJtvvjldwwYAICt5vXYVuECg+31BhmHf93rTPzYAyHaGZXX3n87ku+KKK/TMM8/oxRdflNvt7rHfc889p1NOOUWbNm3SwQcfHHO/vb1d7e3tkeu2tjZ5PB61trbK6XSmZOwAAGSjcHU4KToIGYb9JwefAig0bW1tKi0tjZsN0rIc7sorr9RTTz2l1atX7zMASdLkyZMlSZs2ber2fnFxsZxOZ9QXAACFyOezg05FRXS7200AAoB9SelyOMuydNVVV+n3v/+91qxZo3HjxsV9zfr16yVJLhYxAwAKTChkV3MLBu29PF6vfaDpvvh8UnV1718HAIUspSFo7ty5evTRR/WHP/xBJSUl2rp1qySptLRUgwcP1rvvvqtHH31Up59+uoYPH64333xTV199tU466SQdffTRqRwaAABZxTTtc386l712u6XFi+PP6DgcUlVVSocHAHklpXuCjPCi5C7q6+s1e/ZstbS06Bvf+IbeeustffLJJ/J4PDrnnHP0ox/9KOFlbomu+wMAIFuF9/Z0/X9k9vYAQO8kmg3SVhghVQhBAIBcFgpJlZU9H3warvLW1MQSNwCIJ6sKIwAAgO75/T0HIMmeHWppsfsBAJKDEAQAQAYFg8ntBwCIjxAEAEAGJVoMlaKpAJA8hCAAADLI67X3/PRQS0iGIXk8dj8AQHIQggAAyCCHwy6DLcUGofB1XR1FEQAgmQhBAABkmM9nl8GuqIhud7spjw0AqZDSw1IBACg0oZBdyS0YtPfxeL2JzeL4fFJ1dd9eCwDoHUIQAABJYprSvHnRJa/dbnu5WyKzOQ6HVFWVsuEBAL7AcjgAAJLANKWamtgzfwIBu900MzMuAEAsQhAAAP0UCtkzQJYVey/cVltr9wMAZB4hCACAfvL7Y2eAOrMsqaXF7gcAyDxCEAAA/RQMJrcfACC1CEEAAPSTy5XcfgCA1CIEAQDQT16vXQWu62GnYYYheTx2PwBA5hGCAADoJ4fDLoMtxQah8HVdHWf+AEC2IAQBAJAEPp/U0CBVVES3u912eyLnBAEA0oPDUgEA6EYoZFdzCwbtvTxeb/yZHJ9Pqq7u/esAAOlFCAIAoAvTtM/96Vz22u22l7zFm9FxOKSqqpQODwDQTyyHAwCgE9OUampiz/0JBOx208zMuAAAyUMIAgDgC6GQPQNkWbH3wm21tXY/AEDuIgQBAPAFvz92Bqgzy5JaWux+AIDcRQgCAOALwWBy+wEAshMhCACAL7hcye0HAMhOhCAAAL7g9dpV4LoeeBpmGJLHY/cDAOQuQhAAAF9wOOwy2FJsEApf19Vx7g8A5DpCEAAAnfh8UkODVFER3e522+3xzgkCAGQ/DksFAOS1UMiu5hYM2nt5vN74Mzk+n1Rd3fvXAQByAyEIAJC3TNM+96dz2Wu3217yFm9Gx+GQqqpSOjwAQIawHA4AkJdMU6qpiT33JxCw200zM+MCAGQeIQgAkHdCIXsGyLJi74XbamvtfgCAwkMIAgDkHb8/dgaoM8uSWlrsfgCAwkMIAgDknWAwuf0AAPmFEAQAyDsuV3L7AQDyCyEIAJB3vF67ClzXA0/DDEPyeOx+AIDCQwgCAOQdh8Mugy3FBqHwdV0d5/4AQKEiBAEA8pLPJzU0SBUV0e1ut90e75wgAED+4rBUAEDWC4XsSm7BoL2Px+tNbBbH55Oqq/v2WgBA/iIEAQCymmnaZ/50LnntdtvL3RKZzXE4pKqqlA0PAJCDWA4HAMhapinV1MSe+RMI2O2mmZlxAQByGyEIAJCVQiF7BsiyYu+F22pr7X4AAPQGIQgAkJX8/tgZoM4sS2ppsfsBANAbhCAAQFYKBpPbDwCAMEIQACAruVzJ7QcAQBghCACQlbxeuwpc18NOwwxD8njsfgAA9AYhCACQlRwOuwy2FBuEwtd1dZz5AwDoPUIQACBr+XxSQ4NUURHd7nbb7YmcEwQAQFcclgoASJtQyK7mFgzae3m83vgzOT6fVF3d+9cBANATQhAAIC1M0z73p3PZa7fbXvIWb0bH4ZCqqlI6PABAAWE5HAAg5UxTqqmJPfcnELDbTTMz4wIAFCZCEAAgpUIhewbIsmLvhdtqa+1+AACkAyEIAJBSfn/sDFBnliW1tNj9AABIB0IQACClgsHk9gMAoL+yIgTde++9qqys1H777afJkyfrL3/5S6aHBABIEpcruf0AAOivjIegxx9/XPPnz9eNN96o1157TV/60pc0Y8YMbd++PdNDAwAkgddrV4HreuBpmGFIHo/dDwCAdMh4CLrrrrs0Z84cfetb39Lhhx+u++67T0OGDNFDDz2U6aEBAJLA4bDLYEuxQSh8XVfHuT8AgPTJaAjavXu31q1bp+nTp0faioqKNH36dK1du7bb17S3t6utrS3qCwCQ3Xw+qaFBqqiIbne77fZ45wQBAJBMGT0s9V//+pdCoZBGjRoV1T5q1Cht2LCh29csXLhQN998czqGBwDoRihkV3ILBu19PF5vYrM4Pp9UXd231wIAkEwZDUF9sWDBAs2fPz9y3dbWJo/Hk8ERAUDhME37zJ/OJa/dbnu5WyKzOQ6HVFWVsuEBAJCQjIagAw88UA6HQ9u2bYtq37Ztm8rLy7t9TXFxsYqLi9MxPABAJ6Yp1dTEHnoaCNjtLGsDAOSKjO4JGjRokCZNmqRVq1ZF2jo6OrRq1SpNmTIlgyMDAHQWCtkzQF0DkPTvttpaux8AANku49Xh5s+fr1/96ld6+OGH1djYqCuuuEKffPKJvvWtb2V6aACAL/j90UvgurIsqaXF7gcAQLbL+J6gCy64QP/85z91ww03aOvWrTrmmGO0fPnymGIJAIDMCQaT2w8AgEzKeAiSpCuvvFJXXnllpocBAOiBy5XcfgAAZFLGl8MBALKf12tXget62GmYYUgej90PAIBsRwgCAMTlcNhlsKXYIBS+rqvjzB8AQG4gBAEAEuLz2WWwKyqi291uymMDAHJLVuwJAgCkXyhkV3MLBu29PF5v/Jkcn0+qru796wAAyCaEIAAoQKZpn/vTuey1220veYs3o+NwSFVVKR0eAAApxXI4ACgwpinV1MSe+xMI2O2mmZlxAQCQLoQgACggoZA9A2RZsffCbbW1dj8AAPIVIQgACojfHzsD1JllSS0tdj8AAPIVIQgACkgwmNx+AADkIkIQABQQlyu5/QAAyEWEIAAoIF6vXQWu64GnYYYheTx2PwAA8hUhCAAKiMNhl8GWYoNQ+LqujnN/AAD5jRAEAAXG55MaGqSKiuh2t9tuj3dOEAAAuY7DUgEgx4VCdjW3YNDey+P1xp/J8fmk6urevw4AgHxACAKAHGaa9rk/ncteu932krd4MzoOh1RVldLhAQCQlVgOBwA5yjSlmprYc38CAbvdNDMzLgAAsh0hCAByUChkzwBZVuy9cFttrd0PAABEIwQBQA7y+2NngDqzLKmlxe4HAACiEYIAIAcFg8ntBwBAISEEAUAOcrmS2w8AgEJCCAKAHOT12lXguh54GmYYksdj9wMAANEIQQCQgxwOuwy2FBuEwtd1dZz7AwBAdwhBAJCjfD6poUGqqIhud7vt9njnBAEAUKg4LBUAskAoZFdyCwbtfTxeb2KzOD6fVF3dt9cCAFCoCEEAkGGmaZ/507nktdttL3dLZDbH4ZCqqlI2PAAA8g7L4QAgg0xTqqmJPfMnELDbTTMz4wIAIJ8RggAgQ0IhewbIsmLvhdtqa+1+AAAgeQhBAJAhfn/sDFBnliW1tNj9AABA8hCCACBDgsHk9gMAAIkhBAFAhrhcye0HAAASQwgCgAzxeu0qcF0POw0zDMnjsfsBAIDkIQQBQIY4HHYZbCk2CIWv6+o48wcAgGQjBAFABvl8UkODVFER3e522+2JnBMEAAB6h8NSASCJQiG7mlswaO/l8Xrjz+T4fFJ1de9fBwAA+oYQBABJYpr2uT+dy1673faSt3gzOg6HVFWV0uEBAIAvsBwOAJLANKWamthzfwIBu900MzMuAAAQixAEAP0UCtkzQJYVey/cVltr9wMAAJlHCAKAfvL7Y2eAOrMsqaXF7gcAADKPEAQA/RQMJrcfAABILUIQAPSTy5XcfgAAILUIQQDQT16vXQWu64GnYYYheTx2PwAAkHmEIADoJ4fDLoMtxQah8HVdHef+AACQLQhBAJAEPp/U0CBVVES3u912e7xzggAAQPpwWCoAdCMUsqu5BYP2Xh6vN/5Mjs8nVVf3/nUAACC9CEEA0IVp2uf+dC577XbbS97izeg4HFJVVUqHBwAA+onlcADQiWlKNTWx5/4EAna7aWZmXAAAIHkIQQDwhVDIngGyrNh74bbaWrsfAADIXYQgAPiC3x87A9SZZUktLXY/AACQuwhBAPCFYDC5/QAAQHYiBAHAF1yu5PYDAADZiRAEAF/weu0qcF0PPA0zDMnjsfsBAIDclZIQtHnzZl166aUaN26cBg8erIMPPlg33nijdu/eHdXHMIyYr5dffjkVQwKAuBwOuwy2FBuEwtd1dZz7AwBArkvJOUEbNmxQR0eHfvnLX+qQQw7RW2+9pTlz5uiTTz7RHXfcEdV35cqVOuKIIyLXw4cPT8WQACAhPp/U0ND9OUF1dfHPCQIAANnPsKzuisEm3+23365f/OIXeu+99yTZM0Hjxo3T66+/rmOOOabP79vW1qbS0lK1trbK6XQmabQA8kEoZFdyCwbtfTxeb+KzOP15LQAAyIxEs0FKZoK609raqmHDhsW0n3XWWfr888912GGH6Yc//KHOOuusfb5Pe3u72tvbI9dtbW1JHyuA3Gea3c/mLF6c2GyOwyFVVaVseAAAIIPSUhhh06ZNuueee/Td73430jZ06FDdeeedeuKJJ/R///d/OvHEE3X22Wdr2bJl+3yvhQsXqrS0NPLl8XhSPXwAOcY0pZqa2DN/AgG73TQzMy4AAJAderUc7vrrr9dPf/rTffZpbGzUhAkTIteBQEAnn3yyqqqq9MADD+zztRdffLGamprk38dJhN3NBHk8HpbDAZBkL2OrrOz50FPDsGeEmppY3gYAQL5JyXK4a665RrNnz95nn4MOOijyz1u2bNG0adM0depU3X///XHff/LkyVqxYsU++xQXF6u4uDih8QIoPH5/zwFIkixLammx+7HcDQCAwtSrEDRixAiNGDEiob6BQEDTpk3TpEmTVF9fr6Ki+Cvv1q9fLxenEALoh2Awuf0AAED+SUlhhEAgoKqqKo0dO1Z33HGH/vnPf0bulZeXS5IefvhhDRo0SMcee6wkyTRNPfTQQ3GXzAHAviT69yj8fQsAAIUrJSFoxYoV2rRpkzZt2iS32x11r/MWpFtvvVXvv/++BgwYoAkTJujxxx9XTU1NKoYEoEB4vfaen0DAXvrWVXhPkNeb/rEBAIDskLZzglKFc4IAdBWuDidFByHDsP9saODQUwAA8lGi2SAtJbIBIJ18PjvoVFREt7vdBCAAAJDGw1IBoK9CIbuaWzBo7+XxeuOXt/b5pOrq3r8OAADkP0IQgKxmmtK8edFlr91uafHi+DM6DgdlsAEAQCyWwwHIWuG9PV3P/QkE7HbTzMy4AABAbiMEAchKoZA9A9Rd6ZZwW22t3Q8AAKA3CEEAspLfHzsD1JllSS0tdj8AAIDeIAQByErBYHL7AQAAhBGCAGQllyu5/QAAAMIIQQCyktdrV4ELH3DalWFIHo/dDwAAoDcIQQCyksNhl8GWYoNQ+LqujnN/AABA7xGCAGQtn09qaJAqKqLb3W67Pd45QQAAAN3hsFQAaREK2ZXcgkF7H4/Xm9gsjs8nVVf37bUAAADdIQQBSDnTtM/86Vzy2u22l7slMpvjcEhVVSkbHgAAKDAshwOQUqYp1dTEnvkTCNjtppmZcQEAgMJFCAKQMqGQPQNkWbH3wm21tXY/AACAdCEEAUgZvz92Bqgzy5JaWux+AAAA6UIIApAywWBy+wEAACQDIQhAyrhcye0HAACQDIQgACnj9dpV4LoedhpmGJLHY/cDAABIF0IQgJRxOOwy2FJsEApf19Vx5g8AAEgvQhCAlPL5pIYGqaIiut3tttsTOScIAAAgmTgsFUCvhEJ2Nbdg0N7L4/XGn8nx+aTq6t6/DgAAIBUIQQASZpr2uT+dy1673faSt3gzOg6HVFWV0uEBAAAkhOVwABJimlJNTey5P4GA3W6amRkXAABAbxGCAMQVCtkzQJYVey/cVltr9wMAAMh2hCAAcfn9sTNAnVmW1NJi9wMAAMh2hCAAcQWDye0HAACQSYQgAHG5XMntBwAAkEmEIABxeb12FbiuB56GGYbk8dj9AAAAsh0hCEBcDoddBluKDULh67o6zv0BAAC5gRAEICE+n9TQIFVURLe73XZ7vHOCAAAAsgWHpQIFKhSyq7kFg/ZeHq83/kyOzydVV/f+dQAAANmEEAQUINO0z/3pXPba7baXvMWb0XE4pKqqlA4PAAAgpVgOBxQY05RqamLP/QkE7HbTzMy4AAAA0oUQBBSQUMieAbKs2Hvhttpaux8AAEC+IgQBBcTvj50B6syypJYWux8AAEC+IgQBBSQYTG4/AACAXEQIAgqIy5XcfgAAALmIEAQUEK/XrgLX9cDTMMOQPB67HwAAQL4iBAEFxOGwy2BLsUEofF1Xx7k/AAAgvxGCgALj80kNDVJFRXS72223xzsnCAAAINdxWCqQw0Ihu5JbMGjv4/F6E5vF8fmk6uq+vRYAACDXEYKAHGWa9pk/nUteu932crdEZnMcDqmqKmXDAwAAyFoshwNykGlKNTWxZ/4EAna7aWZmXAAAALmAEATkmFDIngGyrNh74bbaWrsfAAAAYhGCgBzj98fOAHVmWVJLi90PAAAAsQhBQI4JBpPbDwAAoNAQgoAc43Iltx8AAEChIQQBOcbrtavAdT3sNMwwJI/H7gcAAIBYhCAgxzgcdhlsKTYIha/r6jjzBwAAoCeEICAH+XxSQ4NUURHd7nbb7YmcEwQAAFCoOCwVyAKhkF3NLRi09/J4vfFncnw+qbq6968DAAAodCmbCaqsrJRhGFFfixYtiurz5ptvyuv1ar/99pPH49Ftt92WquEAWcs0pcpKado0adYs+8/KysQOPHU4pKoq6cIL7T8JQAAAAPGldCbolltu0Zw5cyLXJSUlkX9ua2vTqaeequnTp+u+++7T3/72N337299WWVmZLrvsslQOC8gapinV1MQefBoI2O0sbQMAAEi+lIagkpISlZeXd3vvkUce0e7du/XQQw9p0KBBOuKII7R+/XrdddddhCAUhFBImjcvNgBJdpthSLW19pI3ZngAAACSJ6WFERYtWqThw4fr2GOP1e233669e/dG7q1du1YnnXSSBg0aFGmbMWOGNm7cqI8//rjH92xvb1dbW1vUF5CL/H7pgw96vm9ZUkuL3Q8AAADJk7KZoO9///s67rjjNGzYMP35z3/WggULFAwGddddd0mStm7dqnHjxkW9ZtSoUZF7BxxwQLfvu3DhQt18882pGjaQNsFgcvsBAAAgMb2aCbr++utjih10/dqwYYMkaf78+aqqqtLRRx+tyy+/XHfeeafuuecetbe392vACxYsUGtra+SrpaWlX+8HZIrLldx+AAAASEyvZoKuueYazZ49e599DjrooG7bJ0+erL1792rz5s0aP368ysvLtW3btqg+4eue9hFJUnFxsYqLi3szbCAreb32uT6BQPf7ggzDvu/1pn9sAAAA+axXIWjEiBEaMWJEn77R+vXrVVRUpJEjR0qSpkyZov/8z//Unj17NHDgQEnSihUrNH78+B6XwgH5xOGQFi+2q8AZRnQQMgz7z7o6iiIAAAAkW0oKI6xdu1Z1dXV644039N577+mRRx7R1VdfrW984xuRgDNr1iwNGjRIl156qd5++209/vjjWrx4sebPn5+KIQFZyeezy2BXVES3u92UxwYAAEgVw7K6W4jTP6+99pq+973vacOGDWpvb9e4ceP0zW9+U/Pnz49ayvbmm29q7ty5evXVV3XggQfqqquu0nXXXder79XW1qbS0lK1trbK6XQm+0cBEhYK2ZXcgkF7H4/Xm/gsTn9eCwAAAFui2SAlISidCEHIBqZpn/nTueS1220vd2M2BwAAID0SzQYpPScIKASmae/r6XrmTyBgt5tmZsYFAACA7hGCgH4IhewZoO7mU8NttbV2PwAAAGQHQhDQD35/7AxQZ5YltbTY/QAAAJAdCEFAPwSDye0HAACA1CMEAf3gciW3HwAAAFKPEAT0g9drV4ELH27alWFIHo/dDwAAANmBEAT0g8Nhl8GWYoNQ+LqujjN/AAAAsgkhCOgnn09qaJAqKqLb3W67nXOCAAAAssuATA8AyDahkF3NLRi09/J4vfFncnw+qbq6968DAABA+hGCgE5M0z73p3PZa7fbXvIWb0bH4ZCqqlI6PAAAACQBy+GAL5imVFMTe+5PIGC3m2ZmxgUAAIDkIgQBspfAzZtnH27aVbitttbuBwAAgNxGCAJk7+XpOgPUmWVJLS12PwAAAOQ2QhAgu5hBMvsBAAAgexGCANnV3JLZDwAAANmLEATILmftdsceeBpmGJLHY/cDAABAbiMEAbLLWy9ebP9z1yAUvq6r49wfAACAfEAIAr7g80kNDVJFRXS72223xzsnCAAAALmBw1KRt0Ihu5pbMGjv5fF648/k+HxSdXXvXwcAAIDcQQhCXjJN+9yfzmWv3W57yVu8GR2HQ6qqSunwAAAAkEEsh0PeMU2ppib23J9AwG43zcyMCwAAANmBEIS8EgrZM0CWFXsv3FZba/cDAABAYSIEIa/4/bEzQJ1ZltTSYvcDAABAYSIEIa8Eg8ntBwAAgPxDCEJecbmS2w8AAAD5hxCEvOL12lXguh54GmYYksdj9wMAAEBhIgQhrzgcdhlsKTYIha/r6jj3BwAAoJARgpB3fD6poUGqqIhud7vt9njnBAEAACC/cVgqslooZFdyCwbtfTxeb2KzOD6fVF3dt9cCAAAgvxGCkLVM0z7zp3PJa7fbXu6WyGyOwyFVVaVseAAAAMhRLIdDVjJNqaYm9syfQMBuN83MjAsAAAC5jxCErBMK2TNAlhV7L9xWW2v3AwAAAHqLEISs4/fHzgB1ZllSS4vdDwAAAOgtQhCyTjCY3H4AAABAZ4QgZB2XK7n9AAAAgM4IQcg6Xq9dBa7rYadhhiF5PHY/AAAAoLcIQcg6DoddBluKDULh67o6zvwBAABA3xCCkJV8PqmhQaqoiG53u+32RM4JAgAAALrDYalIi1DIruYWDNp7ebze+DM5Pp9UXd371wEAAAD7QghCypmmfe5P57LXbre95C3ejI7DIVVVpXR4AAAAKDAsh0NKmaZUUxN77k8gYLebZmbGBQAAgMJFCELKhEL2DJBlxd4Lt9XW2v0AAACAdCEEIWX8/tgZoM4sS2ppsfsBAAAA6UIIQsoEg8ntBwAAACQDIQgp43Iltx8AAACQDIQgpIzXa1eB63rgaZhhSB6P3Q8AAABIF0IQUsbhsMtgS7FBKHxdV8e5PwAAAEgvQhBSyueTGhqkiorodrfbbo93ThAAAACQbByWil4JhexqbsGgvZfH640/k+PzSdXVvX8dAAAAkAqEICTMNO1zfzqXvXa77SVv8WZ0HA6pqiqlwwMAAAASwnI4JMQ0pZqa2HN/AgG73TQzMy4AAACgt1ISgtasWSPDMLr9evXVVyVJmzdv7vb+yy+/nIohoR9CIXsGyLJi74XbamvtfgAAAEC2S8lyuKlTpyrY5QTMH//4x1q1apWOP/74qPaVK1fqiCOOiFwPHz48FUNCP/j9sTNAnVmW1NJi92PJGwAAALJdSkLQoEGDVF5eHrnes2eP/vCHP+iqq66S0aVW8vDhw6P6Ivt0ybP97gcAAABkUlr2BC1btkwffvihvvWtb8XcO+usszRy5EideOKJWrZsWdz3am9vV1tbW9QXUsvlSm4/AAAAIJPSEoIefPBBzZgxQ263O9I2dOhQ3XnnnXriiSf0f//3fzrxxBN19tlnxw1CCxcuVGlpaeTL4/GkevgFz+u1q8B1PfA0zDAkj8fuBwAAAGQ7w7K62+7eveuvv14//elP99mnsbFREyZMiFx/8MEHGjt2rJYuXapzzz13n6+9+OKL1dTUJL/f32Of9vZ2tbe3R67b2trk8XjU2toqp9OZ4E+C3gpXh5OiCySEgxEHnwIAACDT2traVFpaGjcb9GpP0DXXXKPZs2fvs89BBx0UdV1fX6/hw4frrLPOivv+kydP1ooVK/bZp7i4WMXFxXHfC8nl89lBp7tzgurqCEAAAADIHb0KQSNGjNCIESMS7m9Zlurr63XxxRdr4MCBcfuvX79eLjaWpFwoZFdyCwbtfTxer32YaTw+n1Rd3bfXAgAAANkiJdXhwp577jk1NTXpO9/5Tsy9hx9+WIMGDdKxxx4rSTJNUw899JAeeOCBVA6p4Jlm97M5ixcnNpvjcFAGGwAAALktpSHowQcf1NSpU6P2CHV266236v3339eAAQM0YcIEPf7446oJbzxB0oX39XTdBRYI2O3s6wEAAEAh6FVhhGyU6OanQhcKSZWVPR96ahj2jFBTE8vbAAAAkJsSzQZpKZGNzPP7ew5Akj071NJi9wMAAADyGSGoQASDye0HAAAA5CpCUIFItOgexfkAAACQ7whBBcLrtff8hA837cowJI/H7gcAAADkM0JQgXA47DLYUmwQCl/X1VEUAQAAAPmPEFRAfD67DHZFRXS72015bAAAABSOlJ4ThNQKhexqbsGgvZfH640/k+PzSdXVvX8dAAAAkC8IQTnKNKV586LLXrvd9pK3eDM6DodUVZXS4QEAAABZi+VwOcg0pZqa2HN/AgG73TQzMy4AAAAgFxCCckwoZM8AWVbsvXBbba3dDwAAAEAsQlCO8ftjZ4A6syyppcXuBwAAACAWISjHBIPJ7QcAAAAUGkJQjnG5ktsPAAAAKDSEoBzj9dpV4LoeeBpmGJLHY/cDAAAAEIsQlGMcDrsMthQbhMLXdXWc+wMAAAD0hBCUg3w+qaFBqqiIbne77fZ45wQBAAAAhYzDUjMsFLIruQWD9j4erzexWRyfT6qu7ttrAQAAgEJGCMog07TP/Olc8trttpe7JTKb43BIVVUpGx4AAACQl1gOlyGmKdXUxJ75EwjY7aaZmXEBAAAA+Y4QlAGhkD0DZFmx98JttbV2PwAAAADJRQjKAL8/dgaoM8uSWlrsfgAAAACSixCUAcFgcvsBAAAASBwhKANcruT2AwAAAJA4QlAGeL12Fbiuh52GGYbk8dj9AAAAACQXISgDHA67DLYUG4TC13V1nPkDAAAApAIhKEN8PqmhQaqoiG53u+32RM4JAgAAANB7HJaaJKGQXc0tGLT38ni98WdyfD6purr3rwMAAADQd4SgJDBN+9yfzmWv3W57yVu8GR2HQ6qqSunwAAAAAHTCcrh+Mk2ppib23J9AwG43zcyMCwAAAED3CEH9EArZM0CWFXsv3FZba/cDAAAAkB0IQf3g98fOAHVmWVJLi90PAAAAQHYgBPVDMJjcfgAAAABSjxDUDy5XcvsBAAAASD1CUD94vXYVuK4HnoYZhuTx2P0AAAAAZAdCUD84HHYZbCk2CIWv6+o49wcAAADIJoSgfvL5pIYGqaIiut3tttvjnRMEAAAAIL04LDUJfD6putquAhcM2nuAvF5mgAAAAIBsRAhKEodDqqrK9CgAAAAAxMNyOAAAAAAFhRAEAAAAoKAQggAAAAAUFEIQAAAAgIJCCAIAAABQUAhBAAAAAAoKIQgAAABAQSEEAQAAACgohCAAAAAABYUQBAAAAKCgEIIAAAAAFBRCEAAAAICCQggCAAAAUFAIQQAAAAAKCiEIAAAAQEEhBAEAAAAoKAMyPYD+sixLktTW1pbhkQAAAADIpHAmCGeEnuR8CNq5c6ckyePxZHgkAAAAALLBzp07VVpa2uN9w4oXk7JcR0eHtmzZopKSEhmGkdGxtLW1yePxqKWlRU6nM6NjyWd8zunB55wefM7pweecenzG6cHnnB58zumRis/Zsizt3LlTo0ePVlFRzzt/cn4mqKioSG63O9PDiOJ0OvkfTBrwOacHn3N68DmnB59z6vEZpwefc3rwOadHsj/nfc0AhVEYAQAAAEBBIQQBAAAAKCiEoCQqLi7WjTfeqOLi4kwPJa/xOacHn3N68DmnB59z6vEZpwefc3rwOadHJj/nnC+MAAAAAAC9wUwQAAAAgIJCCAIAAABQUAhBAAAAAAoKIQgAAABAQSEEAQAAACgohKA++slPfqKpU6dqyJAhKisr67ZPc3OzzjjjDA0ZMkQjR47UD37wA+3duzeqz5o1a3TcccepuLhYhxxyiJYsWZL6weeoNWvWyDCMbr9effVVSdLmzZu7vf/yyy9nePS5pbKyMuYzXLRoUVSfN998U16vV/vtt588Ho9uu+22DI02N23evFmXXnqpxo0bp8GDB+vggw/WjTfeqN27d0f14Xnuv3vvvVeVlZXab7/9NHnyZP3lL3/J9JBy2sKFC3XCCSeopKREI0eO1Nlnn62NGzdG9amqqop5bi+//PIMjTg33XTTTTGf4YQJEyL3P//8c82dO1fDhw/X0KFDde6552rbtm0ZHHFu6u7/7wzD0Ny5cyXxLPfFCy+8oDPPPFOjR4+WYRh68skno+5blqUbbrhBLpdLgwcP1vTp0/XOO+9E9fnoo4900UUXyel0qqysTJdeeql27dqV1HESgvpo9+7dOu+883TFFVd0ez8UCumMM87Q7t279ec//1kPP/ywlixZohtuuCHSp6mpSWeccYamTZum9evXq7a2Vt/5znf07LPPpuvHyClTp05VMBiM+vrOd76jcePG6fjjj4/qu3Llyqh+kyZNytCoc9ctt9wS9RleddVVkXttbW069dRTNXbsWK1bt0633367brrpJt1///0ZHHFu2bBhgzo6OvTLX/5Sb7/9tn72s5/pvvvu03/8x3/E9OV57rvHH39c8+fP14033qjXXntNX/rSlzRjxgxt374900PLWc8//7zmzp2rl19+WStWrNCePXt06qmn6pNPPonqN2fOnKjnlr8o6b0jjjgi6jN88cUXI/euvvpq/fGPf9QTTzyh559/Xlu2bJHP58vgaHPTq6++GvUZr1ixQpJ03nnnRfrwLPfOJ598oi996Uu69957u71/22236e6779Z9992nV155Rfvvv79mzJihzz//PNLnoosu0ttvv60VK1boqaee0gsvvKDLLrssuQO10C/19fVWaWlpTPvTTz9tFRUVWVu3bo20/eIXv7CcTqfV3t5uWZZl/fCHP7SOOOKIqNddcMEF1owZM1I65nyxe/dua8SIEdYtt9wSaWtqarIkWa+//nrmBpYHxo4da/3sZz/r8f7Pf/5z64ADDog8y5ZlWdddd501fvz4NIwuf912223WuHHjItc8z/335S9/2Zo7d27kOhQKWaNHj7YWLlyYwVHll+3bt1uSrOeffz7SdvLJJ1vz5s3L3KDywI033mh96Utf6vbejh07rIEDB1pPPPFEpK2xsdGSZK1duzZNI8xP8+bNsw4++GCro6PDsiye5f6SZP3+97+PXHd0dFjl5eXW7bffHmnbsWOHVVxcbP32t7+1LMuy/v73v1uSrFdffTXS55lnnrEMw7ACgUDSxsZMUIqsXbtWRx11lEaNGhVpmzFjhtra2vT2229H+kyfPj3qdTNmzNDatWvTOtZctWzZMn344Yf61re+FXPvrLPO0siRI3XiiSdq2bJlGRhd7lu0aJGGDx+uY489VrfffnvUUs61a9fqpJNO0qBBgyJtM2bM0MaNG/Xxxx9nYrh5obW1VcOGDYtp53num927d2vdunVR/50tKirS9OnT+e9sErW2tkpSzLP7yCOP6MADD9SRRx6pBQsW6NNPP83E8HLaO++8o9GjR+uggw7SRRddpObmZknSunXrtGfPnqhne8KECRozZgzPdj/s3r1bv/nNb/Ttb39bhmFE2nmWk6epqUlbt26NenZLS0s1efLkyLO7du1alZWVRa3ymT59uoqKivTKK68kbSwDkvZOiLJ169aoACQpcr1169Z99mlra9Nnn32mwYMHp2ewOerBBx/UjBkz5Ha7I21Dhw7VnXfeqa985SsqKirS7373O5199tl68sknddZZZ2VwtLnl+9//vo477jgNGzZMf/7zn7VgwQIFg0Hdddddkuxnd9y4cVGv6fx8H3DAAWkfc67btGmT7rnnHt1xxx2RNp7n/vnXv/6lUCjU7X9nN2zYkKFR5ZeOjg7V1tbqK1/5io488shI+6xZszR27FiNHj1ab775pq677jpt3LhRpmlmcLS5ZfLkyVqyZInGjx+vYDCom2++WV6vV2+99Za2bt2qQYMGxexJHjVqVOR3DPTek08+qR07dmj27NmRNp7l5Ao/n939d7nz78cjR46Muj9gwAANGzYsqc83IaiT66+/Xj/96U/32aexsTFqYyL6ry+f+wcffKBnn31WS5cujep34IEHav78+ZHrE044QVu2bNHtt99e8L809uZz7vwZHn300Ro0aJC++93vauHChSouLk71UHNaX57nQCCgmTNn6rzzztOcOXMi7TzPyHZz587VW2+9FbVXRVLU2v2jjjpKLpdLp5xyit59910dfPDB6R5mTjrttNMi/3z00Udr8uTJGjt2rJYuXcpfkqbIgw8+qNNOO02jR4+OtPEs5y9CUCfXXHNNVPrvzkEHHZTQe5WXl8dUIApXbSkvL4/82bWSy7Zt2+R0OgvqP3B9+dzr6+s1fPjwhH4RnDx5cmSjYyHrz/M9efJk7d27V5s3b9b48eN7fHalfz/fhaq3n/OWLVs0bdo0TZ06NaHCEjzPiTvwwAPlcDi6fVYL/TlNhiuvvDKyYbnzjHx3Jk+eLMme8eQXx74pKyvTYYcdpk2bNumrX/2qdu/erR07dkTNBvFs993777+vlStXxp3h4Vnun/DzuW3bNrlcrkj7tm3bdMwxx0T6dC1es3fvXn300UdJfb4JQZ2MGDFCI0aMSMp7TZkyRT/5yU+0ffv2yJTeihUr5HQ6dfjhh0f6PP3001GvW7FihaZMmZKUMeSK3n7ulmWpvr5eF198sQYOHBi3//r166P+h1ao+vN8r1+/XkVFRZFnecqUKfrP//xP7dmzJ/LvYMWKFRo/fnzBL4XrzeccCAQ0bdo0TZo0SfX19Soqir9Nk+c5cYMGDdKkSZO0atUqnX322ZLs5VurVq3SlVdemdnB5TDLsnTVVVfp97//vdasWROzNLY769evlySe3X7YtWuX3n33XX3zm9/UpEmTNHDgQK1atUrnnnuuJGnjxo1qbm4uuN8hkqW+vl4jR47UGWecsc9+PMv9M27cOJWXl2vVqlWR0NPW1qZXXnklUnF5ypQp2rFjh9atWxephvrcc8+po6MjEkKTImklFgrM+++/b73++uvWzTffbA0dOtR6/fXXrddff93auXOnZVmWtXfvXuvII4+0Tj31VGv9+vXW8uXLrREjRlgLFiyIvMd7771nDRkyxPrBD35gNTY2Wvfee6/lcDis5cuXZ+rHygkrV660JFmNjY0x95YsWWI9+uijVmNjo9XY2Gj95Cc/sYqKiqyHHnooAyPNTX/+85+tn/3sZ9b69eutd9991/rNb35jjRgxwrr44osjfXbs2GGNGjXK+uY3v2m99dZb1mOPPWYNGTLE+uUvf5nBkeeWDz74wDrkkEOsU045xfrggw+sYDAY+Qrjee6/xx57zCouLraWLFli/f3vf7cuu+wyq6ysLKpyJ3rniiuusEpLS601a9ZEPbeffvqpZVmWtWnTJuuWW26x/vrXv1pNTU3WH/7wB+uggw6yTjrppAyPPLdcc8011po1a6ympibrpZdesqZPn24deOCB1vbt2y3LsqzLL7/cGjNmjPXcc89Zf/3rX60pU6ZYU6ZMyfCoc1MoFLLGjBljXXfddVHtPMt9s3PnzsjvxZKsu+66y3r99det999/37Isy1q0aJFVVlZm/eEPf7DefPNNq7q62ho3bpz12WefRd5j5syZ1rHHHmu98sor1osvvmgdeuih1oUXXpjUcRKC+uiSSy6xJMV8rV69OtJn8+bN1mmnnWYNHjzYOvDAA61rrrnG2rNnT9T7rF692jrmmGOsQYMGWQcddJBVX1+f3h8kB1144YXW1KlTu723ZMkSa+LEidaQIUMsp9NpffnLX44qIYr41q1bZ02ePNkqLS219ttvP2vixInWf//3f1uff/55VL833njDOvHEE63i4mKroqLCWrRoUYZGnJvq6+u7/W9I57+b4nlOjnvuuccaM2aMNWjQIOvLX/6y9fLLL2d6SDmtp+c2/P9fzc3N1kknnWQNGzbMKi4utg455BDrBz/4gdXa2prZgeeYCy64wHK5XNagQYOsiooK64ILLrA2bdoUuf/ZZ59Z3/ve96wDDjjAGjJkiHXOOedE/SUKEvfss89akqyNGzdGtfMs983q1au7/W/EJZdcYlmWXSb7xz/+sTVq1CiruLjYOuWUU2I++w8//NC68MILraFDh1pOp9P61re+FZloSBbDsiwrefNKAAAAAJDdOCcIAAAAQEEhBAEAAAAoKIQgAAAAAAWFEAQAAACgoBCCAAAAABQUQhAAAACAgkIIAgAAAFBQCEEAAAAACgohCAAAAEBBIQQBAAAAKCiEIAAAAAAF5f8Hd9AA7XRFMXgAAAAASUVORK5CYII=\n"
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "67.2385\n",
      "4603.704\n"
     ]
    }
   ],
   "source": [
    "y_pred3 = model3.predict(X_test)\n",
    "plot_predictions(X_train, y_train, X_test, y_test, y_pred2)\n",
    "\n",
    "# evaluation metrics\n",
    "mae_3 = mae(y_test, y_pred3)\n",
    "mse_3 = mse(y_test, y_pred3)\n",
    "print(mae_3)\n",
    "print(mse_3)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# [Module 46: Comparing and tracking experiments](https://youtu.be/tpCFfeUEGs8?t=24709)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_15\"\n",
      "_________________________________________________________________\n",
      " Layer (type)                Output Shape              Param #   \n",
      "=================================================================\n",
      " dense_24 (Dense)            (None, 10)                20        \n",
      "                                                                 \n",
      " dense_25 (Dense)            (None, 1)                 11        \n",
      "                                                                 \n",
      "=================================================================\n",
      "Total params: 31\n",
      "Trainable params: 31\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "None\n",
      "    Model        MAE          MSE\n",
      "0  model1   8.599912    80.465492\n",
      "1  model2   2.985097    11.663383\n",
      "2  model3  67.238503  4603.704102\n"
     ]
    }
   ],
   "source": [
    "model_results = [[\"model1\", mae_1, mse_1],\n",
    "                 [\"model2\", mae_2, mse_2],\n",
    "                 [\"model3\", mae_3, mse_3]]\n",
    "all_results = pd.DataFrame(model_results, columns=[\"Model\", \"MAE\", \"MSE\"])\n",
    "print(model2.summary())\n",
    "print(all_results)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "### Tracking experiments\n",
    "- [TensorBoard](https://tensorboard.dev/): a component of the TensorFlow library to help track modelling experiments.\n",
    "- [Weights & Biases](https://www.wandb.com/): a tool for tracking all kinds of machine learning experiments (the good news for Weights & Biases is it plugs into TensorBoard)."
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "\n",
    "# [Module 47: Saving our models](https://youtu.be/tpCFfeUEGs8?t=25328)\n",
    "\n",
    "### Formats we can use to save our model:\n",
    "1. the SaveModel format (default)\n",
    "2. the HDF5 format"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "WARNING:absl:Found untraced functions such as _update_step_xla while saving (showing 1 of 1). These functions will not be directly callable after loading.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Assets written to: module45_saved_model/assets\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Assets written to: module45_saved_model/assets\n"
     ]
    }
   ],
   "source": [
    "# save model with SaveModel format\n",
    "model2.save(\"module45_saved_model\")  # HDF5 is better for if going use model in other programs outside tensorflow. Add h5 extension to do so."
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# [Module 48: Loading our model](https://youtu.be/tpCFfeUEGs8?t=25892)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_15\"\n",
      "_________________________________________________________________\n",
      " Layer (type)                Output Shape              Param #   \n",
      "=================================================================\n",
      " dense_24 (Dense)            (None, 10)                20        \n",
      "                                                                 \n",
      " dense_25 (Dense)            (None, 1)                 11        \n",
      "                                                                 \n",
      "=================================================================\n",
      "Total params: 31\n",
      "Trainable params: 31\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "None\n",
      "1/1 [==============================] - 0s 10ms/step\n",
      "1/1 [==============================] - 0s 22ms/step\n",
      "[[ True]\n",
      " [ True]\n",
      " [ True]\n",
      " [ True]\n",
      " [ True]\n",
      " [ True]\n",
      " [ True]\n",
      " [ True]\n",
      " [ True]\n",
      " [ True]]\n"
     ]
    }
   ],
   "source": [
    "from_disc = tf.keras.models.load_model(\"module45_saved_model\")\n",
    "# check that it's the same as summary for model2\n",
    "print(from_disc.summary())\n",
    "\n",
    "# check if predictions are the same\n",
    "model2_pred_48 = model2.predict(X_test)\n",
    "from_disc_pred = from_disc.predict(X_test)\n",
    "print(model2_pred_48 == from_disc_pred)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# [Module 49 downloading files from google collab](https://youtu.be/tpCFfeUEGs8?t=26509)\n",
    "I didn't bother taking notes for this because I'm not using google collab. I watched the video and it won't be hard to learn if I want to use Google collab later."
   ],
   "metadata": {
    "collapsed": false
   }
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}