Mercari.ipynb

{
  "cells": [
    {
      "metadata": {
        "_uuid": "8f2839f25d086af736a60e9eeb907d3b93b6e0e5",
        "_cell_guid": "b1076dfc-b9ad-4769-8c92-a6c4dae69d19"
      },
      "cell_type": "markdown",
      "source": "# Mercari Price Suggestion Challenge\n[A Santosh Kumar](https://github.com/santoshk17) - Feb 2018"
    },
    {
      "metadata": {
        "_uuid": "d629ff2d2480ee46fbb7e2d37f6b5fab8052498a",
        "collapsed": true,
        "_cell_guid": "79c7e3d0-c299-4dcb-8224-4455121ee9b0"
      },
      "cell_type": "markdown",
      "source": "### Contents\n1. [Introduction](#intro)\n    * [Packages Needed](#package)\n2. [Exploratory Data Analysis and Feature Engineering](#eda)\n    * [Loading data](#loading_data)\n    * [Price Distribution](#price_distribution)\n    * [Word Count](#word_count)\n    * [Handling Missing Data](#Handling_Missing_Data)\n    * [Categorial Data](#categorical_data)\n    * [Adding SubCategories](#Adding_SubCategories)\n    * [Retrieving Brand Name](#Retrieving_Brand_Name)\n    * [Text to Sequence](#Text_to_Sequence)\n    * [RMSLE](#rmsle)\n \n3. [RNN Model](#rnn_model)\n    * [Model Building](#Model_Building)\n    * [Model Evaluation](#Model_Evaluation)\n    * [Model Output](#Model_Output)\n    \n4. [Ridge Models](#ridge_models)\n     * [Data Preprocessing](#Data_Preprocessing)\n     * [Model Building](#Model_Building1)\n     * [Model Evaluation](#Model_Evaluation1)\n5. [Ensembling](#ensembling)\n6. [Conclusion](#Conclusion)\n7. [Refrences](#Refrences)\n   \n\n\n\n<a id=\"intro\"></a>\n## <center>1. Introduction</center>"
    },
    {
      "metadata": {
        "_uuid": "4346e4428877c2e17e7d45f5d55055a37e6f8fe9",
        "_cell_guid": "379d5391-da52-4828-8317-bd6c69259470"
      },
      "cell_type": "markdown",
      "source": "Mercari, Japan’s biggest community-powered shopping app, knows this problem deeply. They’d like to offer pricing suggestions to sellers, but this is tough because their sellers are enabled to put just about anything, or any bundle of things, on Mercari's marketplace.\n\nIn this competition, Mercari’s challenging you to build an algorithm that automatically suggests the right product prices. You’ll be provided user-inputted text descriptions of their products, including details like product category name, brand name, and item condition."
    },
    {
      "metadata": {
        "_uuid": "f541cededebb6889c5c6bf5e402a3c304ae421ee",
        "_cell_guid": "a9a86f36-e390-40f5-95c7-b186646f5ca6"
      },
      "cell_type": "markdown",
      "source": "<a id=\"package\"></a>\n#### Packages Needed"
    },
    {
      "metadata": {
        "_uuid": "2c309b166f09c7ce5b913e71590748c5a9c24016",
        "_cell_guid": "9babbf4a-6fe1-4727-a372-bd7a99bb0460",
        "trusted": true
      },
      "cell_type": "code",
      "source": "import gc\nimport os\nimport re\nimport sys\nimport time\nimport wordbatch\nimport numpy as np\nimport pandas as pd\nimport seaborn as sns\nfrom sklearn.preprocessing import LabelEncoder\nfrom keras.preprocessing.sequence import pad_sequences\nfrom sklearn.pipeline import FeatureUnion\nfrom sklearn.feature_extraction.text import CountVectorizer, TfidfVectorizer\n\nimport matplotlib.pyplot as plt\n\n\n\n# for text data\nfrom keras.preprocessing.text import Tokenizer\nfrom wordbatch.extractors import WordBag, WordHash\nfrom wordbatch.models import FTRL, FM_FTRL\nfrom nltk.corpus import stopwords\nfrom scipy.sparse import csr_matrix, hstack\n\n#  for splitting data in to train and validation sets\nfrom sklearn.model_selection import train_test_split\n\n# RNN Model related\nfrom keras.models import Model\nfrom keras.optimizers import Adam\nfrom keras.layers import Input, Dropout, Dense, concatenate, GRU, Embedding, Flatten, Activation\n",
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": "/opt/conda/lib/python3.6/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n  from ._conv import register_converters as _register_converters\nUsing TensorFlow backend.\n",
          "name": "stderr"
        }
      ]
    },
    {
      "metadata": {
        "_uuid": "6857b764e97886ecfc13ffb0cfccd5ed80c943a4",
        "collapsed": true,
        "_cell_guid": "8b39ab56-d609-4d32-8fdc-af21d7684b24",
        "trusted": true
      },
      "cell_type": "code",
      "source": "\nsys.path.insert(0, '../input/wordbatch/wordbatch/')\n",
      "execution_count": 2,
      "outputs": []
    },
    {
      "metadata": {
        "_uuid": "860304c7f31d11b9fbf1d48300290423672eac0d",
        "collapsed": true,
        "_cell_guid": "0a224d35-7ab1-4c61-a301-6d8a2ea8dc16",
        "trusted": true
      },
      "cell_type": "code",
      "source": "trainDF = pd.read_table('../input/mercari-price-suggestion-challenge/train.tsv', engine='c')\ntestDF = pd.read_table('../input/mercari-price-suggestion-challenge/test.tsv', engine='c')",
      "execution_count": 3,
      "outputs": []
    },
    {
      "metadata": {
        "_uuid": "90866c85cfcbc0e3ec78fa74697af4ab66bea487",
        "_cell_guid": "b481a9b1-83a1-41a9-b7bf-8a68dce477b4",
        "trusted": true
      },
      "cell_type": "code",
      "source": "print('Train shape: ', trainDF.shape)\nprint('Test shape: ', testDF.shape)",
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "stream",
          "text": "Train shape:  (1482535, 8)\nTest shape:  (693359, 7)\n",
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "_uuid": "f7abb78a2236ef357ddc627a079b7e78f3a118bb",
        "_cell_guid": "d5857076-a36d-4dab-b2d5-a755dba85345",
        "trusted": true
      },
      "cell_type": "code",
      "source": "trainDF.head()",
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 5,
          "data": {
            "text/plain": "   train_id                                 name  item_condition_id  \\\n0         0  MLB Cincinnati Reds T Shirt Size XL                  3   \n1         1     Razer BlackWidow Chroma Keyboard                  3   \n2         2                       AVA-VIV Blouse                  1   \n3         3                Leather Horse Statues                  1   \n4         4                 24K GOLD plated rose                  1   \n\n                                       category_name brand_name  price  \\\n0                                  Men/Tops/T-shirts        NaN   10.0   \n1  Electronics/Computers & Tablets/Components & P...      Razer   52.0   \n2                        Women/Tops & Blouses/Blouse     Target   10.0   \n3                 Home/Home Décor/Home Décor Accents        NaN   35.0   \n4                            Women/Jewelry/Necklaces        NaN   44.0   \n\n   shipping                                   item_description  \n0         1                                 No description yet  \n1         0  This keyboard is in great condition and works ...  \n2         1  Adorable top with a hint of lace and a key hol...  \n3         1  New with tags. Leather horses. Retail for [rm]...  \n4         0          Complete with certificate of authenticity  ",
            "text/html": "<div>\n<style scoped>\n    .dataframe tbody tr th:only-of-type {\n        vertical-align: middle;\n    }\n\n    .dataframe tbody tr th {\n        vertical-align: top;\n    }\n\n    .dataframe thead th {\n        text-align: right;\n    }\n</style>\n<table border=\"1\" class=\"dataframe\">\n  <thead>\n    <tr style=\"text-align: right;\">\n      <th></th>\n      <th>train_id</th>\n      <th>name</th>\n      <th>item_condition_id</th>\n      <th>category_name</th>\n      <th>brand_name</th>\n      <th>price</th>\n      <th>shipping</th>\n      <th>item_description</th>\n    </tr>\n  </thead>\n  <tbody>\n    <tr>\n      <th>0</th>\n      <td>0</td>\n      <td>MLB Cincinnati Reds T Shirt Size XL</td>\n      <td>3</td>\n      <td>Men/Tops/T-shirts</td>\n      <td>NaN</td>\n      <td>10.0</td>\n      <td>1</td>\n      <td>No description yet</td>\n    </tr>\n    <tr>\n      <th>1</th>\n      <td>1</td>\n      <td>Razer BlackWidow Chroma Keyboard</td>\n      <td>3</td>\n      <td>Electronics/Computers &amp; Tablets/Components &amp; P...</td>\n      <td>Razer</td>\n      <td>52.0</td>\n      <td>0</td>\n      <td>This keyboard is in great condition and works ...</td>\n    </tr>\n    <tr>\n      <th>2</th>\n      <td>2</td>\n      <td>AVA-VIV Blouse</td>\n      <td>1</td>\n      <td>Women/Tops &amp; Blouses/Blouse</td>\n      <td>Target</td>\n      <td>10.0</td>\n      <td>1</td>\n      <td>Adorable top with a hint of lace and a key hol...</td>\n    </tr>\n    <tr>\n      <th>3</th>\n      <td>3</td>\n      <td>Leather Horse Statues</td>\n      <td>1</td>\n      <td>Home/Home Décor/Home Décor Accents</td>\n      <td>NaN</td>\n      <td>35.0</td>\n      <td>1</td>\n      <td>New with tags. Leather horses. Retail for [rm]...</td>\n    </tr>\n    <tr>\n      <th>4</th>\n      <td>4</td>\n      <td>24K GOLD plated rose</td>\n      <td>1</td>\n      <td>Women/Jewelry/Necklaces</td>\n      <td>NaN</td>\n      <td>44.0</td>\n      <td>0</td>\n      <td>Complete with certificate of authenticity</td>\n    </tr>\n  </tbody>\n</table>\n</div>"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "_uuid": "4c0ed569c8c9dd971df673c1c170c55c92a3eb00",
        "_cell_guid": "6f03c2c0-0ad3-47a4-8158-8c2275ceb4b4"
      },
      "cell_type": "markdown",
      "source": "<a id=\"price_distribution\"></a>\n#### Price Distribution"
    },
    {
      "metadata": {
        "_uuid": "939ff549d9c1e5cf08195ea4deda8174f61ec299",
        "_cell_guid": "c8c69290-f089-43fa-9f05-32e0d13a2c78",
        "trusted": true
      },
      "cell_type": "code",
      "source": "trainDF.price.describe()",
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 6,
          "data": {
            "text/plain": "count    1.482535e+06\nmean     2.673752e+01\nstd      3.858607e+01\nmin      0.000000e+00\n25%      1.000000e+01\n50%      1.700000e+01\n75%      2.900000e+01\nmax      2.009000e+03\nName: price, dtype: float64"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "_uuid": "1bff636de5d9e757090d26ee55b9f9a18d75d8c8",
        "_cell_guid": "f34ceec0-73ba-47af-89f0-bd8841ef1d61",
        "trusted": true
      },
      "cell_type": "code",
      "source": "plt.subplot(1, 2, 1)\n(trainDF['price']).plot.hist(bins=50, figsize=(20,10), edgecolor='white',range=[0,250])\nplt.xlabel('price+', fontsize=17)\nplt.ylabel('frequency', fontsize=17)\nplt.tick_params(labelsize=15)\nplt.title('Price Distribution - Training Set', fontsize=17)\n\n\nplt.subplot(1, 2, 2)\nnp.log(trainDF['price']+1).plot.hist(bins=50, figsize=(20,10), edgecolor='white')\nplt.xlabel('log(price+1)', fontsize=17)\nplt.ylabel('frequency', fontsize=17)\nplt.tick_params(labelsize=15)\nplt.title('Log(Price) Distribution - Training Set', fontsize=17)\nplt.show()  ",
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": "<matplotlib.figure.Figure at 0x7fa121f1ab38>",
            "image/png": "iVBORw0KGgoAAAANSUhEUgAABMQAAAJtCAYAAADZ86zmAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMS4wLCBo\ndHRwOi8vbWF0cGxvdGxpYi5vcmcvpW3flQAAIABJREFUeJzs3Xm4ZFV57/HvzwYEFAQZYjRoi2hw\nTEwwUTRxRAXEgUBQrwYxisQpijHBgStiDDgEomJENAkqUQJeHECRCAJqECMkwTCKQ4uzjTYiQws2\n7/1j7aJ3V9eZuuv0OU19P89Tz+lae+21V9Wp03vVu9d6d6oKSZIkSZIkaVLcaaE7IEmSJEmSJG1I\nBsQkSZIkSZI0UQyISZIkSZIkaaIYEJMkSZIkSdJEMSAmSZIkSZKkiWJATJIkSZIkSRPFgJi0iCVZ\nmqSSvGCh+zKTJC/o+rp0Ax2vkhwx4viP2UDH32h+NxtakqOTrFzHfTfv3tfDxt0vSZL6kvxGkhuS\n7DPmdh/XncseN852e+1vkeSnSQ6ej/Z7x1mW5MT5PEbvWEckqRHHP3tDHL873olJlm2o420s1nds\ntj7jQmm+GRCT5kkvQDN4rEry4yQnJ3nAQvdvOiP6vjLJj5Kck+Svk9x9jMd6ZpL/O672xq17L165\n0P1YV0O/x+keRyx0XxeLJLsnOSPJ97rP/g+SfCHJG9axvV26gf6Dxt1XSdqYbeiLWSMcDnwbOKPX\npyOGzo+/7s4HH0zymwvUzzVU1c3APwD/N8nms9mnCy4NXtNtSX6R5PIuCPS4cfUtyTbde/jH42pz\nnLoLmkckedhC92VdjPh8TvlY6L4uFknumuTwJF9P8ssk13Wf/X9Osts6tvmiJC8dd1+14aXKvxVp\nPnQzh/4FeAvwDeDOwO8Dfw78EnhoVf1ohjbS7XdrVa2a1w6vedwXsGbfNwF2AB4D7A38DNivqr7U\n22cJsCnwq5rDfyzdlcfnVdUmc+zj5sCvq+rXQ33+o6r68lzamuE45wG/VVW7DJUvyO9mrpI8b6jo\nYODRwIFD5V+vqq+P6ZibAEuq6lfruP/mLND7muTZwEeBS4GTgZ8COwG7A7tX1V3Woc2nAmcC+1fV\nx8fYXUnaqM3XuXuWx7478EPgVVV1fK/8COBNwCuBFcCWwB8B/4cWPHtoF5Caru07AZsBt1TVbfPU\n/+2BHwGHVNU/zaL+MuAG4Oiu6K7AA4BnAUuBjwAH9c+9Se4M3FZVt86hX0uB7wCHV9XfzmG/TYBN\nqmplr2wZ8M2qetJs25nFcR4HnAs8v6pOGtq2KXCndR2/bAhdIK8fzNseOBb4OPCpft3h17eex13n\nsdn6jgvXR/cZ/iqwK3AScDHtb/O3aZ/9d1XV0VO3MGW7FwJ3raqHjLG7WgBz+gIqaZ38e2+Q909J\nrqJd1XsBcNSoHboTcqrqFmAhpxj/+9AA9R1Jfg84C/hkkgdV1U8AuhPkvAYwBgPMqlrZHzAthC7o\nt+inf48Y7D2JFtiZ1SCpC/xtPtPgf+iYvwZ+PaeOrrn/Qr6vfwtcCTxieOCW5F4L0yVJ0jz4P8AS\nWiBhlE9U1fe7f5+Q5FrgVcAzgY+N2iHJFrQLg7cxz2OEqro2yTnAQcCMAbHOj0eMC/4aeC/tgtn3\ngNtnQ2+IAEaSu1TVjes7dhiHuQT+Fkp38fL2C5hdAPJY4JI5jO3uUlU3zvG46/x5XuDf7f7A7wAH\nVtWH+xuSvIIWUNQEc8mktOENciHcF9bIM/FnSd7QXQ1bCTwoU+SpSst58d4k1yT5VZLvJ/lo/wt7\nkk279q7s6vw4yQlZz+WOVfVfwKuBuwMv7x1vrRxiSXZO8rG0JWeDPnw2yUO77efRZiotGZ7i3Xvt\nb0zykiRXAr8Cntptn2qZ352THJvkJ0lu7I43PLtrZI6I4dfQ1XkscL9e/5YN9e8FQ208OMmnuunY\nNyW5MMnThuoMfufPT/KaJN9NW5p3YRdwXBBZnSPiuCQHJPk67T0/sNv+orRlsz/ufp/fSnJkF8Dt\nt7NWrohunzPSliNe2L3ea5K8fKjeWnkqkhzSle2e5B3d7/bmJGcmufeI1/GnSS7rjnFlkgNH9WnE\nfncCdga+OupLQFX9YMQ+f5DkM93v++YkX02yd7/vtNlhAKf2PkfPnq4vkqTVunPD36Ut+7ul+/l3\nabM/+vXSjX2+1/2ffGGSRyc5rxtz9D0L+K+qunaW3Rgevw3GDE9KckySHwI3AltnihxiSe6X5CNp\naSh+leQ7ST6QZKtena2SvL3bdkt3rnxHWrBt2L8Dj07yG7N8DWvpghUvA64C/jLJ1r2+rJVDLG1M\ndkla7rXr05aevanb9jja7DCAt/TOeUd0209MW4K6U5LTklwHfLnbtlYOsd4x/zjJf3a/02VJDh3a\nPnJMNvwauu3ndps+0uvfC3r9Wza0/53S0oVc1f3Ofpg2Bt9mqN55Sb6ZlibhrLQx6E+78ceCfefu\njl9Jdk3y4SQ/A77Zbbtfkvd1Y6Ub05bRnpXkD4baWK+x2agxWGY5Luzqbt/1/bquj6cmuc9wn6Yw\n+A5w/vCGqlo1uLDfO9adk7wpyTe63/ePkvxj//ed5MfAHwIP7n2GrpyhH1qknCEmbXiD/5iHB2B/\nTQtSv5d2FeXnjAhad4OerwL3BD4IXEJbzrh31/YPkgT4f8AetKuGX6d90X8F8AdJHrmes3D+DfgA\n8BRa/o21pAVJ/h24C/A+4PvAb9ACTL8N/C/w1u41jlrCN/Bs4G7A+2nvybIZ+vb3tPfvKNr78pfA\neUkeVlU/n9WrW+1VXTvbAn/Vld0wVeW03HAXALfSZgFeT5sJ+OkkB1TVqUO7/CVt2eW7af8fvxb4\nRJJdFvgq5eNp7/t7actJLuvKXwn8D/BZ2qD/j4E3AveiLQWeyf2AT9CWx3wYeC7wniT/W1VrDVRG\neHd33L8F7gG8BvhQ118AkjyDttTxctpV7rsA7wTWCmYNq6rbklwDPCnJfarqu9PVT/JE2nvxP7Tl\nxbcAzwFOT7JfVZ0GnAO8g/a7fTfwtW73r8zi9UrSxOvGNKcBe9KWPH2Ftoz9dcBDgX5C/Ld25Z8H\nPk0bF51OGz98v9fmJsAf0M5FszXV+O0faKkw3kZbXnnLFK/jgcB/0M73J9BSUvwWLTC3HfDLtGVp\nXwDu39X5Jm153KuAhyTZaygtxX92Px9DG/etk6r6dZJ/BY7s2vrsFK/hIOB42rn8fUBoY7o/6qpc\nQbtoOryEr5+SIbSVBl8HDmPmCRr3of0uT6Qt69wX+Pskm1bV22b9IpsvAn8HvL7r/wVd+QVT7gH/\nCLyElmfu3cCDgEOARyZ5VLeaY2ArWuD0TNp79BTgb2hBwvfPsa/j9nHaGPpw2ucU4FG0v6VTgWuA\nHYEX08bND6+qq2bR7oxjs2nMOC5MS8nyWVramRNoKS2eDHxyFu3D6u8NByU5crolzF3g8tO0z/MH\nu2PdnxYwfkSS3bvx+ctof+93pv1/A/CLWfZHi01V+fDhYx4etEBI0QJV29MCWPvQ/mNeBfxeV+9x\nXb0fAFsNtbG02/aCXtk/d2WPH3HMQV7A53R19hja/uSu/MWz7PtjpqlzCfDzEfss7Z7/Tvd8/xmO\ndSItF9hw+eC130TL4TW8vYAjRhz/SmCLXvkeXflRQ8dcNs3rXtorO4+Wv2Kq/vV/Nx+nBeMe3Cvb\nipZz5Ae03Bj93/m3h/r6zMFnZh4/lyPf727b5t3xV9FypAxv33JE2d92r3nHXtnRwMqhej8e/tzS\nBmQ/Az4yog+H9coO6cq+QMvtMSg/rCu/X6/syu693rpX9tDuNa0c9bqH+vmirs1baAPno7rP0KZD\n9ZYA36J96cpQ+X8C3+qVPbVrc7/5+r368OHDx8b4YHbjjacNn8e78nd05Xt1z3fs/u8+a+hc8edd\nvfN6ZTt3ZX814nhHdNseShu/7UT7ov5z2hf/ew71/eIR54jBef5xvbIvADcDu4445mD89jraKoGH\nDm0/mNHjunt25YfP4r1eBpw9zfZndW29cmifE3vPPwFcOsNxlnbtvHHEthO7be+e6n0f0eei5TYb\nlC3pzs83AdsMHfMFU7zu/msY/G6eN0X/lvWeP6Sr+7Ghei/ryl/aKzuvKztkqO7/AF+bx7+hKd/v\nbvvR3fZTRmwbNa7bsfusv6dXtr5js/UZFw6+07xmaP9Thvs0xevfkhZ8LtqS4I8AfwHce0Tdg4Db\ngMcOlT+92//5vbILZ/pb8LFxPFwyKc2/M4DltC/pn6adVJ5fbelh30eq6pfTNdRdudgX+HxVnTu8\nvbr/oYEDaMGW/+6mGW+flnz1v2hXMJ6wPi+o80tasGcqgyslT01y1/U4zqdrdQ6P2Xh/9fJdVdXn\nabOFnjb1Luuvu4L1VOCzVTWYUUX3O30fbdA6vBzyxFozN9dgltTO89nXWfhSVf3vcGFV3QTttabd\nRWp7Wp+XAL87i3av6H9uu/YuYvav9/ha88reGu9Xkp1pV6o/VFXX947zv7QB24yq6oO0wPX5tOnw\nh9FmOv4oyQG9qo/ojvuvwHa9v7Ftgc8BOye5zyxflyRpaoPz9zuHyt8+tP1JtJv7vHfoXPEh4Lqh\nfQd5g1ZMc9yv08Zv19D+r78OeEZV/XCo3gdqhlnd3fnhcbSx3lpLq4bGb1+hnXP647fBcs3h8dtg\n5vs48iANxqDTje2uA3ZK8qj1PNY/zqHuz2hBDOD2nLXvAbYAnrie/ZjJ4LP1jqHyD9Dei+Gx5a20\nmUV957Pw4zoY8Z4PxnUASbZMsh0tIHQxbUbWbEw7NpvBbMaFe9IuvB4/tO+7ZtO5rs3daf9/3AY8\nj/ZefDfJJ5Ls0Kt+AO3C6mVDf39foQWzx/H9SYuMSyal+fdq2pTbVbSB1RU1+g4t35pFWzvQlg+u\nFawY8gDayWT5FNt3nMWxZrIVqwdPa6mqZUneTlsK+rwkX6FNIf/XOQa4ZvO+9I2a3n0VbZbPfNqB\ntjxvVA6By7uf92X18gaANZbkVdWKtjKEafO8peWB26xXdEvNfTnodEa+50keQ1uO8sih4wNss/Ye\naxm1BHEFsFYesFnuP/giM3i/BgGoq0fsezWrl3RMq6rOAM5Iy00zWI7zauBjSb5XVRfQ/sagTfOf\nyo4j+ixJmpulwPKq+lm/sKqWpyW6v29XNPIcUG054HemaDvTHPc5tOWRv6bdzfEbvcBV32zGKffr\njjWb8dsWzH78Nl3/52oQCJvu4uzRtKDABV2KgXNos8bOmOK9mcq351K3ujuK9wzGekvn0M66GLS/\nxtiuqm5J8k1Wf/YGfjCiryuYeVy32Yg6P681l2Our7U+p90S3TfTbjAxfOOgK2bZ7kxjs7nsO9i/\nPy68D/DDWvsmAKPGeiNVyxP4WuC1Se5JWxb8F7SVGVvQ5Sem/f3dl/n9/qRFxoCYNP8uqtndSnw2\nd/EbDHxmGnTciXbyfsUU26e7IjpzJ9qJ+wHMMLCrqr9J8s+0qcZPop10D0/yjKo6Z5aHm/XdDQeH\nHVE2PGCc6v1bMsdjzdZUv7ep7so50wD3NFoutoHzaVeex2Wt97zLj3Y27XP1atoV85W0weIHmN1N\nWtb19Y5j/zl/aaiWWP8i4KIkXwU+Q7uyeAGrX+9racshRplN7g1J0roLM4+JBvX6BnnAtp1mny/P\n8gLeuMdv59PSEYwyPDtt0P/Z3hhgOg/pfn5zqgpVdVWSXWm5sZ7c/TwIODPJ02qa/Ew9q+YY6Fmf\ncR3M79hutuO6mezO6mT/A4+nLcMcl1Gf0/cBz6fNuPsKbdbbbcD/ZfbBn/UZm23QcR1AN8PzlCSn\n0l7zU5LsWFU/pf39/S9w6BS7j+PvTIuMATFp4/JT2lLEh81Q75u05V5fmOXgZK7+lLb083MzVayW\nkPMdwDuS7AT8Ny0R+yAgNpcrirOxK6vv6jfwANZMxr+C0TOalo4om23/ltNyi+w6RZ9g5hsCzNZr\nWHMQv14Bzll6Fi156FOr6seDwiRP3wDHnq3Blcb7j9g2qmwuBjP77tn9HHxhuL6qzh5Rv2/cn3FJ\nmiTLaF9at+vPEuuWMm3H6nNr/xxwRa/eJrTz+yW9Nq+hnbPvN1+dHjI4Z8xm/Lb1LM4rA4P+Xz5t\nrRl079Fzae/JtBdxu1QPnwQ+2d3w4Cha4vg/ogXzxn3Ou1+STYZmXg1maS/rfg5myQ/f+fHOwG8O\ntTeX/g3a35WWdmTQ7qa0lRgXzqGt6VzC2isZLhlVccwOoC35fXW/sFvhsVh8F9g9yV2GZomt17iu\nqirJRbTvS/ekfcf6JvBA4JxZzHh0bHcHYQ4xaSPSBbdOA/ZIstbdW7qBCbS77G1PuyvRcJ0l3ZK7\ndZLk92h3D/o57S6EU9Xbuhtg9fv/PVrgqD9guRFYsp55xvoOTu/W5En2oN0R6DO9Ot8E7pbk4b16\nd2X0nS5vZBbLAbtlsGcCe3Z3kuq3+xe0q7rDeePWSVVdXFVn9x4Xj6PdGQwCq7efN7q8aVNdRdvg\nqurbtMSpB2bN28Y/lNnd7YgkT55i097dz8GyiQtpg7TX9o/Va6efk2IwgJvNslJJ0ppO734On29e\nO7T9bNryxpd1OVcHDmTo/98uuPKftHyQ865bsnUu8GfdDKs1DI3fHp5k3xF1Nk8ynN9r0P/ZrEQY\nqRurvZcW9HlXPwfniLrb9Z93QYPBLOnBezzuc952tFlMgz4soa2AWEmXH7TL17qctc/1h7D2DLG5\n9O+M7ufwZ+9FtAuTpzMGVbViaFx3dlXN68XO7jO3iqF4QJInAQ8fudPCOJM2ieeQofJXzmbnJL/X\nBc+Hy7egLf+9hdVLeE+mBcf+YkT9TZL0L0bP6vuBFj9niEkbn9fTriKdleSDtCtIdwf2os28Op+W\n/PVPaLelfkxXtop2JfFPaFOhT5zFsZ6cZCltMLEDbc3902gJTvetqp9Ms+8TgPcl+Tht6divu313\npV1JHBgEc96T5GzaVPqTZ9G3qfwK+FKSj9Cme/8lLfdHPyHqx2h5MD6R5F20JLwvBH5Cu5tU38XA\n3kne2f37hqqaagD0RtoSgi8mOQ64nnYXqvsCB4zIK7Ex+Swtf9hZSU6gvWfPYf6WIqyrw2hB4wu6\n5bp3BV5OS448avbesM8muYo2CL6aNituN9qV8x/Tbi8+yElzEO19uSzJv9BmHPwm7Tbm9wEe3LV5\nKW3A9cokt9HujHVBVV2z/i9Xku4Q/izJ40aUn0T7f/ZzwOuT/BbwVVouy+fTcledCVBVP03y97Qx\nxueSfArYBfgzWv6k4RkdnwSO6S2Xmm+vAP4D+M/uPHoV7ZyxLy2X0TJa4u+nAacmOYkWtNuUNiPq\nT4H9WHMZ3ZOBr8wwHuu7R5Lndf++S9fuvrQZdCfRxofT+XyS5d3r+AFtzPQy2vnxPGjBvy6/2HOT\nfIs2i/3Sqrp0ln0c9k3g2CQPo/0e96XNRnv9UNDo/cAbu/PxhbRz9xNYe5nbFbTlgy9NcjMtsPHV\nqlorz1xVXZrk/cBLuotfn6NdZD2EdpHzn9bxNS24bobU6cCfJ/kVbangQ2jj1stZPOO7U2krI96e\n5H60MdVTWD1en2mm1l60/zvOoH0uVtDypT2XNhvsLb0g8L/QPl/v7SYefKlrfxfa395raEEzaN8J\nnpjkKNoY8xdV9dn1fK1aAAbEpI1MVf04yR/Q8nE9k3aV6ie0oNfVXZ1Ksh9t8PUC2h1abqHNaPk3\nZnnHPeDw7ucttLwCl9ECcv80nNx2hEtoQYWn0IJNv6bN3nlhVfUTkZ9EG9juS7uKG1afbNbFa2hJ\n0N9AC4Z8EXhFv79d8vpnAscAb6MN6o6hJZIdTpJ+DC2Q8udd299liiuCXW6N3WnLBw6lJZ6/BHh6\nl6h9o1VVlyV5FvAW2nt2He2z9BFWBzUXXFV9IslzgDfRfg/fpuU8exSzS757IG022LNoVwk3pd2m\n+4PAUf27i1XVuUkeSQuEHkK74cVPaFfL39Srd12SF9K+aJxAG2Q+hxZAkyTBi6cov7C7Sc++tP9D\nn0v7//NHtP/jjxyq/3pagOMltKDJf9P+Tz+ONqOo7yTanSr3Z5oZ7+PSnUcH47cDaUnsfwh8ni5o\nU1UrkzyBdkOiZ3ePG4DvdK/h64P2ulkvT2TEbJZpPJh23q6u3R/Qxo8vrBF3Lx/hfbT3/+WsPued\nARxZVb/o1TuQNn76B9pY6M20QMa6+C5tDPZO2mv9CfDaqhq+6+hbaasjDqAFD8+lvT/n9StV1Y1J\nDqR9do6nfR8+iPYej/JS2ljiRbTk6z+jncvfMOak9wvhpbS/lz+lvceX0L5bvITZ3T183nUXIPek\nrU75P13x52g5XS9j7b/rYR+jjeX2oP1dbUcb718C/G1VfbR3rFVdKpBX0QLpT6NdaP8u7e/mS712\n30YLlP0F7W/hKlrwXhuZzO2GIJIkbXySnAXcs6oeutB9kSRtON0Su58Cp1XVi4e2HUe7Sc3D5niX\nxAWX5HW0wNQuXV4vaWIkeRTtJkf7VdX/W+j+aONlDjFJ0h1Gkk27Lz/9sofR7nI625mRkqSNUD+H\naM9BtNQSo84Bb6GlNdhnPvs1bt3rfBVtZpbBMN2hDf9dd/nPXk1bffKlkTtJs+QMMUnSHUaSXWhT\n6f8V+D6rp7OvAn7HvF2SdMfV5cc6hLaM7zpaHqmDaPmR/uAOsMRNmjhdXrjNaDnANqEFsB8PHFtV\ni+bmTto4mUNMknRH8jNaTrM/p90I4iZaHpE3GAyTpDu8S2k5kQ6l3QXwWlri89cbDJM2WufQZkQ+\nDdiclu/tb2h55aT14gwxSZIkSZIkTRRniC2A7bffvpYuXbrQ3ZAkSfPo4osvvraqdljofmg1x2CS\nJN2xzWX8ZUBsASxdupSLLrpoobshSZLmUZLvLnQftCbHYJIk3bHNZfzlXSYlSZIkSZI0UQyISZIk\nSZIkaaIYEJMkSZIkSdJEMSAmSZIkSZKkiWJATJIkSZIkSRPFgJgkSZIkSZImigExSZIkSZIkTRQD\nYpIkSZIkSZooBsQkSZIkSZI0UQyISZIkSZIkaaIYEJMkSZIkSdJEMSAmSZIkSZKkiWJATJIkSZIk\nSRPFgJgkSZIkSZImigExSZIkSZIkTRQDYpIkSZIkSZooBsQkSZIkSZI0UQyISZIkSZIkaaIYEJMk\nSZIkSdJEMSAmSZIkSZKkiWJATJIkSZIkSRPFgJgkSZIkSZImigExSZIkSZIkTRQDYpIkSZIkSZoo\nBsQ0pZW3rhpLHUmSJEmzHzs7xpak+bfJQndAi9fmmy5h6WGfmbbOsqP33kC9kSRJkjZusxlfg2Ns\nSdoQnCEmSZIkSZKkiWJATJIkSZIkSRPFgJgkSZIkSZImigExSZKkCZNklyTvT3JJklVJzpui3kOT\nnJHkF0l+meQ/k/z+UJ0HJTknyU1JfpjkyCRLhuokyeuTfC/JzUm+mOR3RxxvbG1JkiRNx4CYJEnS\n5HkwsBfwje6xli7IdAFwHXAAsD9wOrBFr862wNlAAc8AjgReA7x5qLnDgMOBtwH7ADcAZye5x3y0\nJUmSNBPvMilJkjR5Tq+qTwEk+Tiw/Yg6x3f1ntcr+9xQnUNoAbJ9q+p64PNJtgaOSPL2qro+yea0\nINZRVXVcd8yvAMuAlwNvnIe2JEmSpuUMMUmSpAlTVbdNtz3Jg4A/BN4zQ1N7Amd1AayBk2mBrcd2\nz3cHtgZO6R3/Rtpssz3nqS1JkqRpGRCTJEnSsD/sfm7b5Rn7dZJvJfnzoXq7Alf2C6rqGuCmbtug\nzirg6qF9r+jVGXdbkiRJ0zIgJkmSpGGDfFwfBv4V2IO2XPKDSfbq1duWlmNs2Ipu26DODVW1akSd\nLZNsNg9t3S7JwUkuSnLR8uXLRzQvSZImkQExSZIkDRuMET9YVW+vqnOr6mXAucDrhurWiP0zVD5V\nneFt42yrFVSdUFW7VdVuO+yww4hdJUnSJDIgJkmSpGE/736eO1T+BeBBvecrgG1G7H83Vs/2WgFs\nlWTJUJ1tgJuq6tZ5aEuSJGlaBsQkSZI07IopygP0E/JfyVDuriQ7AXdhdT6wK4ElwC5DbQ3nDBtn\nW5IkSdMyICZJkqRhF9BmYz1xqPyJwCW952cCT0myVa/sAOBm4PxeW9cD+w8qJNkS2Kfbfz7akiRJ\nmtYmC90BSZIkbVhdEGmQHP9ewNZJ9uuef7aqbkpyJPD2JNcBXwP+BPhj4LG9po4HXgmcluRtwM7A\nEcAxVXU9QFWtTHI0cHiSFbSZXIfSLsy+Z57akiRJmpYBMUmSpMmzI3DqUNng+X2BZVX1D0nuBLyC\nFpi6Ctivqr402KGqViR5InAccDot19exXf2+o2lBq9cB2wEXAXtU1U/moy1JkqSZGBCTJEmaMFW1\njNV3Zpyu3jHAMTPUuRx4wgx1Cnhr99ggbUmSJE3HHGKSJEmSJEmaKAbEJEmSJEmSNFEMiEmSJEmS\nJGmiGBCTJEmSJEnSRDEgJkmSJEmSpIliQEySJEmSJEkTxYCYJEmSJEmSJooBMUmSJEmSJE0UA2KS\nJEmSJEmaKAbEJEmSJEmSNFEMiEmSJEmSJGmiGBCTJEmSJEnSRDEgJkmSJEmSpIliQEySJEmSJEkT\nxYCYJEmSJEmSJooBMUmSJEmSJE0UA2KSJEmSJEmaKAbEJEmSJEmSNFEMiEmSJEmSJGmiGBCTJEmS\nJEnSRDEgJkmSJEmSpIliQEySJEmSJEkTxYCYJEmSJEmSJooBMUmSJEmSJE0UA2KSJEmSJEmaKAbE\nJEmSJEmSNFEMiEmSJEmSJGmiLFhALMl+SS5I8rMkK5NcleSNSTbr1UmS1yf5XpKbk3wxye+OaOtB\nSc5JclOSHyY5MsmSoTobvC1JkiRJkiQtPgs5Q2w74FzgRcCewD8DbwCO6dU5DDgceBuwD3ADcHaS\newwqJNkWOBso4BnAkcBrgDcPHW+DtiVJkiRJkqTFaZOFOnBVvX+o6NwkWwMvS/IK4M60wNNRVXUc\nQJKvAMuAlwNv7PY7BNgC2Leqrgc+37VzRJK3V9X1STZfgLYkSZIkSZK0CC22HGI/AwZLJncHtgZO\nGWysqhuB02kzygb2BM7qAlj1g+7AAAAgAElEQVQDJ9MCW49dwLYkSZIkSZK0CC14QCzJkiRbJnkM\n8ErgfVVVwK7AKuDqoV2u6LYN7Apc2a9QVdcAN/XqLURbkiRJkiRJWoQWPCAG3Ng9vgScD7y2K98W\nuKGqVg3VXwFs2Uu+vy1w3Yh2V3TbFqqtNSQ5OMlFSS5avnz5qCqSJEmSJEnaABZDQGx34I9oyeuf\nARzX21Yj6mfEtqnqzabOfLa1eoeqE6pqt6rabYcddhhVRZIkSZIkSRvAgiXVH6iq/+r++eUk1wIf\nSvL3tBlXWyVZMjQbaxvgpqq6tXu+oisbdjdWz/ZaiLYkSZIkSZK0CC2GGWJ9g+DYfWm5vJYAuwzV\nGc7zdSVDubuS7ATcpVdvIdqSJEmSJEnSIrTYAmKP7n5+B7gAuB7Yf7AxyZbAPsCZvX3OBJ6SZKte\n2QHAzbScZCxQW5IkSZIkSVqEFmzJZJLPAWcDl9Hu2vhoWh6xf6uqb3V1jgYOT7KCNvvqUFoQ7z29\npo6n3Z3ytCRvA3YGjgCOqarrAapq5QK0JUmSJEmSpEVoIXOIfQ14AbAU+DXwbeB1tKDUwNG0QNPr\ngO2Ai4A9quongwpVtSLJE2nJ+E+n5fo6lhbIYqHakiRJkiRJ0uK0YAGxqjocOHyGOgW8tXtMV+9y\n4AmLrS1JkiRJkiQtPosth5gkSZIkSZI0rwyISZIkSZIkaaIYEJMkSZIkSdJEMSAmSZIkSZKkiWJA\nTJIkSZIkSRPFgJgkSZIkSZImigExSZIkSZIkTRQDYpIkSZIkSZooBsQkSZIkSZI0UQyISZIkSZIk\naaIYEJMkSZowSXZJ8v4klyRZleS8Ger/Q5JK8s4R2x6U5JwkNyX5YZIjkywZqpMkr0/yvSQ3J/li\nkt+dz7YkSZKmY0BMkiRp8jwY2Av4RveYUpIHAS8Erh+xbVvgbKCAZwBHAq8B3jxU9TDgcOBtwD7A\nDcDZSe4xH21JkiTNxICYJEnS5Dm9qnaqqv2By2ao+27gXcCKEdsOAbYA9q2qz1fV8bQA1qFJtgZI\nsjktiHVUVR1XVWcD+9MCXy+fp7YkSZKmZUBMkiRpwlTVbbOpl2Q/4IHA0VNU2RM4q6r6s8dOpgW2\nHts93x3YGjild/wbgdO7/eejLUmSpGkZEJMkSdJakmwB/D1wWBd0GmVX4Mp+QVVdA9zUbRvUWQVc\nPbTvFb06425LkiRpWgbEJEmSNMrrgB8BJ01TZ1vguhHlK7ptgzo3VNWqEXW2TLLZPLR1uyQHJ7ko\nyUXLly+f5qVIkqRJYkBMkiRJa0hyX+CvgFdVVc1QfdT2DJVPVWd42zjbagVVJ1TVblW12w477DBi\nV0mSNIkMiEmSJGnY0cCZwJVJtkmyDW3ceOfu+SAAtQLYZsT+d2P1bK8VwFZJlgzV2Qa4qapunYe2\nJEmSpmVATJIkScN+G9iXFoAaPHai3clxBXCvrt6VDOXuSrITcBdW5wO7ElgC7DJ0jOGcYeNsS5Ik\naVoGxCRJkjTsRcDjhx4/od3d8fHAIBnXmcBTkmzV2/cA4Gbg/O75BcD1wP6DCkm2BPbp9h8YZ1uS\nJEnT2mShOyBJkqQNqwsi7dU9vRewdZL9uuefraqLRuyzEvheVZ3XKz4eeCVwWpK3ATsDRwDHVNX1\nAFW1MsnRwOFJVtBmch1KuzD7nnlqS5IkaVoGxCRJkibPjsCpQ2WD5/cFls2mkapakeSJwHHA6bRc\nX8fSAll9R9OCVq8DtgMuAvaoqp/MR1uSJEkzMSAmSZI0YapqGavvzDjbfZZOUX458IQZ9i3grd1j\nunpja0uSJGk65hCTJEmSJEnSRDEgJkmSJEmSpIliQEySJEmSJEkTxYCYJEmSJEmSJooBMUmSJEmS\nJE0UA2KSJEmSJEmaKAbEJEmSJEmSNFEMiEmSJEmSJGmiGBCTJEmSJEnSRDEgJkmSJEmSpIliQEyS\nJEmSJEkTxYCYJEmSJEmSJooBMUmSJEmSJE0UA2KSJEmSJEmaKAbEJEmSJEmSNFEMiEmSJEmSJGmi\nGBCTJEmSJEnSRDEgJkmSJEmSpIliQEySJEmSJEkTxYCYJEmSJEmSJooBMUmSJEmSJE0UA2KSJEmS\nJEmaKAbEJEmSJEmSNFEMiEmSJEmSJGmiGBCTJEmSJEnSRDEgJkmSJEmSpIliQEySJEmSJEkTxYCY\nJEmSJEmSJooBMUmSJEmSJE0UA2KSJEmSJEmaKAbEJEmSJEmSNFEMiEmSJEmSJGmiGBCTJEmSJEnS\nRDEgJkmSJEma1spbV42ljiQtFpssdAckSZIkSYvb5psuYelhn5m2zrKj995AvZGk9ecMMUmSJEmS\nJE0UA2KSJEmSJEmaKAbEJEmSJEmSNFEMiEmSJEmSJGmiGBCTJEmSJEnSRDEgJkmSJEmSpIliQEyS\nJEmS7oBW3rpqLHUk6Y5ok4U6cJL9gecDvw/cDbgKeGdVfaxX5zzgsSN236KqVvbq3Qs4DtgDWAmc\nDPx1Vd00dMwXA38N7ARc1tU5Z6jO2NqSJEmSpIWy+aZLWHrYZ6ats+zovTdQbyRpcVmwgBhwKPAd\n4NXAtcBewEeTbF9V7+nVOxd4/dC+vxr8I8kmwFnALcABwDbAMd3P5/XqPRs4HjgC+DJwEHBGkkdU\n1aXjbkuSJEmSJEmL00IGxPapqmt7z7+Q5J60QFk/IPbzqrpwmnb2Bx4I7FJV3wFIcitwcpI3V9XV\nXb03Ax+qqrd0dc4HHg4cxupg1zjbkiRJkiRJ0iK0YDnEhoJhA/8N7DjHpvYEvjYIYHU+SZvl9VSA\nJDsDDwBO6R3/NuDUbv/5aEuSJEmSJEmL0GJLqr87cPlQ2ZOT3NQ9zkrysKHtuwJX9guq6hbgW902\nej/XqAdcAdw9yQ7z0JYkSZIkSZIWoUUTEEvyROAZwHt7xecDfwk8BTgYuDfwpSRLe3W2Ba4b0eSK\nbhu9n8P1VgxtH2dba0hycJKLkly0fPnyUVUkSZIkSZK0ASyKgFgX4Poo8KmqOnFQXlVvqqp/qaov\nVdVJwOOBAl411ESNanZE+fDzjCgfZ1urK1edUFW7VdVuO+zgJDJJkiRJkqSFsuABsSR3B84ErmGG\nhPRV9WPgP4Df6xWvoN0Fctg2rJ7FtaJXNlyHoXrjakuSJEmSJEmL0IIGxJJsCZwBbAbsXVU3znLX\n/iysK1md12vQ7mbAzqzO8zX4uUa97vnPq2p5r9642pIkSZIkSdIitGABsSSb0O7MeH9gz6r66Sz2\n+Q3g0cDFveIzgUckuU+v7OnAnYHPAVTVt4FvAPv32rpT9/zMeWpLkiRpUUqyS5L3J7kkyaok5w1t\n/80k7+i235Dke0k+lOSeI9q6V5JPdPWuTXJcd9FzuN6Lk1ydZGWSi7v8sfPWliRJ0nQ2WcBj/yOw\nFy1p/t2TPLK37b+B3waOogXNvktLqP864DbgH3p1Pw68ATgtyeHA3YBjgY9W1dW9ekcAJyVZRlt2\neSAtGPfceWpLkiRpsXowbRx2IW2m/rDfB54FfBD4KvAbtPHPBUkeUlU3wO0XOM8CbgEOoKWQOKb7\neXsqjCTPBo7v2vgycBBwRpJHVNWl425LkiRpJgsZEHty9/NdI7bdF/gZLVH9UcB2wC+B84BnVtU1\ng4pVdWuSpwLHAacAvwJOBl7bb7CqPpbkrsDfAIcDlwFP6w+cxtmWJEnSInZ6VX0KIMnHge2Htn8Z\n2LWqfj0oSPJfwFXAnwAf6or3Bx4I7FJV3+nq3QqcnOTNvQuKbwY+VFVv6eqcDzwcOIzVwa5xtiVJ\nkjStBQuIVdXSWVTba5ZtfR945izqfQD4wIZqS5IkaTGqqttm2L7WTYKq6htJbgJ27BXvCXxtEMDq\nfJI2y+upwNVJdgYeQFsVcPvxk5zaLxtzW5IkSdNa8LtMSpIkafFL8jBgS+DyXvGurL7hEABVdQvw\nLVbfgGjwc416wBW0tBk7zENbkiRJ0zIgJkmSpGl1NxB6F3A18O+9TdsCa80mA1Z02+j9HK63Ymj7\nONvq9/3gJBcluWj5cm8GLkmSGgNikiRJmslRwKOA51fVrUPbakT9jCgffp4R5eNsqxVUnVBVu1XV\nbjvs4AQySZLUGBCTJEnSlJK8lHaDoQOr6qtDm1fQ7gI5bBtWz+Ja0SsbrsNQvXG1JUmSNC0DYpIk\nSRopyZ8A7wH+uqr+bUSVK1md12uwz2bAzqzO8zX4uUa97vnPq2p5r9642pIkSZqWAbEJtfLWVQvd\nBUmStIgleRzwr8BxVfXOKaqdCTwiyX16ZU8H7gx8DqCqvg18A9i/1/aduudnzlNb0kZtNmN1x/OS\ntH42WegOaGFsvukSlh72mWnrLDt67w3UG0mStCEl2RLYq3t6L2DrJPt1zz8L3Af4JG1G1r8leWRv\n9+VV9a3u3x8H3gCcluRw4G7AscBHq+rq3j5HACclWQb8B3AgcH/gub0642xL2qg5Vpek+WdATJIk\nafLsCJw6VDZ4fl/gD2kBqd+hBZ36PgS8AKCqbk3yVOA44BTgV8DJtJxjt6uqjyW5K/A3wOHAZcDT\nqurSXp2xtSVJkjQTA2KSJEkTpqqWsfrOjKOc2D1m09b3gWfOot4HgA9sqLYkSZKmYw4xSZIkSZIk\nTRQDYpIkSZKk9ebNACRtTFwyKUmSJElab94MQNLGxBlikiRJkiRJmigGxCRJkiRJkjRRDIhJkiRJ\nkiRpohgQkyRJkiRJ0kQxICZJkiRJkqSJYkBMkiRJkiRJE8WAmCRJkiRJkiaKATFJkiRJkiRNFANi\nkiRJkiRJmigGxCRJkiRpI7Py1lUL3QVJ2qhtstAdkCRJkiTNzeabLmHpYZ+Zts6yo/feQL2RpI2P\nM8QkSZIkSZI0UQyISZIkSZIkaaIYEJMkSZIkSdJEMSAmSZIkSZKkiWJATJIkSZIkSRPFgJgkSZIk\nSZImigExSZIkSZIkTRQDYpIkSZIkSZooBsS0XlbeumosdSRJkiRJkjaUTRa6A9q4bb7pEpYe9plp\n6yw7eu8N1BtJkiRJkqSZOUNMkiRJkiRJE8WAmCRJkiRJkiaKATFJkiRJkiRNFANikiRJkiRJmigG\nxCRJkiRJkjRRDIhJkiRJkiRpohgQkyRJkiRJ0kQxICZJkiRJkqSJYkBMkiRJkiRJE8WAmCRJkiRJ\nkiaKATFJkiRJkiRNFANikiRJkiRJmigGxCRJkiRJkjRRDIhJkiRJkiRpohgQkyRJkiRJ0kQxICZJ\nkiRJkqSJYkBMkiRJkiRJE8WAmCRJkiRJkiaKATFJkiRJkiRNFANikiRJkiRJmigGxCRJkiRJkjRR\nDIhJkiRJkiRpohgQkyRJkiRJ0kQxICZJkiRJkqSJYkBMkiRJkiRJE8WAmCRJkiRJkiaKATFJkqQJ\nk2SXJO9PckmSVUnOG1EnSV6f5HtJbk7yxSS/O6Leg5Kck+SmJD9McmSSJQvdliRJ0nQMiEmSJE2e\nBwN7Ad/oHqMcBhwOvA3YB7gBODvJPQYVkmwLnA0U8AzgSOA1wJsXsi1JkqSZGBCTJEmaPKdX1U5V\ntT9w2fDGJJvTAk9HVdVxVXU2sD8tWPXyXtVDgC2Afavq81V1PC2AdWiSrRewLUmSpGkZEJMkSZow\nVXXbDFV2B7YGTuntcyNwOrBnr96ewFlVdX2v7GRaYOuxC9iWJEnStAyISZIkadiuwCrg6qHyK7pt\n/XpX9itU1TXATb16C9GWpFlaeeuqhe6CJC2ITRa6A5IkSVp0tgVuqKrhb8orgC2TbFZVt3T1rhux\n/4pu20K1dbskBwMHA9z73vce/WqlCbb5pktYethnZqy37Oi9N0BvJGnDWbAZYkn2T/LpJD9IckOS\ni5M8Z0S9Fye5OsnKrs4TR9S5V5JPdO1cm+S4JFsudFuSJEkbsRpRlhHbpqo3mzrz2VYrqDqhqnar\nqt122GGHEbtKkqRJtJBLJg+l3RXo1cDTgXOBjyZ5xaBCkmcDxwMfpuWFuAw4I8lDenU2Ac4C7gMc\nAPwlLbnqCf2Dbei2JEmSNmIrgK2SLBkq3wa4qapu7dXbZsT+d2P1bK+FaEuSJGlaC7lkcp+qurb3\n/AtJ7kkLlL2nK3sz8KGqegtAkvOBh9PuLvS8rs7+wAOBXarqO129W4GTk7y5qq5eoLYkSZI2VlcC\nS4BdgKt65cN5vq5kKHdXkp2Au/TqLURbkiRJ01qwGWJDwbCB/wZ2BEiyM/AA1ryL0G3Aqax9R6Kv\nDQJYnU8CtwBPXcC2JEmSNlYXANfTLhYC0KWQ2Ac4s1fvTOApSbbqlR0A3Aycv4BtSZIkTWuxJdXf\nHbi8+/fgCuHw1b4rgLsn2aGqlnf1Lu9XqKpbknyLNe9ItKHbkiRJWpS6INJe3dN7AVsn2a97/tmq\nuinJ0cDhSVbQxj2H0i6mvqfX1PHAK4HTkrwN2Bk4Ajimqq4HqKqVC9CWJEnStBZNQKxLSv8M4IVd\n0eBuQsN3G1rR276c2d+RaEO3tQbvcCRJkhaRHWmz2/sGz+8LLAOOpgWaXgdsB1wE7FFVPxnsUFUr\nujHcccDptPHRsbRAVt8GbUuSJGkmcwqIJTmSljvrW+PsRJKlwEeBT1XViUObh+8WtK53JFqItlZX\nrjqBLjn/brvtNrKOJEnSsPkYf1XVMlaPXaaqU8Bbu8d09S4HnrDY2pIkSZrOXHOIvR74RpILkrwk\nyag7Ac1JkrvTcj5cw5oJ6QczroaPMXjev9vQqH5sM1RnQ7clSZI0DmMff0mSJE26uQbEfgv4G2BL\n4H3Aj5KckuRpI25/PaMuf8UZwGbA3lV1Y2/zIEfXrkO77Qr8vJena9QdiTaj5Z3o35FoQ7clSZI0\nDmMdf0mSJGmOAbGq+nFVvbOqfhd4OPBe4NHAp4EfJjk2ye/Npq0km9ByVdwf2LOqfjp0rG8D32DN\nuwjdqXs+fEeiRyS5T6/s6cCdgc8tYFuSJEnrbZzjL0mSJDVznSF2u6q6pKr+CtgJeCpwIe3OQF9L\ncmmS18wwpf8faXc3egvt7oyP7D3u3NU5AjgoyRuTPB74Z1oA7eheOx+nzdo6LcleSZ5DS8b60aq6\nuldvQ7clSZI0VmMYf0mSJIn1CIgBJAmwB/BnwJNoyVkvBK4F3gF8O8mTp9h9UP4u4CtDj98EqKqP\nAYcAL6DN0HoY8LSqunTQSFXdShsQfg84hRbA+n90d3Ts1dugbUmSJM2H9Rx/SZIkiTneZXIgye8A\nzweeC9wD+AEtsPUvg5lUSXYFTgLezdr5tqiqpbM5VlV9APjADHW+DzxzsbUlSZI0LuMYf0mSJKmZ\nU0AsyV/RBmIPAX4FfBI4Efh8dwvs21XVlUneTVtOKEmSpHXg+EuSJGn85jpD7O3AV4GXAidX1S9m\nqH8xLUeYJEmS1o3jL0mSpDGba0DsgVV11WwrV9VlwGVzPIYkSZJWc/wlSZI0ZnNNqv/9JPeeamOS\neyfZcj37JEmSpNUcf0mSJI3ZXANixwKnT7P9U8A71707kiRJGuL4S5IkaczmGhB7EvCJabZ/AvA2\n35IkSePj+EuSJGnM5hoQuyfw/Wm2/7CrI0mSpPFw/CVJkjRmcw2I/Ry4/zTbfxv45bp3R5IkSUMc\nf0mSJI3ZXANi5wAvSbLWoCzJbwMHA2ePo2OSJEkCHH9JkiSN3SZzrP8mYG/gf5J8GLgUKOBhwJ8B\nK7s6kiRJGg/HX5IkSWM2p4BYVX07yaOB99KuRqa3+VzgFVX1zTH2T5IkaaI5/pIkSRq/uc4Qo6qu\nAJ6QZDvgfrRB2Ter6mfj7pwkSZIcf0mSJI3bnANiA90AzEGYJEnSBuL4S5IkaTzWKSCW5IHAzsDd\nWXPaPgBV9eH17JckSZJ6HH9JkiSNz5wCYkmWAicBj2LEQKxTgAMySZKkMXD8JUmSNH5znSF2AvBw\n4FDgfGDF2HskSZKkPsdfkiRJYzbXgNhjgLdX1bvmozOSJElai+MvSZKkMbvTHOv/Arh2PjoiSZKk\nkRx/SZIkjdlcA2InAn+aZKr8FZIkSRqvE3H8JUmSNFZzXTJ5FvAk4Pwk7weuAVYNV6qqC8bQN0mS\nJDn+kiRJGru5BsS+0Pv3o0dsD+0uR0vWuUeSJEnqc/wlSZI0ZnMNiB00L72QJEnSVBx/SbrDWHnr\nKjbfdPr4/WzqSNL6mlNArKo+NF8dkSRJ0tocf0m6I9l80yUsPewz09ZZdvTeG6g3kibZXJPq3y7J\n/ZM8OsndxtkhSZIkjeb4S5IkaTzmHBBLckCS7wJXAl8Efr8r3z7J1Un2H3MfJUmSJprjL0mSpPGa\nU0AsyTOAj9HubnQ4LYkrAFV1LXAF8PxxdlCSJGmSOf6SJEkav7nOEHsj8MWq+iPg/SO2fxX4nfXu\nlSRJkgYcf0mSJI3ZXANiDwZOmWb7j4HfWPfuSJIkaYjjL0mSpDGba0BsJbD5NNuXAtetc28kSZI0\nzPGXJEnSmM01IPZl4DmjNnR3O3oh8IX17ZQkSZJu5/hLkiRpzOYaEDsCeHCSc4F9u7Ldkrwc+B9g\na+At4+ueJEnSxDsCx1//n717j5Osqu+9//k6DA6jXEbAGD0oIioi8ZKMMTEXozxegCB5TAiao4ma\nIyHnGEzwkAwEFDDEIYmgARMkJtHEKI8XjCISAhoRxCQSL9HAIF5GRNQMOjjCMDAOv+ePvZspiuru\n6u7qru7en/frtV81tfZvr1p7TXf1ql+tvbYkSdJIzSghVlWfAZ4HPIydi7quB/4cuBt4XlVdP9IW\nSpIkdZjjL0mSpNHbZaYHVNVVwBOSPBl4HE1S7cvAZ6qqRtw+SZKkznP8JUmSNFozTohNqKrPA58f\nYVskSZI0BcdfkiRJozGjhFiSnx8mrqo+MbvmSJIkqZfjL0mSpNGb6QyxjwPDTMtfMfOmSJIkaYCP\n4/hLkiRppGaaEHvWgLIVwKOB44B7gJPm2ihJkiTdy/GXJEnSiM0oIVZVV062L8nbgWuAnwU+Nrdm\nSZIkCRx/SZIkzYcHjKqiqtoBvAs4dlR1SpIkaXKOvyRJkmZnZAmx1m7A3iOuU5IkSZNz/CVJkjRD\nM11DbKAke9Csb/F/gU+Pok5JkiRNzvGXJEnS7M0oIZbkHia/y1GAjcD/mWObJEmS1HL8JUmSNHoz\nnSF2BvcfkBWwGfgy8M/tWhbSvbZt38GqlVPfCX6YGEmSOsrxlyRJ0ojN9C6Tp81TO7SMrVq5gv3X\nXTJlzMb1RyxQayRJWlocf0mSJI3eqBfVlyRJkiRJkha1ma4h9tpZvEZV1etncZwkSVLnjXP8leRF\nwO8DjwO+D3wUWFdVt/TEBDgJ+G1gH5oF/o+vqs/11XUwcC7w08BtwNuA03sv9xxlXZIkSVOZ6Rpi\np7FzDYv07Zuq3ISYJEnS7JzGGMZfSV4AvBt4C3Ai8KPAHwEfTrK2qu5pQ9cBp7YxG4ATgCuSHFJV\n327rWgNcAVwHHAU8BngjzdUKp/S87CjrkiRJmtRML5l8OPB5msHRTwJ7ttvTgQvbfQ+vqgf0bK6U\nLkmSNHvjGn/9GvCZqnpVVX20qt4JHA88FXg8QJJVNEmsN1TVeVV1BXA0TULuVT11HQfsBrywqi6v\nqvOB04ETkuwx6rokSZKmM9OE2LnA16rqJVV1bVX9oN0+XVX/k+a23+eOvJWSJEndNa7x10qayyR7\n3dY+TsxIewawB/CeiYCqugO4GDis57jDgMuqaktP2YU0ia1nzkNdkiRJU5ppQuy5NFPUJ3M58JzZ\nN0eSJEl9xjX++hvg55L8epI9kjyO5pLJf6mq69qYg4AdwI19x17f7qMnbkNvQFXdBGztiRtlXZIk\nSVOaaULshzTT5Cfz48A9U+yXJEnSzIxl/FVVlwAvAy6gmSl2A7ACeGFP2Brg9gGL2W8GVifZtSfu\nNu5vc7tv1HXdK8mxSa5Ncu2mTZsGHCZJkrpopgmx9wO/meS1SfaaKEyyV5LXAa8A3jfKBkqSJHXc\nWMZfSZ4FnA+8GXgW8CLgIcAHkvSuUVaDDh+wb7K4YWJmU1cTWHVBVa2tqrX77rvvgMMkSVIXzfQu\nk6+huZPPacDrkmyiGXg8lGYQcmUbI0mSpNEY1/jrjcCHquoPJgqSfI7mcsWjgItoZmXtnmRF38yu\nvYCtVbW9fb65Leu3Jztne42yLkmSpCnNKCFWVT8ADk1yJPCLwCNpBmIbgUuq6uKRt1CSJKnDxjj+\nOojmzpa9bbkhyZ00CTpokmMrgANpLqnsPbZ3na8N9K3vlWQ/4EE9caOsS5IkaUoznSEGQDvwMvkl\nSZK0QMYw/vo6zfpk90ryBJq7OW5si64BtgBH0yy4T5LVwJE0a49NuBQ4McnubYIP4BjgTpoZbqOu\nS5IkaUqzSoglOZBmLYmHAv9QVRvbhU4fBny7qu4eYRslSZI6bwzjr/OBc5LcQpOE+hHgtTTJsI8A\nVNW2JOuBU5NsppmhdQLNOrXn9tV1PHBRkrOAA2guAT27qraMui5JkqTpzCghliTAW4BjaQYnBXyK\nZmC0K/AF4AyaNSckSZI0R2Mcf/05cDfw28BxNOtzXQ2cVFV39MStb9t1ErA3cC3wnKr6zkRAVW1O\ncihwHs0st9uAc2gSWcxTXZIkSZOa6QyxP6AZEL0B+GfgXyZ2VNXtSS4CfgkTYpIkSaMylvFXVRXw\nl+02XdyZ7TZV3HXAsxeqLkmSpKk8YIbxvwn8XVX9IfDFAfu/CDx2zq2SJEnSBMdfkiRJIzbThNgj\ngU9Osf8HDL4NtiRJkmbH8ZckSdKIzTQhtgl4+BT7nwTcPPvmSJIkqY/jL0mSpBGbaULsw8CxSR7a\nvyPJWpop/R8cRcMkSZIEOP6SJEkauZkmxF5Lc7ehL9Is3FrA/0ryPuAa4BvAH420hZIkSd3m+EuS\nJGnEZpQQq6r/BtYCHwCOBAK8CPh/gHcAz6iqzaNupCRJUlc5/pIkSRq9XWZ6QFV9F/gt4LeS7EuT\nVNtUVfeMunGSJEly/DjBNlwAACAASURBVCVJkjRqQ88QS7I6yXeTnDhRVlWbquo7sx2MJTkwyVuT\nfD7JjiQfHxCzMUn1bd8eEHdwko8m2ZrkliRnJFnRF5MkJyf5RpI7k3wiyVPmsy5JkqTZmo/xlyRJ\nkmYwQ6yqtia5B7h9hK//ROBw4F+BXaeIexdwbs/zu3t3JlkDXAFcBxwFPIZmjY0HAKf0hK4DTgVO\nBDYAJwBXJDmkqr496rokSZLmYp7GX5IkSZ0300X1L6ZZu2JULq6q/arqaOC/poj7VlX9a8/2mb79\nxwG7AS+sqsur6nzgdOCEJHsAJFlFk8R6Q1WdV1VXAEfTLEz7qnmqS5Ikaa5GPf6SJEnqvJkmxM4G\n9ktyUZLnJnlMkof3b8NWNsKp/ocBl1XVlp6yC2kSW89snz8D2AN4T8/r30EzyDxsnuqSJEmaq5GO\nvyRJkjTzRfX/s318Is3lhJNZMcW+2XhFkuOBO4HLgddU1dd79h8EfKz3gKq6KcnWdt/F7eMO4Ma+\nuq8HjpmnuiRJkuZqXOMvSZKkZWvKhFiSFwDXVtUtbdHp89+k+/kgzRpjNwNPAF4HXJXkx6rq+23M\nGuC2AcdubvdNxNxeVTsGxKxOsmtV3T3iuu6V5FjgWIBHPvKRU52vJEnqsEUy/pIkSVrWppsh9gHg\npTSL2gP8BvC7VfWheW1Vj6p6dc/Tq5JcA3wOeDnwpt7QAYenr3yymP59o6yrKai6ALgAYO3atYOO\nlSRJgkUw/pIkSVrupltD7HZg957n+wMPnrfWDKGqvgjcAPx4T/FmYK8B4Xuyc7bXZmD3JP2XE+wF\nbK2q7fNQlyRJ0kwtuvGXJEnScjPdDLHPAScm2RWYuDzx55JMeVxV/d0oGjeN3llWG2jW9bpXkv2A\nB7X7JmJWAAfSJNQmHNQTM+q6JEmSZmoxj78kSZKWhekSYscD7wfe3D4v4LfabTIFzNuALMkhwOOB\nt/YUX0ozcNy9qn7Qlh1Dswj/le3za4AtwNHAH7V1raa5jfkF81SXJEnSTC268ZckSdJyM903jZ9P\n8jjg0cCPAFcDZ9Dc6XHO2iTS4e3TRwB7JPmV9vlHgGcBLwE+DNxCMwPrFOAm4O09VZ1PM3i8KMlZ\nwAHAacDZVbWlPZdtSdYDpybZTDOT6wSay0bPnae6JEmSZmS+x1+SJEmafoYYVXUP8BXgK0neAVxa\nVf82otd/KPDevrKJ548GvtHGvIlmfa7vAv8EnDyRnGrbuDnJocB5wMU0a32dQ5PI6rWeJml1ErA3\ncC3wnKr6znzUJUmSNBvzPP6SJEnqvGkTYr2q6uWjfPGq2sjOOzNO5tAh67oOePY0MQWc2W4LUpck\nSdJcjHr8JUmSpOnvMilJkiRJkiQtKybEJEmSJEmS1CkmxCRJkiRJktQpJsQkSZIkSZLUKSbEJEmS\nJEmS1CkmxCRJkiRJktQpJsQkSZIkSZLUKSbEJEmSJEmS1CkmxCRJkiRJktQpJsQkSZIkSZLUKSbE\nJEmSJEmS1CkmxCRJkiRJktQpJsQkSZIkSZLUKSbEJEmSJEmS1CkmxCRJkiRJktQpJsQkSZIkSZLU\nKSbEJEmSJEmS1CkmxCRJkiRJktQpJsQkSZIkSZLUKSbEJEmSJEmS1CkmxCRJkiRJktQpJsQkSZIk\nSZLUKSbEJEmSJEmS1CkmxCRJkiRJktQpJsQkSZIkSZLUKSbEJEmSJEmS1CkmxCRJkiRJktQpJsQk\nSZIkSYvGtu07RhIjSVPZZdwNkCRJkiRpwqqVK9h/3SVTxmxcf8QCtUbScuUMMUmSJEmSJHWKCTFJ\nkiQNlGSXJOuS3JjkriQ3JzmnLyZJTk7yjSR3JvlEkqcMqOvgJB9NsjXJLUnOSLJivuqSFpqX8EnS\n0uIlk5IkSZrM3wKHAqcDG4D9gIP7YtYBpwIntjEnAFckOaSqvg2QZA1wBXAdcBTwGOCNNF/OnjJP\ndUkLysv8JGlpMSEmSZKk+0nyfOBFwJOr6rpJYlbRJLHeUFXntWWfAjYCr2Jnguo4YDfghVW1Bbg8\nyR7AaUn+pKq2jLKuUfaDJElanrxkUpIkSYO8AvjYZMmw1jOAPYD3TBRU1R3AxcBhPXGHAZf1Jasu\npElsPXMe6pIkSZqSCTFJkiQN8nTgS0nOS7KlXa/roiQP74k5CNgB3Nh37PXtvt64Db0BVXUTsLUn\nbpR1SZIkTcmEmCRJkgZ5GPAy4Ck0l06+HPgJ4ANJ0sasAW6vqv7VxDcDq5Ps2hN324DX2NzuG3Vd\n90pybJJrk1y7adOmgScqSZK6xzXEJEmSNEja7aiq+i5Akm8BVwLPBj7axtUkx/bvmyxumJjZ1NUE\nVl0AXACwdu3aQcdJkqQOcoaYJEmSBtkMfGEiGda6GribnXea3AzsnmRF37F7AVurantP3F4DXmNP\nds72GmVdkiRJUzIhJkmSpEGun6Q8wD3tvzcAK4AD+2L61/naQN/6Xkn2Ax7UEzfKuiRJkqZkQkyS\nJEmDfBh4UpJ9esp+HlgJfL59fg2wBTh6IiDJauBI4NKe4y4Fnpdk956yY4A7aS7BHHVdkgTAtu39\nyxLOLkbS8uMaYpIkSRrkAuB44OIkfwzsDpwFXFFVVwNU1bYk64FTk2ymmaF1As2Xruf21HV+W9dF\nSc4CDgBOA86uqi2jrkuSJqxauYL9110yZczG9UcsUGskLSYmxCRJknQ/VbUlybOBPwcupFk77IPA\n7/WFrqdJWp0E7A1cCzynqr7TU9fmJIcC5wEX06z1dQ5NImu+6pIkSZqUCTFJkiQNVFVfBg6fJqaA\nM9ttqrjraO5OuSB1SZIkTcU1xCRJkiRJktQpJsQkSZIkSZLUKSbEJEmSJEmS1CkmxCRJkiRJktQp\nJsQkSZIkSZLUKSbEJEmSJEmS1CkmxCRJkiRJktQpJsS0KGzbvmOkcZIkSZIkSZPZZdwNkABWrVzB\n/usumTZu4/ojFqA1kiRJkiRpOXOGmCRJkiRJkjrFhJgkSZIkSZI6xYSYJEmSJEmSOsWEmCRJkiRJ\nkjrFhJgkSZIkSZI6xYSYJEmSJEmSOsWEmCRJkiRJkjrFhJgkSZIkSZI6xYSYJEmSJE1h2/Yd426C\nJGnEdhl3AyRJkiRpMVu1cgX7r7tkypiN649YoNZIkkbBGWKSJEmSJEnqFBNikiRJkiRJ6hQTYpIk\nSZKkJcV13STN1VjXEEtyIHAi8FPAIcBVVfULfTEBTgJ+G9gH+DRwfFV9ri/uYOBc4KeB24C3AadX\n1Y5x1iVJkiRJGq1h1nUD13aTNLlxzxB7InA48KV2G2QdcCpwFnAkcDtwRZKHTQQkWQNcARRwFHAG\n8Brg9HHWJUmSJEmSpMVn3Amxi6tqv6o6Gviv/p1JVtEknt5QVedV1RXA0TTJqlf1hB4H7Aa8sKou\nr6rzaRJYJyTZY4x1SZIkSZIkaZEZa0Ksqu6ZJuQZwB7Ae3qOuQO4GDisJ+4w4LKq2tJTdiFNYuuZ\nY6xLkiRJkiRJi8y4Z4hN5yBgB3BjX/n17b7euA29AVV1E7C1J24cdUmSJEmSJGmRWewJsTXA7b2L\n2bc2A6uT7NoTd9uA4ze3+8ZV172SHJvk2iTXbtq0aUD1kiRJkiRJWgiLPSEGzbpc/TJg32Rxw8TM\nZ11NQdUFVbW2qtbuu+++Aw6VJEmSJEnSQljsCbHNwO5JVvSV7wVsrartPXF7DTh+T3bO9hpHXZIk\nSZIkSVpkFntCbAOwAjiwr7x/na8N9K3dlWQ/4EE9ceOoS5IkSZIkSYvMYk+IXQNsAY6eKEiyGjgS\nuLQn7lLgeUl27yk7BrgTuHKMdUmSJEmSJGmR2WWcL94mkQ5vnz4C2CPJr7TPP1JVW5OsB05Nsplm\n9tUJNIm8c3uqOh84HrgoyVnAAcBpwNlVtQWgqraNoS5JkiRJkiQtMmNNiAEPBd7bVzbx/NHARmA9\nTaLpJGBv4FrgOVX1nYkDqmpzkkOB84CLadb6OocmkdVrQeuSJEmSJEnS4jPWhFhVbWTnnRkniyng\nzHabKu464NmLrS5JkiRJkiQtLot9DTFJkiRJkiRppEyISZIkSZIkqVNMiEmSJEmSJKlTTIhJkiRJ\nkiSpU0yISZIkSZIkqVNMiEmSJEmSJKlTTIhJkiRJkiSpU0yISZIkSZIkqVNMiEmSJEmSJKlTTIhJ\nkiRJkiSpU0yISZIkSZIkqVNMiEmSJEmSJKlTTIhJkiRJkiSpU0yISZIkSZIkqVNMiEmSJEmSJKlT\nTIhJkiRJkiSpU0yISZIkaUpJHpHk9iSV5ME95UlycpJvJLkzySeSPGXA8Qcn+WiSrUluSXJGkhV9\nMSOrS5IkaTomxCRJkjSdPwVuH1C+DjgVOAs4so25IsnDJgKSrAGuAAo4CjgDeA1w+jzWJUmSNCUT\nYpIkSZpUkp8Dng/8WV/5Kpok1huq6ryqugI4miZZ9aqe0OOA3YAXVtXlVXU+TQLrhCR7jLouSZKk\nYZgQkyRJ0kDtpYjn0szEurVv9zOAPYD3TBRU1R3AxcBhPXGHAZdV1ZaesgtpElvPnIe6JEmSpmVC\nTJIkSZM5DlgFvGXAvoOAHcCNfeXXt/t64zb0BlTVTcDWnrhR1iVJkjQtE2KSJEm6nyR7A68HTqiq\n7QNC1gC3V9WOvvLNwOoku/bE3Tbg+M3tvlHX1X8exya5Nsm1mzZtGhQiSZI6yISYJEmSBjkT+Leq\n+sgUMTWgLAP2TRY3TMxs6toZXHVBVa2tqrX77rvvoBBJktRBu4y7AZIkSVpckjwReAXw80n2aotX\nt497JtlBMytr9yQr+mZ27QVs7ZlVtrkt67cnO2d7jbIuSZKkaZkQkyRJUr/HAiuBTw3YdzPw18C7\ngBXAgcANPfv71/naQN/6Xkn2Ax7UE7dhhHVJkiRNy0smJUmS1O9q4Fl921ntvsOBPwWuAbYAR08c\nlGQ1cCRwaU9dlwLPS7J7T9kxwJ3Ale3zUdYlSZI0LWeISZIk6T6q6lbg471lSfZv/3lVVd3elq0H\nTk2ymWaG1gk0X7ie23Po+cDxwEVJzgIOAE4Dzq6qLe3rbRtVXZIkScMwIaYlZdv2HaxauWLOMZIk\naSTW0yStTgL2Bq4FnlNV35kIqKrNSQ4FzgMuplnr6xyaRNZ81SVJkjQlE2JaUlatXMH+6y6ZMmbj\n+iMWqDWSJHVHVb0deHtfWdHcjfLMaY69Dnj2NDEjq0uSJGk6riEmSZIkSZKkTjEhJkmSJEmSpE4x\nISZJkiRJkqROMSEmSZIkSZKkTjEhJkmSJEmSpE4xISZJkiRJkqROMSEmSZIkSZKkTjEhJkmSJEmS\npE4xISZJkiRJkqROMSEmSZIkSZKkTjEhJkmSJEmSpE4xISZJkiRJkqROMSEmSZIkSZKkTjEhJkmS\nJEmSpE4xISZJkiRJkqROMSEmSZIkSZKkTjEhJkmSJKmztm3fMe4mSJLGYJdxN0CSJEmSxmXVyhXs\nv+6SKWM2rj9igVojSVoozhCTJEmSJElSp5gQkyRJkrQseTmkJGkyXjIpSZIkaVnyckhJ0mScISZJ\nkiRJkqROMSEmSZIkSZKkTjEhJkmSJEmSpE4xISZJkiRJkqROMSEmSZIkSZKkTjEhJkmSJEmSpE4x\nISZJkiRJ6qxt23csWIykxWOXcTdAkiRJkqRxWbVyBfuvu2TKmI3rjxgqRtLS4QwxSZIkSZIkdYoJ\nMUmSJEmSJHWKCTEtO17fL0mSJEmSpuIaYlp2hl0DQJIkSZIkdZMzxCRJkiRJktQpJsQkSZIkSZLU\nKSbEJEmSJEmS1CmLPiGW5GVJasB2XE9Mkpyc5BtJ7kzyiSRPGVDXwUk+mmRrkluSnJFkRV/MyOqS\nJEmSJEnS4rOUFtV/NnBnz/Ov9vx7HXAqcCKwATgBuCLJIVX1bYAka4ArgOuAo4DHAG+kSQqeMk91\nSZIkSZIkaZFZSgmxT1fV7f2FSVbRJLHeUFXntWWfAjYCr2Jnguo4YDfghVW1Bbg8yR7AaUn+pKq2\njLKu0Z++JEmSJEmSRmHRXzI5hGcAewDvmSioqjuAi4HDeuIOAy7rS1ZdSJPYeuY81CVJkiRJkqRF\naCklxL6S5IdJbkjyWz3lBwE7gBv74q9v9/XGbegNqKqbgK09caOsS5IkSZIkSYvQUrhk8ls0a3r9\nO7ACeDFwfpLVVXUOsAa4vap29B23GVidZNequruNu21A/ZvbfYy4rvtIcixwLMAjH/nIqc5XkiRJ\nkiRJ82jRJ8Sq6jLgsp6iS5M8EDglyZsnwgYcmgH7JosbJmY2de0MrroAuABg7dq1A2MkSZIkSZI0\n/5bSJZO93gc8BNifZlbW7klW9MXsBWytqu3t881tWb892Tnba5R1SZIkSZIkaRFaqgmxCUWzltcK\n4MC+ff3rfG2gb32vJPsBD+qJG2VdkiRJkiRJWoSWakLsl4Fbga8D1wBbgKMndiZZDRwJXNpzzKXA\n85Ls3lN2DHAncGX7fJR1SZIkSZI6Ytv2/qWo5xYnaX4t+jXEkryfZkH9/6SZvXVMux1fVfcA25Ks\nB05NsplmhtYJNMm+c3uqOh84HrgoyVnAAcBpwNlVtQWgqkZWlyRJkiSpO1atXMH+6y6ZNm7j+iMW\noDWSprPoE2LADcArgP1oFq2/Dvj1qvr7npj1NEmrk4C9gWuB51TVdyYCqmpzkkOB84CLadb6Oocm\nkcU81SVJkiRJkqRFZtEnxKrqZODkaWIKOLPdpoq7Dnj2QtUlSZIkSZKkxWepriEmSZIkSZIkzYoJ\nMUmSJN1PkqOTfCjJN5PcnuQ/krx4QNwrk9yYZFsbc+iAmEck+UBbz61JzmtvXDRvdUmSJE3FhJgk\nSZIGOQG4Hfg94AXAvwDvSvI7EwFJXkRzs6G/Aw4D/gv4cJJDemJ2AS4DHkVzY6RX09zR+4LeFxtl\nXZIkSdNZ9GuISfNh2/YdrFq5Ys4xkiQtY0dW1a09zz+W5OE0ibKJu2+fDryjql4PkORK4KnAOuAl\nbczRwBOAA6vqa23cduDCJKdX1Y3zUJckSdKUTIipk4a5JbK3Q5YkdVlfMmzCZ4GjAJIcADyOZpbW\nxDH3JHlvbxnNbK9PTySwWv8I3A08H7hxlHXN9DwlSVI3ecmkJEmShvUM4Lr23we1jxv6Yq4HHpJk\n3564+8RU1d3AV3rqGGVdkiRJ0zIhJkmSpGm1C9wfBbylLVrTPt7WF7q5b/+aATETcWv6YkdRV3+7\nj01ybZJrN23aNChEkiR1kAkxSZIkTSnJ/sC7gA9W1dv7dld/+IDy/piJuP7yUdbVBFddUFVrq2rt\nvvvuOyhEkiR1kAkxSZIkTSrJQ4BLgZvYubg97Jy9tVffIRPPb+uJ64+ZiOuNGVVdkiRJ0zIhJkmS\npIGSrAY+DOwKHFFVd/TsnljLq3/troOA71XVpp64+8Qk2RU4oKeOUdYlSZI0LRNikiRJup8kuwDv\nBR4LHFZV/927v6q+CnwJOLrnmAe0zy/tCb0UeFqSR/WUvQB4IPBPo65LS9+27TtGEiNJ0lR2GXcD\nJEmStCj9BXA48GqaOz3+VM++z1bVXcBpwDuTbAQ+CfwGTQLt13pi3wf8IXBRklOBPYFzgHdV1Y09\ncaOsS0vYqpUr2H/dJVPGbFx/xAK1RpK0XJkQkyRJ0iDPbR/fPGDfo4GNVfXuJA8G/gA4Ffgv4Ber\n6osTgVW1PcnzgfOA9wB3ARcCJ/ZWOMq6JEmSpmNCTJIkSfdTVfsPGfdXwF9NE3Mz8EsLWZckSdJU\nXENsmXE9BUmSJEmSpKk5Q2yZGWbNBXDdBUmSJEmS1F3OEJMkSZIkSVKnmBCTJEmSJElSp5gQkyRJ\nkiRJUqeYEJMmMewNCryRgSRJ0sJy/CVJmisX1Zcm4Q0KJEmSFifHaZKkuXKGmCRJkiRJkjrFhJgk\nSZIkSZI6xYSYJEmSJEmSOsWEmCRJkiRJkjrFhJgkSZIkSZI6xYSYJEmSJEkLZNv2HSOJkTQ3u4y7\nAZIkSZIkdcWqlSvYf90lU8ZseP3zp61n2/YdrFq5YlTNkjrHhJgkSZIkSYvIMEmzjeuPWKDWSMuT\nl0xKkiRJkiSpU0yISXPkGgCSJEmSJC0tXjIpzZHTmSVJkiRJWlqcISZJkiRJkqROMSEmSZIkSZKk\nTjEhJkmSJEmSpE4xISZJkiRJkqROMSEmSZIkSZKkTjEhJkmSJEmSpE4xISZJkiRJkqROMSEmLYBt\n23eMJEaSJEmSJM3dLuNugNQFq1auYP91l0wZs3H9EQvUGkmSJEmSus0ZYpIkSZIkSeoUE2KSJEmS\nJEnqFBNikiRJkiQtMa5TLM2Na4hJkiRJkrTEuE6xNDfOEJMkSZIkSVKnmBCTFgmnPEuSJEmStDC8\nZFJaJJzyLEmSJGmUtm3fwaqVK+YcIy1HJsQkSZIkSVqG/NJdmpyXTEpLyLCXTHpppSRJkiRJk3OG\nmLSEDPMND/gtjyRJkiRJU3GGmCRJkiRJkjrFhJgkSZIkSR3lsizqKi+ZlCRJkiSpo1yWRV3lDDFJ\nkiRJkiR1igkxaRkaZjqzU54lSZIkSV3lJZPSMjTMtOcNr3/+tPVs276DVStXjKpZkiRJkiQtCibE\npI4aJmnmOgGSJEmSYLgvy/1CXUuJCTFJk/KPniRJkiTwC3UtPybEJE3KP3qSJEmSpOXIRfUlzYkL\n+EuSJEkCPxtoaXGGmKQ5cRaZJEmSJPCzgZYWZ4hJmnfDfgvkt0WSJEmSpIXgDDFJ826Yb4pguG+L\nRrXQvzcDkCRJkhaeY3UtFibEJC0aw/zhG3YatlO1JUmSpMXHyyq1WJgQm4MkBwPnAj8N3Aa8DTi9\nqrzuS5oF/zhKkobhGEySlrdhZ4g5k0xzYUJslpKsAa4ArgOOAh4DvJFmXbZTxtg0SSO0kJdo+gdd\nkqbnGEySlr9hl1zZ8PrnTxvjGFuTMSE2e8cBuwEvrKotwOVJ9gBOS/InbZmkRWrYP4yjvERzuphR\n/kEf1R/+USbyHIxIGpFFOQbziw9JWnheYaK5MCE2e4cBl/UNui4EzgKeCVw8llZJGsooF/oflVEl\n32B0ybXF2KaFnpG3kB9gR9kmP3hrGVuUYzA/lEnS4uTVHJqMCbHZOwj4WG9BVd2UZGu7z4SYpLFZ\njDcfGFWbhkmsLdVE3kwStQs143Chk4uLsU1adByDSZKGtpBj0MU4JtLkUlXjbsOSlGQ7cGJVvamv\n/Gbg76rq5L7yY4Fj26ePB26Yp6btA9w6T3VrMPt8YdnfC8v+Xlj298Kbzz5/VFXtO091d5ZjsEWl\na+fctfMFz7kLuna+4Dkvd0OPv5whNjeDsokZVF5VFwAXzHeDklxbVWvn+3W0k32+sOzvhWV/Lyz7\ne+HZ50uWY7BFoGvn3LXzBc+5C7p2vuA5a6cHjLsBS9hmYK8B5XvS3P5bkiRJo+cYTJIkzZkJsdnb\nQLNOxb2S7Ac8qN0nSZKk0XMMJkmS5syE2OxdCjwvye49ZccAdwJXjqdJwAJcEqD7sc8Xlv29sOzv\nhWV/Lzz7fOlxDLZ4dO2cu3a+4Dl3QdfOFzxntVxUf5aSrAGuA75Ic5vvA4CzgTdV1SnjbJskSdJy\n5RhMkiSNggmxOUhyMHAe8NM0a1a8DTitqnaMtWGSJEnLmGMwSZI0VybEJEmSJEmS1CmuIbZMJDk4\nyUeTbE1yS5IzkqwYd7uWuiQvS1IDtuN6YpLk5CTfSHJnkk8keco4271UJDkwyVuTfD7JjiQfHxAz\nVP/6OzCcIft844Cf+W8PiLPPp5Dk6CQfSvLNJLcn+Y8kLx4Q98okNybZ1sYcOiDmEUk+0NZza5Lz\nkqxemDNZGobp7yQfn+Q9fVVfnP2toXTxfXCYvyPLybDv5ctFkl9Jck2S77Z/l25IckqSXcfdtoXS\n/g24vf378OBxt2c+DPMZZ7lJskuSde2Y664kNyc5Z9ztmi9TjHkqyU+Pu32LxS7jboDmLs1aGlfQ\nrKdxFPAY4I00CU/X0hiNZ9Ms1jvhqz3/XgecCpxIc3erE4ArkhxSVfdLIug+nggcDvwrMNlAa9r+\n9XdgRobpc4B3Aef2PL+7d6d9PpQTgK8BvwfcStPv70qyT1WdC5DkRcD5wGnA1cDLgQ8neVpVfbGN\n2QW4jOb/4BhgL5r1kvYCXrKQJ7TITdvfrX8BTu479q6Jf9jfGlaH3weH/TuyXAz73rJc7E3zPvmn\nNJcj/yTN36iHAa8aX7MW1J8Ct9PcuXa5m+ozznLzt8ChwOk0nyn2Aw4ea4vm1/8G9ugrOwN4KvDp\nhW/OIlVVbkt8A04CNgN79JT9PrC1t8xtVn37MqCAB0+yfxXwfeC1PWUPAjYBfzTu9i/2DXhAz7/f\nB3x8Nv3r78Do+rwt3wj82TT12OfT9/U+A8reBXyt5/kNwN/0/v8AXwDe2VP2YmAH8Oiesl8F7gEe\nO+7zXCzbkP39ceB909Rjf7sNtXX1fXCYvyPLaRvmvWW5b8CZNMmxjLstC3CuPwd8D/i/U30GWOrb\ndJ9xltsGPB/YDhw87raMsQ92bX+2/3LcbVlMm5dMLg+HAZdV1ZaesguB3YBnjqdJnfEMmsz7eyYK\nquoO4GKa/xdNoarumSZk2P71d2BIQ/T5sOzzaVTVrQOKPws8FCDJAcDjuO/P9z3Ae7n/z/enq+pr\nPWX/SDOD6fkjbvaSNV1/z4D9rWF18n1whH9HloQRvrcsZd+lA7MB28udz6WZRTPo/11L1yuAj1XV\ndeNuyBg9H1gDvHvcDVlMTIgtDwfRTPu8V1XdRPMN5UFjadHy85UkP2zXUfitnvKDaGYS3NgXfz32\n/SgM27/+DozeK5LcneT7Sd6X5FF9++3z2XkGzeVVsLOfNvTFXA88JMm+PXH9fX038BXs6+n09veE\n57brPW1NclmSJ/Xtt781LN8Hu2vQe8uykmRFktVJfhY4nmZWyXK/G9txNFcnvGXcDVlAk33GWW6e\nDnypXRN0SzsGaeVuqwAADsFJREFUuCjJw8fdsAX0IuCbwFXjbshi4hpiy8MammnM/Ta3+zR736JZ\nv+rfgRU0l9Kcn2R1VZ1D07+31/1v874ZWJ1k1/aDlGZn2P71d2C0PkizNszNwBOA1wFXJfmxqvp+\nG2Ofz1C7WP5RNN9Sws5+6u/HzT37N2Ffz8qA/ga4EngH8GXgUcAf0vxsP7mqNrYx9reG5c9KB03y\n3rIc3QE8sP3339Gs5bpsJdkbeD3wkqranmTcTZpv033GWW4eRnOZ6OdpEkO7A38CfCDJTy33ZG97\nY6AjgQuW+7nOlAmx5WPQD3YmKdeQquoymsWVJ1ya5IHAKUnePBE24NBMsU8zM2z/+jswIlX16p6n\nVyW5BvgczYLvb+oNHXC4fT5Akv1p1pz5YFW9vW93f3/58z1Hk/V3Vb2uJ+yqJFfQzPD53Xa7N3RQ\ntZOUq9v8WemQad7Ll5tnAKtpFtV/LXAezSLdy9WZwL9V1UfG3ZCFMN1nnGV4aXTa7aiq+i5Akm/R\nfFH2bOCjY2zbQjgSeDBeLnk/XjK5PGymuftVvz0Z/M2l5uZ9wEOA/Wn6fvcBt1jfC9haVdsXuG3L\nzbD96+/APKrmboc3AD/eU2yfDynJQ4BLgZu4710KJ2aC9ffjxPPbeuIG9fVe2Nf3M0V/3081d6r9\nJMP9bNvf6uf7YIfM5L1lOaiqz1TV1VV1Ns0lk7+d5DHjbtd8SPJEmhl/pyfZK8leNMlAgD2T7Da+\n1i2o3s84y81m4AsTybDW1TTrgy7nO01OeBHw5aq6dtwNWWxMiC0PG+hbqyLJfjR34+tfm0ajUzT9\nuwI4sG/f/dYV0awM27/+DiyM3hkP9vkQ2inqH6ZZjPiI9qYQEyb6qX+toYOA71XVpp64/r7eFTgA\n+/o+punvqUz3s21/axDfBztiDu8ty8Vn2sdHj7UV8+exwErgUzSJk83sXEfsZpqF9rtkOc5wvX6S\n8tDcRXrZSrInzU1gnB02gAmx5eFS4HlJdu8pOwa4k2YaqEbrl2nuPPN14BpgC3D0xM6ea7QvHUvr\nlpdh+9ffgXmU5BDg8cB/9BTb59NIsgvNHSMfCxxWVf/du7+qvgp8ifv+fD+gfd7/8/20vhsbvIBm\nbZd/mp/WLz3T9fckx/wI8DPc/2fb/tYwfB/sgNm8tyxDP9M+fm3KqKXrauBZfdtZ7b7DgT8dU7sW\nWu9nnOXmw8CTkuzTU/bzNInQz4+nSQvm/6UZw5gQGyCuqbb0JVlDc6ebL9K8eR8AnA28qapOGWfb\nlrok76dZbPI/aWYqHUMzTf74qjq3jTmJZlHKE2m+ET6B5k4mT6yq74yj3UtFm9w6vH36GmAPmgXc\nAT5SVVuH6V9/B4Y3XZ/TDAJfQjNwuIVm9sMpwF3AU6pqS1uPfT6NJBcArwReTfM+0uuzVXVXkhcD\n76T5P/gk8Bs07zNPay9VJclK4LM0/wen0lyOdQ5wRVUt+8t2hjVdf9Mkdd9A88H268AjgZOAh9L8\nbN/U1mN/ayhdfR8c5m/3WBo2T4Z5L1/4Vs2fJP8EXAH8F82dvn+G5v/5w1X1onG2bSEleRnwt8Du\nVXX7mJszcsN8xllOkuxB8179TeCPaRbVPwvYUFXPGWfb5lv7O/2wqnrKuNuyKFWV2zLYaK59/hjN\nt5LforlLyopxt2upbzRvmDfQ3EL9TppZBC/tiwnNncpubmOuAp467rYvhY1mjYKaZNt/Jv3r78Bo\n+hx4Es3CopuA7cC3gbcDD7fPZ9zXG6f7+W7jXklz18O7aC5LOXRAXf8D+EfgduC7NJdyrB73OS6m\nbbr+Bh5Bk/T9Fs2aId8F3g8cZH+7zXbr4vvgMH+7l9M27Hv5ctnan+Evtu9/t7V/l34HWDnuti1w\nP7ys/T9+8LjbMk/nN+1nnOW20SzB8hGaO6hubse3a8bdrnk+533a8fy6cbdlsW7OEJMkSZIkSVKn\nuIaYJEmSJEmSOsWEmCRJkiRJkjrFhJgkSZIkSZI6xYSYJEmSJEmSOsWEmCRJkiRJkjrFhJgkSZIk\nSZI6xYSYJM1Bkrcn2TjudkiSJI1CkpclqST7L8BrvS/J++eh3gU7h/mS5OFJ7kry3HG3RVquTIhJ\nkiRJkhZUkrXAC4E/HndbFlKS5yb5qySfS/LDJDUorqpuAd5Bx/pHWkgmxCRpbl4JPH7cjZAkSVpi\nfh/496r6j3mo+++B3YCvz0Pdc/VrwEuBu4CvTRP7F8BPJDl03lsldZAJMUmahSQPAqiq7VV11wK/\n9mlepilJkpaqJHsDRwEXjrjeifHZjqraVlUDZ1/NhyS/MORlmn8I7FFVTweunCqwqj4H3AC8fCSN\nlHQfJsQkdVabWKokhyR5R5LNSbYkeXeSh/bEfTzJl9u4y5P8APiHdt/ANcSS/HKSq5P8oK3z2iS/\n2Rfz40k+1L7unW3ML833eUuSJM1EkiOT/GuSrUluS/LBJE8YEPezSf4tybYkNyU5OcnLBySKjgB2\nBS7rO34iqfTrSU5J8o12jHR1kp/oi50Yx/1Ykr9Jcitwc7tv4BpiSZ6a5ANJbm3r/VKSc/pifiTJ\n+Um+meTudgx4UpKRfHauqm9W1d0zOOSfgRckWTGK15e00y7jboAkLQLvBL4LnAo8FvjfwBOS/GTP\ngGUP4HLgQ8B7gTsnqyzJOuANwGdp1n34PvBk4Ejgr9uYn6MZ4FwHnAlsA34V+ECSX6uqd4/4HCVJ\nkmYsyYtpvgj8AnAKzZjod4Brkjytqr7cxj2ZZmyzGXg9cDfN0hJ3DKj2Z9ryGyZ52ROA1cCfA6uA\nVwEfS7K2qm7si303TSLsdcCDpziPXwAupRmXnQ/cBBxAM/76vTZmH+Bf29e8ALilbesfA48Cjpus\n/nn07zT9/RRgPi4vlTrLhJgkwfeA51bVDoAkX6QZBL0ceGsbsy/wmqo6e6qKkjwa+COa5NkRVbW9\nZ196Hi8APg08q+d13wJcBfxJkgsXcpq/JElSvyQrgbOBrwA/U1W3t+UXsfOLv19tw88AVgA/W1Vf\na+P+FvjSgKqfAGysqnsmeen9gMdV1Xfbet5Hk5A7A3hxX+xXgF+aatzUzu76K2AL8OSq+k7PvpN7\nQs8Edgd+rKq+1Za9NcnNwLokZ1fVoPOZT19tHw/GhJg0Ul4yKUlw7kRSqvV24DbgF3vK7qH5NnE6\nL6QZDJ7WmwwD6BmoPRk4iObb1jVJ9mm/kdwb+AjwP4DHQTOAm9jfE7cauF95ktUzPG9JkqSp/ATw\nMOAvJ5JhAFX1n8A/AYe1Y5UVwHOASyaSYW3crbTLTPTZh2Ym2WTeOZEMa+u5nubyysMnvmDs8ZdD\nfIn4VOBA4M29ybC27nvg3i8sj6YZi23vG3tdBgR41sRxSdb0xezZ7lrTNz7bk7n5Xvu4zxzrkdTH\nGWKS1Dddv6q2J/kasH9P8berausQdR3YPn5hipjHtY/nM3mS7aFtux7J5Hcg2tT3/HTgtOmbKEmS\nNJT928cNA/ZdBxxOM4v+ATR3dey/nJFJyqBJME1m0KWUN7Sv9xCapS4mfGWKeiYMMz7bF1hDcwfI\nl04S89Cef3+W5jLKfp/pe34l8AvTN3FSU/WTpDkwISZJMOhbxf7Bx6Rrhk1z3CATs3NPprlscpAv\nto/fpvnGtdevA88FXtJX/lUkSZIWxsSYZ7rZWYPGRrfSzIyfzDBjswnDjNGGaevE+Oz/A942SUzv\nWOt/0iQBJzwZ+DOa8VnvLLSpZsINY037eOsc65HUx4SYJDWXL14/8aRdL2N/4OpZ1DXxLeiPAddM\nEvPl9vGOqrpiqsqqahtwn5gkPwtsm+5YSZKkOdrYPh5Ecylhr4OA22kSNaFJTD12QB2Dyq4HXpJk\nRd+yFb1193sczRpg3xuwbzoT47MnAR+eJGZTW/+uw4yxquqTvc+T/LD95yerauMs2jiZx7SP142w\nTkm4hpgkAfxO362sXwbsBVwyi7ouAnYAp7eJtXv1rHnxGZqB2WuS7NVfQZJ9Z/G6kiRJo3YtzWz1\n45I8aKIwySHA84GPVNU9bVLrcuCI9gZDE3H7AL82oN5P0qyJevAkr/uSJPfOIEvyBOB5wKWzvOnQ\nZ2kurXx1kh/p3TExPmvP4b3AC5I8rb+CJLsneeAsXnuunkaTePzcGF5bWtacISZJzVoU/5zkAzRr\nTPwfmksW/3amFVXV15K8juZOk/+W5D00t/c+BPhR4IVVdU+Sl9Pcmvy6JH9D8w3sw4Cn0wwOHzOo\nfkmSpIVSVT9McgLNwvifTPIOYA/gd4AfAH/YE/46mqTV1e2ds7cDr6QZ46zhvpcrfgS4m2YJiEHr\nen0D+FSSvwIe2L7encBrZ3ke9yQ5FrgU+HyStwFfp1kD7EXsXGPsJJr1vq5K8tfAf9LcdfKJwK/Q\nXAGwcTZtmJDkScAL2qdPactOaZ9/var+vu+Q5wIXTzKTTtIcmBCTpGath98HXk/zvngR8Oqqums2\nlVXVmUm+AvwuzcDthzQLwf5FT8wnk/wkcCpwLM2MtO8An+e+g0tJkqSxqap3J7mDZnzyxzSJrI8D\nJ1XVl3viPpfkuTTraL2OZmbZW2gSWU8FtvXE3prkgzTJqDcOeNmzaZJVr6a5u+J/AL9bVV+aw3l8\nrF124rXAq2gSbTcBH+qJ2ZTk6cApwFE0Cb3baGb2v749p7n68bauXhPPrwTuTYgleTLw+La9kkYs\ns5txKklLX5LTaAZs+1XVzWNujiRJ0rKT5M00X/49uHeWU5K1wL8DP1lV17ZlvwD8C/DSqnrnGJq7\nqCR5K7AWWDvLS0UlTcE1xCRJkiRJc5Zkt77n+wIvBT7Rf8lfmwS7iOau2+qT5OHAbwAnmwyT5oeX\nTEqSJEmSRuHLSf6B5hLDRwD/i2YNrjMGBVfVryxg25aUqroFWDXudkjLmQkxSZIkSdIofIRm8fkf\npVlD9VrgJVX1ybG2SpIGcA0xSZIkSZIkdYpriEmSJEmSJKlTTIhJkiRJkiSpU0yISZIkSZIkqVNM\niEmSJEmSJP3/7diBAAAAAIAgf+tBLoxYEWIAAAAArAQuVQpaxI1KAQAAAABJRU5ErkJggg==\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "_uuid": "8e9959faf29022be5e65344f1dad3abee3f1260e",
        "_cell_guid": "d0b30c17-6c71-400e-b7af-23892a51303a",
        "trusted": true
      },
      "cell_type": "code",
      "source": "plt.figure(figsize = (15, 8))\nplt.hist(np.log(trainDF[trainDF.shipping == 1].price+1), bins = 50, alpha = 0.5, label = \"log price with free shipping\")\nplt.hist(np.log(trainDF[trainDF.shipping == 0].price+1), bins = 50, alpha = 0.5, label = \"log price with shipping\")\nplt.legend(fontsize = 10)\nplt.xlabel('log(price+1)', fontsize=17)\nplt.ylabel('frequency', fontsize=17)\nplt.title('Price distribution wrt Shipping', fontsize=17)\nplt.show()",
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": "<matplotlib.figure.Figure at 0x7fa121f1a470>",
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAA5MAAAH8CAYAAABM0vsvAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nJKkzw6QkSZIkqTPDpCRJkiSps/8DhQr+hdeeo2QAAAAASUVORK5CYII=\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "_uuid": "408b0be382d322968e7be2908fbac5849f70bfbb",
        "_cell_guid": "b4b2aab0-7ae7-4947-939d-d6a860fa81bc"
      },
      "cell_type": "markdown",
      "source": "The minimum price of a product is zero, few products have price close to 0. These seem to be outliers. \nFor the purpose of analysis these will be dropped."
    },
    {
      "metadata": {
        "_uuid": "fd8a43e18c4a2ae79c39a6f7cb3a6904809b2369",
        "_cell_guid": "0ced221e-b82e-4d71-a44b-3f8e1b458bc1",
        "trusted": true
      },
      "cell_type": "code",
      "source": "# Dropping low price \ntrainDF = trainDF.drop(trainDF[(trainDF.price < 1.0)].index)\ntrainDF.shape",
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 9,
          "data": {
            "text/plain": "(1481661, 8)"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "_uuid": "2995d0ecdc9df894f1623e01c63bb5d3836a5236",
        "_cell_guid": "cf71e201-9dbe-4dc9-8586-30ddf8dd27b5"
      },
      "cell_type": "markdown",
      "source": "<a id=\"word_count\"></a>\n#### Word Count"
    },
    {
      "metadata": {
        "_uuid": "4ffac58efb30b9117adbf1944c9c133e679f0a69",
        "_cell_guid": "eca60e18-e359-4928-82ac-6da149f12c99"
      },
      "cell_type": "markdown",
      "source": "wordCount finds the number of words in given text.  "
    },
    {
      "metadata": {
        "_uuid": "9e3bde994615eda483e339a2f79eec1abcfb6248",
        "_cell_guid": "3cab97da-490e-49dd-a366-48a16ccda3cf",
        "trusted": true
      },
      "cell_type": "code",
      "source": "# get name and description lengths\ndef WordCount(text):\n    try:\n        if text == 'No description yet':\n            return 0\n        else:\n            text = text.lower()\n            words = [w for w in text.split(\" \")]\n            return len(words)\n    except: \n        return 0\ntrainDF['desc_len'] = trainDF['item_description'].apply(lambda x: WordCount(x))\ntestDF['desc_len'] = testDF['item_description'].apply(lambda x: WordCount(x))\ntrainDF['name_len'] = trainDF['name'].apply(lambda x: WordCount(x))\ntestDF['name_len'] = testDF['name'].apply(lambda x: WordCount(x))\ntrainDF.head()",
      "execution_count": 10,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 10,
          "data": {
            "text/plain": "   train_id                                 name  item_condition_id  \\\n0         0  MLB Cincinnati Reds T Shirt Size XL                  3   \n1         1     Razer BlackWidow Chroma Keyboard                  3   \n2         2                       AVA-VIV Blouse                  1   \n3         3                Leather Horse Statues                  1   \n4         4                 24K GOLD plated rose                  1   \n\n                                       category_name brand_name  price  \\\n0                                  Men/Tops/T-shirts        NaN   10.0   \n1  Electronics/Computers & Tablets/Components & P...      Razer   52.0   \n2                        Women/Tops & Blouses/Blouse     Target   10.0   \n3                 Home/Home Décor/Home Décor Accents        NaN   35.0   \n4                            Women/Jewelry/Necklaces        NaN   44.0   \n\n   shipping                                   item_description  desc_len  \\\n0         1                                 No description yet         0   \n1         0  This keyboard is in great condition and works ...        36   \n2         1  Adorable top with a hint of lace and a key hol...        29   \n3         1  New with tags. Leather horses. Retail for [rm]...        32   \n4         0          Complete with certificate of authenticity         5   \n\n   name_len  \n0         7  \n1         4  \n2         2  \n3         3  \n4         4  ",
            "text/html": "<div>\n<style scoped>\n    .dataframe tbody tr th:only-of-type {\n        vertical-align: middle;\n    }\n\n    .dataframe tbody tr th {\n        vertical-align: top;\n    }\n\n    .dataframe thead th {\n        text-align: right;\n    }\n</style>\n<table border=\"1\" class=\"dataframe\">\n  <thead>\n    <tr style=\"text-align: right;\">\n      <th></th>\n      <th>train_id</th>\n      <th>name</th>\n      <th>item_condition_id</th>\n      <th>category_name</th>\n      <th>brand_name</th>\n      <th>price</th>\n      <th>shipping</th>\n      <th>item_description</th>\n      <th>desc_len</th>\n      <th>name_len</th>\n    </tr>\n  </thead>\n  <tbody>\n    <tr>\n      <th>0</th>\n      <td>0</td>\n      <td>MLB Cincinnati Reds T Shirt Size XL</td>\n      <td>3</td>\n      <td>Men/Tops/T-shirts</td>\n      <td>NaN</td>\n      <td>10.0</td>\n      <td>1</td>\n      <td>No description yet</td>\n      <td>0</td>\n      <td>7</td>\n    </tr>\n    <tr>\n      <th>1</th>\n      <td>1</td>\n      <td>Razer BlackWidow Chroma Keyboard</td>\n      <td>3</td>\n      <td>Electronics/Computers &amp; Tablets/Components &amp; P...</td>\n      <td>Razer</td>\n      <td>52.0</td>\n      <td>0</td>\n      <td>This keyboard is in great condition and works ...</td>\n      <td>36</td>\n      <td>4</td>\n    </tr>\n    <tr>\n      <th>2</th>\n      <td>2</td>\n      <td>AVA-VIV Blouse</td>\n      <td>1</td>\n      <td>Women/Tops &amp; Blouses/Blouse</td>\n      <td>Target</td>\n      <td>10.0</td>\n      <td>1</td>\n      <td>Adorable top with a hint of lace and a key hol...</td>\n      <td>29</td>\n      <td>2</td>\n    </tr>\n    <tr>\n      <th>3</th>\n      <td>3</td>\n      <td>Leather Horse Statues</td>\n      <td>1</td>\n      <td>Home/Home Décor/Home Décor Accents</td>\n      <td>NaN</td>\n      <td>35.0</td>\n      <td>1</td>\n      <td>New with tags. Leather horses. Retail for [rm]...</td>\n      <td>32</td>\n      <td>3</td>\n    </tr>\n    <tr>\n      <th>4</th>\n      <td>4</td>\n      <td>24K GOLD plated rose</td>\n      <td>1</td>\n      <td>Women/Jewelry/Necklaces</td>\n      <td>NaN</td>\n      <td>44.0</td>\n      <td>0</td>\n      <td>Complete with certificate of authenticity</td>\n      <td>5</td>\n      <td>4</td>\n    </tr>\n  </tbody>\n</table>\n</div>"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "_uuid": "10df81911a2e80e71bf8e2020f97604bdc240812",
        "_cell_guid": "f9ef58be-078a-44a3-9e73-510e88054afb"
      },
      "cell_type": "markdown",
      "source": "<a id=\"Adding_SubCategories\"></a>\n#### Adding SubCategories"
    },
    {
      "metadata": {
        "_uuid": "4b7fc1b9110be153c3582c28d2b3ffe7eb02efd8",
        "_cell_guid": "0e0e3e10-fca3-4064-8944-f8ab64c6d6b0"
      },
      "cell_type": "markdown",
      "source": "Mostly there are three subcategories in category name. \n**SplitCategory** function generates three new columns by splitting category_name in to three sub categories."
    },
    {
      "metadata": {
        "_uuid": "a2d2942d167d1eb3a539035e67ef9a8950eaf4f1",
        "collapsed": true,
        "_cell_guid": "2f30c815-ed53-40b3-a916-a23f7df03d34",
        "trusted": true
      },
      "cell_type": "code",
      "source": "# split category name into 3 parts\ndef split_cat(text):\n    try: return text.split(\"/\")\n    except: return (\"No Label\", \"No Label\", \"No Label\")\ntrainDF['subcat_0'], trainDF['subcat_1'], trainDF['subcat_2'] = \\\nzip(*trainDF['category_name'].apply(lambda x: split_cat(x)))\ntestDF['subcat_0'], testDF['subcat_1'], testDF['subcat_2'] = \\\nzip(*testDF['category_name'].apply(lambda x: split_cat(x)))",
      "execution_count": 11,
      "outputs": []
    },
    {
      "metadata": {
        "_uuid": "3cea510970b24c50873296e4b821b3c0c1e73a2b",
        "_cell_guid": "d438e9e9-9a49-4e35-9b8c-1c16456647f8",
        "trusted": true
      },
      "cell_type": "code",
      "source": "trainDF.head()",
      "execution_count": 12,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 12,
          "data": {
            "text/plain": "   train_id                                 name  item_condition_id  \\\n0         0  MLB Cincinnati Reds T Shirt Size XL                  3   \n1         1     Razer BlackWidow Chroma Keyboard                  3   \n2         2                       AVA-VIV Blouse                  1   \n3         3                Leather Horse Statues                  1   \n4         4                 24K GOLD plated rose                  1   \n\n                                       category_name brand_name  price  \\\n0                                  Men/Tops/T-shirts        NaN   10.0   \n1  Electronics/Computers & Tablets/Components & P...      Razer   52.0   \n2                        Women/Tops & Blouses/Blouse     Target   10.0   \n3                 Home/Home Décor/Home Décor Accents        NaN   35.0   \n4                            Women/Jewelry/Necklaces        NaN   44.0   \n\n   shipping                                   item_description  desc_len  \\\n0         1                                 No description yet         0   \n1         0  This keyboard is in great condition and works ...        36   \n2         1  Adorable top with a hint of lace and a key hol...        29   \n3         1  New with tags. Leather horses. Retail for [rm]...        32   \n4         0          Complete with certificate of authenticity         5   \n\n   name_len     subcat_0             subcat_1            subcat_2  \n0         7          Men                 Tops            T-shirts  \n1         4  Electronics  Computers & Tablets  Components & Parts  \n2         2        Women       Tops & Blouses              Blouse  \n3         3         Home           Home Décor  Home Décor Accents  \n4         4        Women              Jewelry           Necklaces  ",
            "text/html": "<div>\n<style scoped>\n    .dataframe tbody tr th:only-of-type {\n        vertical-align: middle;\n    }\n\n    .dataframe tbody tr th {\n        vertical-align: top;\n    }\n\n    .dataframe thead th {\n        text-align: right;\n    }\n</style>\n<table border=\"1\" class=\"dataframe\">\n  <thead>\n    <tr style=\"text-align: right;\">\n      <th></th>\n      <th>train_id</th>\n      <th>name</th>\n      <th>item_condition_id</th>\n      <th>category_name</th>\n      <th>brand_name</th>\n      <th>price</th>\n      <th>shipping</th>\n      <th>item_description</th>\n      <th>desc_len</th>\n      <th>name_len</th>\n      <th>subcat_0</th>\n      <th>subcat_1</th>\n      <th>subcat_2</th>\n    </tr>\n  </thead>\n  <tbody>\n    <tr>\n      <th>0</th>\n      <td>0</td>\n      <td>MLB Cincinnati Reds T Shirt Size XL</td>\n      <td>3</td>\n      <td>Men/Tops/T-shirts</td>\n      <td>NaN</td>\n      <td>10.0</td>\n      <td>1</td>\n      <td>No description yet</td>\n      <td>0</td>\n      <td>7</td>\n      <td>Men</td>\n      <td>Tops</td>\n      <td>T-shirts</td>\n    </tr>\n    <tr>\n      <th>1</th>\n      <td>1</td>\n      <td>Razer BlackWidow Chroma Keyboard</td>\n      <td>3</td>\n      <td>Electronics/Computers &amp; Tablets/Components &amp; P...</td>\n      <td>Razer</td>\n      <td>52.0</td>\n      <td>0</td>\n      <td>This keyboard is in great condition and works ...</td>\n      <td>36</td>\n      <td>4</td>\n      <td>Electronics</td>\n      <td>Computers &amp; Tablets</td>\n      <td>Components &amp; Parts</td>\n    </tr>\n    <tr>\n      <th>2</th>\n      <td>2</td>\n      <td>AVA-VIV Blouse</td>\n      <td>1</td>\n      <td>Women/Tops &amp; Blouses/Blouse</td>\n      <td>Target</td>\n      <td>10.0</td>\n      <td>1</td>\n      <td>Adorable top with a hint of lace and a key hol...</td>\n      <td>29</td>\n      <td>2</td>\n      <td>Women</td>\n      <td>Tops &amp; Blouses</td>\n      <td>Blouse</td>\n    </tr>\n    <tr>\n      <th>3</th>\n      <td>3</td>\n      <td>Leather Horse Statues</td>\n      <td>1</td>\n      <td>Home/Home Décor/Home Décor Accents</td>\n      <td>NaN</td>\n      <td>35.0</td>\n      <td>1</td>\n      <td>New with tags. Leather horses. Retail for [rm]...</td>\n      <td>32</td>\n      <td>3</td>\n      <td>Home</td>\n      <td>Home Décor</td>\n      <td>Home Décor Accents</td>\n    </tr>\n    <tr>\n      <th>4</th>\n      <td>4</td>\n      <td>24K GOLD plated rose</td>\n      <td>1</td>\n      <td>Women/Jewelry/Necklaces</td>\n      <td>NaN</td>\n      <td>44.0</td>\n      <td>0</td>\n      <td>Complete with certificate of authenticity</td>\n      <td>5</td>\n      <td>4</td>\n      <td>Women</td>\n      <td>Jewelry</td>\n      <td>Necklaces</td>\n    </tr>\n  </tbody>\n</table>\n</div>"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "_uuid": "872a6d8310495ab1c78c36b520efe872fee9c8d0",
        "_cell_guid": "bb3d8bef-661f-482b-92b2-8fc4c0d25790"
      },
      "cell_type": "markdown",
      "source": "<a id=\"Retrieving_Brand_Name\"></a>\n#### Retreiving Brand Name"
    },
    {
      "metadata": {
        "_uuid": "ff44261eb0aada91f6466f6f0721c999614116a1",
        "_cell_guid": "8327d8ec-392c-4c84-abb2-61a03aec9b94"
      },
      "cell_type": "markdown",
      "source": "Retriving missing brand names"
    },
    {
      "metadata": {
        "_uuid": "f7bb0fc8955e673b0ff3813c41dbd80ea256471a",
        "_cell_guid": "99563020-b1c5-4a8a-acd5-13f0e66e1866",
        "collapsed": true,
        "trusted": true
      },
      "cell_type": "code",
      "source": "full_set = pd.concat([trainDF,testDF])\nbrands = set(full_set['brand_name'].values)\ntrainDF.brand_name.fillna(value=\"missing\", inplace=True)\ntestDF.brand_name.fillna(value=\"missing\", inplace=True)",
      "execution_count": 13,
      "outputs": []
    },
    {
      "metadata": {
        "scrolled": true,
        "_uuid": "35253194d51357fb37053bbe849cade91d77c35f",
        "_cell_guid": "56b94742-0a2e-48c8-be99-2ff85c464ad0",
        "trusted": true
      },
      "cell_type": "code",
      "source": "print(\"Missing brand names\")\nprint (\"Number of items with missing brand names in training set : {missingTrain} \".format(missingTrain=len(trainDF.loc[trainDF['brand_name'] == 'missing'])))\nprint (\"Number of items with missing brand names in test set : {missingTest} \".format(missingTest=len(testDF.loc[testDF['brand_name'] == 'missing'])))",
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "stream",
          "text": "Missing brand names\nNumber of items with missing brand names in training set : 632336 \nNumber of items with missing brand names in test set : 295525 \n",
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "_uuid": "06b8facfad26cb61cefa4c742bbefa9a6a8157f1",
        "_cell_guid": "6a60f581-2aba-443c-a98e-453f337c7df6",
        "trusted": true
      },
      "cell_type": "code",
      "source": "uniqueBrands = trainDF[\"brand_name\"].value_counts()\nprint(\"There are\", uniqueBrands.size, \"unique known brands.\")",
      "execution_count": 15,
      "outputs": [
        {
          "output_type": "stream",
          "text": "There are 4808 unique known brands.\n",
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "_uuid": "66e2e32b69b05f4400bd5441a395741ae6c3b77c",
        "_cell_guid": "a9572a2d-37a4-41d5-a3fc-4dae46ef6edf"
      },
      "cell_type": "markdown",
      "source": "For some items brand names could be made out from their name.\n *BrandFinder* is the function which will do this.  "
    },
    {
      "metadata": {
        "_uuid": "de7978aaf12c4d92ce323d8d4f5eaf32e011eef3",
        "collapsed": true,
        "_cell_guid": "6dba67ca-14c1-4f91-8598-fc15ca892d72",
        "trusted": true
      },
      "cell_type": "code",
      "source": "# get to finding!\nmissingBrandNamesTrain = len(trainDF.loc[trainDF['brand_name'] == 'missing'])\nmissingBrandNamesTest = len(testDF.loc[testDF['brand_name'] == 'missing'])\ndef brandfinder(line):\n    brand = line[0]\n    name = line[1]\n    namesplit = name.split(' ')\n    if brand == 'missing':\n        for x in namesplit:\n            if x in brands:\n                return name\n    if name in brands:\n        return name\n    return brand",
      "execution_count": 16,
      "outputs": []
    },
    {
      "metadata": {
        "_uuid": "ec9d3b6247261beb1b607cd92d0d60963ce18d07",
        "collapsed": true,
        "_cell_guid": "a831b96f-954e-4178-b55e-24a1a5976a1a",
        "trusted": true
      },
      "cell_type": "code",
      "source": "trainDF['brand_name'] = trainDF[['brand_name','name']].apply(brandfinder, axis = 1)\ntestDF['brand_name'] = testDF[['brand_name','name']].apply(brandfinder, axis = 1)\nfoundBrandNamesTrain = missingBrandNamesTrain-len(trainDF.loc[trainDF['brand_name'] == 'missing'])\nfoundBrandNamesTest = missingBrandNamesTest-len(testDF.loc[testDF['brand_name'] == 'missing'])",
      "execution_count": 17,
      "outputs": []
    },
    {
      "metadata": {
        "_uuid": "f0e0c93e28089bed37ed4517cb727c33d00de669",
        "_cell_guid": "329af3aa-8f32-4e34-aaa0-08571ea317da",
        "trusted": true
      },
      "cell_type": "code",
      "source": "print (\"Found brand names for {foundTrain} items in training set \".format(foundTrain = foundBrandNamesTrain))\nprint (\"Found brand names for {foundTest} items in test set \".format(foundTest = foundBrandNamesTest))",
      "execution_count": 18,
      "outputs": [
        {
          "output_type": "stream",
          "text": "Found brand names for 137342 items in training set \nFound brand names for 64154 items in test set \n",
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "_uuid": "2e66f29341c6eeafbf924a012e454e5cd8dfcbf7",
        "_cell_guid": "13df4760-c72d-449a-9b19-89970ee37bc3",
        "trusted": true
      },
      "cell_type": "code",
      "source": "# Scale target variable to log.\ntrainDF[\"target\"] = np.log1p(trainDF.price)\n\n# Split training examples into train/dev examples.\ntrainDF, validationDF = train_test_split(trainDF, random_state=123, train_size=0.99)",
      "execution_count": 19,
      "outputs": [
        {
          "output_type": "stream",
          "text": "/opt/conda/lib/python3.6/site-packages/sklearn/model_selection/_split.py:2026: FutureWarning: From version 0.21, test_size will always complement train_size unless both are specified.\n  FutureWarning)\n",
          "name": "stderr"
        }
      ]
    },
    {
      "metadata": {
        "_uuid": "89508884b04236210d5cc1a75b867e5e0cd2395f",
        "_cell_guid": "e99ed964-0acc-415b-988d-c92a318b99ec",
        "trusted": true
      },
      "cell_type": "code",
      "source": "# Calculate number of train/dev/test examples.\nprint(\"Training on\",trainDF.shape[0], \"samples\")\nprint(\"Validating on\", validationDF.shape[0], \"samples\")\nprint(\"Testing on\", testDF.shape[0], \"samples\")",
      "execution_count": 20,
      "outputs": [
        {
          "output_type": "stream",
          "text": "Training on 1466844 samples\nValidating on 14817 samples\nTesting on 693359 samples\n",
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "_uuid": "da7011311372ac8e3cd2f9e1f68bd370aed92b0b",
        "collapsed": true,
        "_cell_guid": "d3d698bc-518e-4037-95ea-fb4080bf6b59",
        "trusted": true
      },
      "cell_type": "code",
      "source": "fullDF = pd.concat([trainDF, validationDF, testDF])\n",
      "execution_count": 21,
      "outputs": []
    },
    {
      "metadata": {
        "_uuid": "8517dbd68b8177c03a84777bbd0f0effeb200e3c",
        "_cell_guid": "1f9e912d-1168-4a3c-9680-070f6c99e75b"
      },
      "cell_type": "markdown",
      "source": "<a id=\"Handling_Missing_Data\"></a>\n#### Handling Missing Data"
    },
    {
      "metadata": {
        "_uuid": "e64d567cd354c4e33910aba7f6d219b19fec2f6e",
        "_cell_guid": "0fb74cf0-021f-479a-ba5b-192873a8be24",
        "trusted": true
      },
      "cell_type": "code",
      "source": "def FillMissingValues(df):\n    df.category_name.fillna(value=\"missing\", inplace=True)\n    df.brand_name.fillna(value=\"missing\", inplace=True)\n    df.item_description.fillna(value=\"missing\", inplace=True)\n    df.item_description.replace('No description yet',\"missing\", inplace=True)\n    return df\n\nprint(\"Filling missing data...\")\nfull_df = FillMissingValues(fullDF)\n",
      "execution_count": 22,
      "outputs": [
        {
          "output_type": "stream",
          "text": "Filling missing data...\n",
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "_uuid": "1b2fa1e5cdb4d022601639bb2044dae684868f19",
        "_cell_guid": "4ede583d-0423-44d0-a779-18d99c9c443a"
      },
      "cell_type": "markdown",
      "source": "<a id=\"categorical_data\"></a>\n#### Categorical Data"
    },
    {
      "metadata": {
        "_uuid": "3a3ec42fe0fe49e209879adddf1e629ba38d9b7c",
        "_cell_guid": "c31ab3c6-c31b-46bb-83f2-e87720d92057",
        "trusted": true
      },
      "cell_type": "code",
      "source": "print(\"Processing categorical data...\")\nle = LabelEncoder()\n# full_df.category = full_df.category_name\nle.fit(fullDF.category_name)\nfullDF['category'] = le.transform(fullDF.category_name)\n\nle.fit(fullDF.brand_name)\nfullDF.brand_name = le.transform(fullDF.brand_name)\n\nle.fit(fullDF.subcat_0)\nfullDF.subcat_0 = le.transform(fullDF.subcat_0)\n\nle.fit(fullDF.subcat_1)\nfullDF.subcat_1 = le.transform(fullDF.subcat_1)\n\nle.fit(fullDF.subcat_2)\nfullDF.subcat_2 = le.transform(fullDF.subcat_2)\n\ndel le",
      "execution_count": 23,
      "outputs": [
        {
          "output_type": "stream",
          "text": "Processing categorical data...\n",
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "_uuid": "a69009430465551001d54e66c433e60a957700eb",
        "_cell_guid": "00eeda54-a9a3-4c41-8dcb-811654abf73a"
      },
      "cell_type": "markdown",
      "source": "<a id=\"Text_to_Sequence\"></a>\n#### Text to Sequence"
    },
    {
      "metadata": {
        "_uuid": "1dd68bc47e6c0b08525a1ae1e2fadd61a7f4a55c",
        "_cell_guid": "89254563-d087-4a9d-83c0-e55fc4765d01",
        "trusted": true
      },
      "cell_type": "code",
      "source": "print(\"Transforming text data to sequences...\")\nraw_text = np.hstack([fullDF.item_description.str.lower(), fullDF.name.str.lower(), fullDF.category_name.str.lower()])\n\nprint(\"   Fitting tokenizer...\")\ntok_raw = Tokenizer()\ntok_raw.fit_on_texts(raw_text)\n\nprint(\"   Transforming text to sequences...\")\nfullDF['seq_item_description'] = tok_raw.texts_to_sequences(fullDF.item_description.str.lower())\nfullDF['seq_name'] = tok_raw.texts_to_sequences(fullDF.name.str.lower())\n# full_df['seq_category'] = tok_raw.texts_to_sequences(full_df.category_name.str.lower())\n\ndel tok_raw",
      "execution_count": 24,
      "outputs": [
        {
          "output_type": "stream",
          "text": "Transforming text data to sequences...\n   Fitting tokenizer...\n   Transforming text to sequences...\n",
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "_uuid": "2f9652fadb3cb4bea5386a928ce416bcdb4fe3ab",
        "_cell_guid": "889855fc-9081-4f3d-80ad-20ebfc72cd4b",
        "trusted": true
      },
      "cell_type": "code",
      "source": "\nbrands = trainDF[\"brand_name\"].value_counts()\nplt.figure(1,figsize = (18,8))\nplt.subplot(121)\nsns.barplot(brands[1:11].values, brands[1:11].index)\nplt.title(\"Top 10  brand in store: Training set\")\n#plt.show()\n\nplt.subplot(122)\nsns.barplot(brands[1:11].values, brands[1:11].index)\nplt.title(\"Top 10  brand in store: Test set\")\nplt.subplots_adjust(top=0.92, bottom=0.08, left=0.10, right=0.95, hspace=6.55, wspace=0.55)\nplt.show()\n\n",
      "execution_count": 25,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": "<matplotlib.figure.Figure at 0x7fa10bcf57b8>",
            "image/png": "iVBORw0KGgoAAAANSUhEUgAABLgAAAITCAYAAAAErqVBAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nPaD9EOGnAAAAAElFTkSuQmCC\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "_uuid": "0f3f9948e42d1c435b9ba61918506c19063248d4",
        "_cell_guid": "5336d4c7-89e9-4ec3-8dea-a51b84cb2553"
      },
      "cell_type": "markdown",
      "source": "It can be seen that data distribution is same for test and train sets"
    },
    {
      "metadata": {
        "_uuid": "87dac5eb7daae7363899ac90f2ed9a6045521d41",
        "_cell_guid": "7078e5fc-6343-42a0-838b-17dc1260e79f"
      },
      "cell_type": "markdown",
      "source": "<a id=\"rmsle\"></a>\n#### RMSLE\n\n\nRoot Mean Squared Logarithmic Error (RMSLE)\n$$\\sqrt{ \\frac{1}{N} \\sum_{i=1}^N (\\log(Y_{p}) - \\log(Y_{t}))^2 }$$\n$ Y_{p}:  predicted    $ \n\n$ Y_{t}:  true  $"
    },
    {
      "metadata": {
        "_uuid": "22588a69e3c725a3260112de71d9988c52654ca8",
        "_cell_guid": "cfcf463b-0b39-4d5d-b7d9-8622845f6335",
        "collapsed": true,
        "trusted": false
      },
      "cell_type": "code",
      "source": "def RMSLE (predicted, true):\n    assert len(predicted) == len(true)\n    return np.sqrt(np.mean(np.power(np.log1p(predicted) - np.log1p(true), 2)))",
      "execution_count": null,
      "outputs": []
    },
    {
      "metadata": {
        "_uuid": "3416d300e00176689783e41fe0df83a010c0a1b2",
        "_cell_guid": "191f3892-2b06-42d5-b6a7-f78bead50dd2"
      },
      "cell_type": "markdown",
      "source": "<a id=\"rnn_model\"></a>\n## <center>RNN Model</center>"
    },
    {
      "metadata": {
        "_uuid": "4d71b03e8ee79f53a79e787f4d03f4f54d54d333",
        "_cell_guid": "9f3785c1-14f1-4ca8-887c-5b1d00736925"
      },
      "cell_type": "markdown",
      "source": "<a id=\"Model_Building\"></a>\n#### Model Building"
    },
    {
      "metadata": {
        "_uuid": "ca7b66c95da0bef9dda21570a5e45cbd423f342f",
        "_cell_guid": "f0197b77-db98-4627-88c2-adea22896a0b"
      },
      "cell_type": "markdown",
      "source": "These are some of the constants related to data."
    },
    {
      "metadata": {
        "_uuid": "879824273ac8d736a442aefcdf03422ef8a4d1ad",
        "collapsed": true,
        "_cell_guid": "4cc97cb8-dc67-4720-8445-2dfbf227b968",
        "trusted": false
      },
      "cell_type": "code",
      "source": "MAX_NAME_SEQ = 10 \nMAX_ITEM_DESC_SEQ = 75 \nMAX_CATEGORY_SEQ = 8 \nMAX_TEXT = np.max([ np.max(fullDF.seq_name.max()),np.max(fullDF.seq_item_description.max())]) + 100\nMAX_CATEGORY = np.max(fullDF.category.max()) + 1\nMAX_BRAND = np.max(fullDF.brand_name.max()) + 1\nMAX_CONDITION = np.max(fullDF.item_condition_id.max()) + 1\nMAX_DESC_LEN = np.max(fullDF.desc_len.max()) + 1\nMAX_NAME_LEN = np.max(fullDF.name_len.max()) + 1\nMAX_SUBCAT_0 = np.max(fullDF.subcat_0.max()) + 1\nMAX_SUBCAT_1 = np.max(fullDF.subcat_1.max()) + 1\nMAX_SUBCAT_2 = np.max(fullDF.subcat_2.max()) + 1",
      "execution_count": null,
      "outputs": []
    },
    {
      "metadata": {
        "_uuid": "6c7a0ad8b5de11087c55cd0f10df7a723f3d6ddd",
        "collapsed": true,
        "_cell_guid": "2a33fdd7-f1ea-46d7-9d00-001dc8a7f13d",
        "trusted": false
      },
      "cell_type": "code",
      "source": "n_trains = trainDF.shape[0]\nn_validate = validationDF.shape[0]\nn_tests = testDF.shape[0]",
      "execution_count": null,
      "outputs": []
    },
    {
      "metadata": {
        "_uuid": "ddc067493e1c73fa603d2d271858740f0e470436",
        "collapsed": true,
        "_cell_guid": "dc6ae5d1-89a5-491d-9ee8-dbbc60486884",
        "trusted": false
      },
      "cell_type": "code",
      "source": "def get_rnn_data(dataset):\n    X = {\n        'name': pad_sequences(dataset.seq_name, maxlen=MAX_NAME_SEQ),\n        'item_desc': pad_sequences(dataset.seq_item_description, maxlen=MAX_ITEM_DESC_SEQ),\n        'brand_name': np.array(dataset.brand_name),\n        'category': np.array(dataset.category),\n#         'category_name': pad_sequences(dataset.seq_category, maxlen=MAX_CATEGORY_SEQ),\n        'item_condition': np.array(dataset.item_condition_id),\n        'num_vars': np.array(dataset[[\"shipping\"]]),\n        'desc_len': np.array(dataset[[\"desc_len\"]]),\n        'name_len': np.array(dataset[[\"name_len\"]]),\n        'subcat_0': np.array(dataset.subcat_0),\n        'subcat_1': np.array(dataset.subcat_1),\n        'subcat_2': np.array(dataset.subcat_2),\n    }\n    return X\n\n",
      "execution_count": null,
      "outputs": []
    },
    {
      "metadata": {
        "_uuid": "a10350fb04cd8bb0dfada385526a3080994ee3e0",
        "_cell_guid": "d379c434-8ded-474c-b859-1618dd9eef56",
        "trusted": false,
        "collapsed": true
      },
      "cell_type": "code",
      "source": "train = fullDF[:n_trains]\nvalidate = fullDF[n_trains:n_trains+n_validate]\ntest = fullDF[n_trains+n_validate:]\n\nX_train = get_rnn_data(train)\nY_train = train.target.values.reshape(-1, 1)\n\nX_validate = get_rnn_data(validate)\nY_validate = validate.target.values.reshape(-1, 1)\n\nX_test = get_rnn_data(test)",
      "execution_count": null,
      "outputs": []
    },
    {
      "metadata": {
        "_uuid": "b59a191630106946eaf18e6e76dc95624157601e",
        "_cell_guid": "9b081fdb-9e2b-4974-96af-0c9d99d60a04",
        "trusted": false,
        "collapsed": true
      },
      "cell_type": "code",
      "source": "# set seed again in case testing models adjustments by looping next 2 blocks\nnp.random.seed(123)\n\ndef new_rnn_model(lr=0.001, decay=0.0):\n    # Inputs\n    name = Input(shape=[X_train[\"name\"].shape[1]], name=\"name\")\n    item_desc = Input(shape=[X_train[\"item_desc\"].shape[1]], name=\"item_desc\")\n    brand_name = Input(shape=[1], name=\"brand_name\")\n#     category = Input(shape=[1], name=\"category\")\n#     category_name = Input(shape=[X_train[\"category_name\"].shape[1]], name=\"category_name\")\n    item_condition = Input(shape=[1], name=\"item_condition\")\n    num_vars = Input(shape=[X_train[\"num_vars\"].shape[1]], name=\"num_vars\")\n    desc_len = Input(shape=[1], name=\"desc_len\")\n    name_len = Input(shape=[1], name=\"name_len\")\n    subcat_0 = Input(shape=[1], name=\"subcat_0\")\n    subcat_1 = Input(shape=[1], name=\"subcat_1\")\n    subcat_2 = Input(shape=[1], name=\"subcat_2\")\n\n    # Embeddings layers (adjust outputs to help model)\n    emb_name = Embedding(MAX_TEXT, 20)(name)\n    emb_item_desc = Embedding(MAX_TEXT, 60)(item_desc)\n    emb_brand_name = Embedding(MAX_BRAND, 10)(brand_name)\n#     emb_category_name = Embedding(MAX_TEXT, 20)(category_name)\n#     emb_category = Embedding(MAX_CATEGORY, 10)(category)\n    emb_item_condition = Embedding(MAX_CONDITION, 5)(item_condition)\n    emb_desc_len = Embedding(MAX_DESC_LEN, 5)(desc_len)\n    emb_name_len = Embedding(MAX_NAME_LEN, 5)(name_len)\n    emb_subcat_0 = Embedding(MAX_SUBCAT_0, 10)(subcat_0)\n    emb_subcat_1 = Embedding(MAX_SUBCAT_1, 10)(subcat_1)\n    emb_subcat_2 = Embedding(MAX_SUBCAT_2, 10)(subcat_2)\n    \n\n    # rnn layers (GRUs are faster than LSTMs and speed is important here)\n    rnn_layer1 = GRU(16) (emb_item_desc)\n    rnn_layer2 = GRU(8) (emb_name)\n#     rnn_layer3 = GRU(8) (emb_category_name)\n\n    # main layers\n    main_l = concatenate([\n        Flatten() (emb_brand_name)\n#         , Flatten() (emb_category)\n        , Flatten() (emb_item_condition)\n        , Flatten() (emb_desc_len)\n        , Flatten() (emb_name_len)\n        , Flatten() (emb_subcat_0)\n        , Flatten() (emb_subcat_1)\n        , Flatten() (emb_subcat_2)\n        , rnn_layer1\n        , rnn_layer2\n#         , rnn_layer3\n        , num_vars\n    ])\n    # (incressing the nodes or adding layers does not effect the time quite as much as the rnn layers)\n    main_l = Dropout(0.1)(Dense(512,kernel_initializer='normal',activation='relu') (main_l))\n    main_l = Dropout(0.1)(Dense(256,kernel_initializer='normal',activation='relu') (main_l))\n    main_l = Dropout(0.1)(Dense(128,kernel_initializer='normal',activation='relu') (main_l))\n    main_l = Dropout(0.1)(Dense(64,kernel_initializer='normal',activation='relu') (main_l))\n\n    # the output layer.\n    output = Dense(1, activation=\"linear\") (main_l)\n    \n    model = Model([name, item_desc, brand_name , item_condition, \n                   num_vars, desc_len, name_len, subcat_0, subcat_1, subcat_2], output)\n\n    optimizer = Adam(lr=lr, decay=decay)\n    # (mean squared error loss function works as well as custom functions)  \n    model.compile(loss = 'mse', optimizer = optimizer)\n\n    return model\n\nmodel = new_rnn_model()\nmodel.summary()\ndel model",
      "execution_count": null,
      "outputs": []
    },
    {
      "metadata": {
        "_uuid": "df6680a03030603432001ff11706cd4b8ff79f41",
        "_cell_guid": "506b7afe-903e-4fe7-8a1a-d5a18542c1a5",
        "trusted": false,
        "collapsed": true
      },
      "cell_type": "code",
      "source": "BATCH_SIZE = 512 * 3\nepochs = 2\n\n# Calculate learning rate decay.\nexp_decay = lambda init, fin, steps: (init/fin)**(1/(steps-1)) - 1\nsteps = int(len(X_train['name']) / BATCH_SIZE) * epochs\nlr_init, lr_fin = 0.005, 0.001\nlr_decay = exp_decay(lr_init, lr_fin, steps)\n\n# Create model and fit it with training dataset.\nrnn_model = new_rnn_model(lr=lr_init, decay=lr_decay)\nrnn_model.fit(\n        X_train, Y_train, epochs=epochs, batch_size=BATCH_SIZE,\n        validation_data=(X_validate, Y_validate), verbose=1,\n)\n",
      "execution_count": null,
      "outputs": []
    },
    {
      "metadata": {
        "_uuid": "b2914624f00600859319ea9250deeaf025b2436b",
        "_cell_guid": "7df8ab0d-463c-4fd2-a7b9-ea71420ba5ac"
      },
      "cell_type": "markdown",
      "source": "<a id=\"Model_Evaluation\"></a>\n#### Model Evaluation"
    },
    {
      "metadata": {
        "_uuid": "2b9b006949d9ade1f325eea3c0ff0cdbef15046c",
        "collapsed": true,
        "_cell_guid": "c3f10896-a0d6-458b-af18-b1e6baff8e40",
        "trusted": false
      },
      "cell_type": "code",
      "source": "print(\"Evaluating the model on validation data...\")\nY_validate_preds_rnn = rnn_model.predict(X_validate, batch_size=BATCH_SIZE)\nprint(\" RMSLE error:\", RMSLE(Y_validate, Y_validate_preds_rnn))",
      "execution_count": null,
      "outputs": []
    },
    {
      "metadata": {
        "_uuid": "d0b58a6af269ecbd956947e25fcc8619d0a458ba",
        "_cell_guid": "364b517f-9390-435c-bc64-10720e5c2db0"
      },
      "cell_type": "markdown",
      "source": "<a id=\"Model_Output\"></a>\n#### Model Output"
    },
    {
      "metadata": {
        "_uuid": "c529a90b6af84d2f0628ad39423b2158a93b35c0",
        "collapsed": true,
        "_cell_guid": "f6293325-aa89-4d9c-ac16-6f52ba0c95cb",
        "trusted": false
      },
      "cell_type": "code",
      "source": "rnn_preds = rnn_model.predict(X_test, batch_size=BATCH_SIZE, verbose=1)\nrnn_preds = np.expm1(rnn_preds)",
      "execution_count": null,
      "outputs": []
    },
    {
      "metadata": {
        "_uuid": "f9354832c3dcdf17a0cc18052f206af1cd190db1",
        "_cell_guid": "6e9f3426-32c3-4fe3-bd26-34f459a6608d"
      },
      "cell_type": "markdown",
      "source": "<a id=\"ridge_model\"></a>\n## <center>Ridge Model</center>"
    },
    {
      "metadata": {
        "_uuid": "e49812396e1f860a0260d4a7a97db014ff8997d8",
        "collapsed": true,
        "_cell_guid": "2123c1db-000a-4e8d-86dd-401b83e7ba82",
        "trusted": false
      },
      "cell_type": "code",
      "source": "fullDF = pd.concat([trainDF, validationDF, testDF])",
      "execution_count": null,
      "outputs": []
    },
    {
      "metadata": {
        "_uuid": "9e6ecee4afc26e83e2d88ac69b847612cbfebe66",
        "_cell_guid": "de6ba5ee-5162-4b87-b681-9a652e7a164c"
      },
      "cell_type": "markdown",
      "source": "Converting to string"
    },
    {
      "metadata": {
        "_uuid": "07ff449421111110fbd41788275311d7a111583f",
        "collapsed": true,
        "_cell_guid": "cee72fce-f733-439d-9cd6-59e8f45886f6",
        "trusted": false
      },
      "cell_type": "code",
      "source": "print(\"Handling missing values...\")\nfullDF['category_name'] = fullDF['category_name'].fillna('missing').astype(str)\nfullDF['subcat_0'] = fullDF['subcat_0'].astype(str)\nfullDF['subcat_1'] = fullDF['subcat_1'].astype(str)\nfullDF['subcat_2'] = fullDF['subcat_2'].astype(str)\nfullDF['brand_name'] = fullDF['brand_name'].fillna('missing').astype(str)\nfullDF['shipping'] = fullDF['shipping'].astype(str)\nfullDF['item_condition_id'] = fullDF['item_condition_id'].astype(str)\nfullDF['desc_len'] = fullDF['desc_len'].astype(str)\nfullDF['name_len'] = fullDF['name_len'].astype(str)\nfullDF['item_description'] = fullDF['item_description'].fillna('No description yet').astype(str)",
      "execution_count": null,
      "outputs": []
    },
    {
      "metadata": {
        "_uuid": "0bc8a0b641f982621126e128126f1651aa76afd4",
        "_cell_guid": "0781540b-b2b3-4d63-82ae-ff83ddb741a5"
      },
      "cell_type": "markdown",
      "source": "**Vectorizig Data**"
    },
    {
      "metadata": {
        "_uuid": "80fa55fa792ebe274ba882bd81c0d0b8aa6e48df",
        "_cell_guid": "b1635a8f-0de7-4eef-8878-2259ac78efb5"
      },
      "cell_type": "markdown",
      "source": "**CountVectorizer** :Converts a collection of text documents to a matrix of token counts"
    },
    {
      "metadata": {
        "_uuid": "d71fab584c44b24d4979d73748f69e31012a09ea",
        "_cell_guid": "a6f58643-4a7b-4189-8bcf-b66ab22e86c9",
        "trusted": false,
        "collapsed": true
      },
      "cell_type": "code",
      "source": "\ndefault_preprocessor = CountVectorizer().build_preprocessor()\ndef build_preprocessor(field):\n    field_idx = list(fullDF.columns).index(field)\n    return lambda x: default_preprocessor(x[field_idx])\n\nvectorizer = FeatureUnion([\n    ('name', CountVectorizer(\n        ngram_range=(1, 2),\n        max_features=50000,\n        preprocessor=build_preprocessor('name'))),\n#     ('category_name', CountVectorizer(\n#         token_pattern='.+',\n#         preprocessor=build_preprocessor('category_name'))),\n    ('subcat_0', CountVectorizer(\n        token_pattern='.+',\n        preprocessor=build_preprocessor('subcat_0'))),\n    ('subcat_1', CountVectorizer(\n        token_pattern='.+',\n        preprocessor=build_preprocessor('subcat_1'))),\n    ('subcat_2', CountVectorizer(\n        token_pattern='.+',\n        preprocessor=build_preprocessor('subcat_2'))),\n    ('brand_name', CountVectorizer(\n        token_pattern='.+',\n        preprocessor=build_preprocessor('brand_name'))),\n    ('shipping', CountVectorizer(\n        token_pattern='\\d+',\n        preprocessor=build_preprocessor('shipping'))),\n    ('item_condition_id', CountVectorizer(\n        token_pattern='\\d+',\n        preprocessor=build_preprocessor('item_condition_id'))),\n    ('desc_len', CountVectorizer(\n        token_pattern='\\d+',\n        preprocessor=build_preprocessor('desc_len'))),\n    ('name_len', CountVectorizer(\n        token_pattern='\\d+',\n        preprocessor=build_preprocessor('name_len'))),\n    ('item_description', TfidfVectorizer(\n        ngram_range=(1, 3),\n        max_features=100000,\n        preprocessor=build_preprocessor('item_description'))),\n])\n\nX = vectorizer.fit_transform(fullDF.values)\n\nX_train = X[:n_trains]\nY_train = trainDF.target.values.reshape(-1, 1)\n\nX_dev = X[n_trains:n_trains+n_devs]\nY_dev = validationDF.target.values.reshape(-1, 1)\n\nX_test = X[n_trains+n_devs:]\nprint(X.shape, X_train.shape, X_dev.shape, X_test.shape)",
      "execution_count": null,
      "outputs": []
    },
    {
      "metadata": {
        "_uuid": "a0fdfd7e913c5aa1440354cb5a02c8f1a73b56ff",
        "_cell_guid": "60a5cfdc-7b4b-4e41-b9d2-fd25a66a6d5d"
      },
      "cell_type": "markdown",
      "source": "This model solves a regression model where the loss function is the linear least squares function and regularization is given by the l2-norm."
    },
    {
      "metadata": {
        "_uuid": "ba806789781211d05d66ba24c688e3c4b2b602b5",
        "_cell_guid": "c2e02e12-5489-43ae-9198-701b7b9b7c4a"
      },
      "cell_type": "markdown",
      "source": "<a id=\"Model_Building1\"></a>\n#### Model Building"
    },
    {
      "metadata": {
        "_uuid": "f3981d8e418f057f075cd4dbadeb0f4659591e3a",
        "collapsed": true,
        "_cell_guid": "50fcbf32-908c-44e0-9edd-1d43343d5c92",
        "trusted": false
      },
      "cell_type": "code",
      "source": "print(\"Fitting Ridge model on training examples...\")\nridge_model = Ridge(\n    solver='auto', fit_intercept=True, alpha=1.0,\n    max_iter=100, normalize=False, tol=0.05, random_state = 1,\n)\nridge_modelCV = RidgeCV(\n    fit_intercept=True, alphas=[5.0],\n    normalize=False, cv = 2, scoring='neg_mean_squared_error',\n)\nridge_model.fit(X_train, Y_train)\nridge_modelCV.fit(X_train, Y_train)",
      "execution_count": null,
      "outputs": []
    },
    {
      "metadata": {
        "_uuid": "ae1ec88b154d53f47045412c5d283c56f7d375f1",
        "_cell_guid": "678842c0-f389-4a4f-ae17-c291350b6b4a"
      },
      "cell_type": "markdown",
      "source": "<a id=\"Model_Evaluation1\"></a>\n#### Model Evaluation"
    },
    {
      "metadata": {
        "_uuid": "752283b8c495d793d022f24fa3c6b25fd3105ce6",
        "collapsed": true,
        "_cell_guid": "f63173f5-8939-4232-bd30-910130b56cfe",
        "trusted": false
      },
      "cell_type": "code",
      "source": "Y_dev_preds_ridge = ridge_model.predict(X_dev)\nY_dev_preds_ridge = Y_validation_preds_ridge.reshape(-1, 1)\nprint(\"RMSL error on validation set:\", rmsle(Y_dev, Y_dev_preds_ridge))",
      "execution_count": null,
      "outputs": []
    },
    {
      "metadata": {
        "_uuid": "9bdd9659e8de98f9b911bd0bce5218ebb2474dc0",
        "collapsed": true,
        "_cell_guid": "a99cce67-743a-4e45-a475-041129835f6f",
        "trusted": false
      },
      "cell_type": "code",
      "source": "Y_dev_preds_ridgeCV = ridge_modelCV.predict(X_dev)\nY_dev_preds_ridgeCV = Y_dev_preds_ridgeCV.reshape(-1, 1)\nprint(\"CV RMSL error on dev set:\", rmsle(Y_dev, Y_dev_preds_ridgeCV))",
      "execution_count": null,
      "outputs": []
    },
    {
      "metadata": {
        "_uuid": "97d0080d80b7aca613f309bbe730fbf01070f8bd",
        "_cell_guid": "f7ec0448-13ad-420a-aa58-ec328c827b0f"
      },
      "cell_type": "markdown",
      "source": "<a id=\"Model_Output1\"></a>\n#### Model Output"
    },
    {
      "metadata": {
        "_uuid": "607154f3f1df5676865bcf4558bb1759238416d8",
        "collapsed": true,
        "_cell_guid": "fd7f58ec-c446-43c1-ac54-b84f58224159",
        "trusted": false
      },
      "cell_type": "code",
      "source": "ridge_preds = ridge_model.predict(X_test)\nridge_preds = np.expm1(ridge_preds)\nridgeCV_preds = ridge_modelCV.predict(X_test)\nridgeCV_preds = np.expm1(ridgeCV_preds)",
      "execution_count": null,
      "outputs": []
    },
    {
      "metadata": {
        "_uuid": "5fd7dfb70a6558948356e1b21a6cd8d07fe01618",
        "_cell_guid": "945f194e-6065-45b5-a062-55de0057c3f0"
      },
      "cell_type": "markdown",
      "source": "<a id=\"ensemblingl\"></a>\n## <center>Ensembling</center>"
    },
    {
      "metadata": {
        "_uuid": "67c5a03b80c25d65b4c80b12b0ca229e280b721d",
        "_cell_guid": "400a6164-62d4-4bc5-a38c-5164078342eb"
      },
      "cell_type": "markdown",
      "source": "Here all the three models are considered. This model will give good result for a pair value of ratio1 and ratio2."
    },
    {
      "metadata": {
        "_uuid": "53ce6b1dbeebbf3a390ba35d5e507999e3454431",
        "collapsed": true,
        "_cell_guid": "bb9d17aa-e0c3-4b31-9847-6d228cb79045",
        "trusted": false
      },
      "cell_type": "code",
      "source": "def aggregate_predicts3(Y1, Y2, Y3, ratio1, ratio2):\n    assert Y1.shape == Y2.shape\n    return Y1 * ratio1 + Y2 * ratio2 + Y3 * (1.0 - ratio1-ratio2)",
      "execution_count": null,
      "outputs": []
    },
    {
      "metadata": {
        "_uuid": "fe24ec02f078660bed2b27d83f1d266af924948d",
        "collapsed": true,
        "_cell_guid": "0405a793-4d28-4570-a503-c1becdc7d707"
      },
      "cell_type": "markdown",
      "source": "<a id=\"Conclusion\"></a>\n## <center>Conclusion</center>"
    },
    {
      "metadata": {
        "_uuid": "04df60f030f7adf4408ecdf8d400debac06dfbef",
        "_cell_guid": "42a9dd75-a703-4cf8-aa12-232786071344"
      },
      "cell_type": "markdown",
      "source": " Gradient boost models like LGBM and XGBoost could be considered.   Given the resource constraint this couldn't be done.\n A better model could be obtained by tweaking RNN parameters. \n"
    },
    {
      "metadata": {
        "_uuid": "c1366a2019f7a948265356a91c0b1822d12892ac",
        "_cell_guid": "83e6a8e4-4573-44bc-9a69-41d1a141d2df"
      },
      "cell_type": "markdown",
      "source": "<a id=\"Refrences\"></a>\n## <center>Refrences</center>"
    },
    {
      "metadata": {
        "_uuid": "7124c5daea567e7903ed443eec414932625e71cb",
        "_cell_guid": "469faccf-d7d3-4646-ad31-4504baac2519"
      },
      "cell_type": "markdown",
      "source": "https://www.kaggle.com/mercari-team"
    },
    {
      "metadata": {
        "_uuid": "92ec0a1c615eaa22f352b87ba966faf44070e389",
        "_cell_guid": "2a1f8a13-2327-40b4-b93a-7c2553194c4b"
      },
      "cell_type": "markdown",
      "source": ""
    }
  ],
  "metadata": {
    "language_info": {
      "name": "python",
      "version": "3.6.4",
      "mimetype": "text/x-python",
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "pygments_lexer": "ipython3",
      "nbconvert_exporter": "python",
      "file_extension": ".py"
    },
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 1
}