Donar choose GIT.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "Dnubs-Nnb3cw"
   },
   "source": [
    "# NAIVE BAYES on different Vectorization Techniques on Donar Choose Dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.feature_extraction.text import CountVectorizer\n",
    "from sklearn.feature_extraction.text import TfidfVectorizer\n",
    "from sklearn.preprocessing import Normalizer\n",
    "\n",
    "%matplotlib inline\n",
    "import warnings\n",
    "warnings.filterwarnings(\"ignore\")\n",
    "\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import nltk\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "from sklearn.feature_extraction.text import TfidfVectorizer\n",
    "from sklearn.feature_extraction.text import CountVectorizer\n",
    "from sklearn.metrics import confusion_matrix\n",
    "from sklearn import metrics\n",
    "from sklearn.metrics import roc_curve, auc\n",
    "\n",
    "import re\n",
    "# Tutorial about Python regular expressions: https://pymotw.com/2/re/\n",
    "import pickle\n",
    "from tqdm import tqdm\n",
    "import os\n",
    "\n",
    "from chart_studio import plotly\n",
    "# import plotly.offline as offline\n",
    "import plotly.graph_objs as go\n",
    "from collections import Counter"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "FYqCIxzFb3c5"
   },
   "source": [
    "## 1. Loading Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "dA1jknyqb3c_"
   },
   "outputs": [
    {
     "data": {
      "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>school_state</th>\n",
       "      <th>teacher_prefix</th>\n",
       "      <th>project_grade_category</th>\n",
       "      <th>teacher_number_of_previously_posted_projects</th>\n",
       "      <th>clean_categories</th>\n",
       "      <th>clean_subcategories</th>\n",
       "      <th>essay</th>\n",
       "      <th>price</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>ca</td>\n",
       "      <td>mrs</td>\n",
       "      <td>grades_prek_2</td>\n",
       "      <td>53</td>\n",
       "      <td>math_science</td>\n",
       "      <td>appliedsciences health_lifescience</td>\n",
       "      <td>i fortunate enough use fairy tale stem kits cl...</td>\n",
       "      <td>725.05</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "  school_state teacher_prefix project_grade_category  \\\n",
       "0           ca            mrs          grades_prek_2   \n",
       "\n",
       "   teacher_number_of_previously_posted_projects clean_categories  \\\n",
       "0                                            53     math_science   \n",
       "\n",
       "                  clean_subcategories  \\\n",
       "0  appliedsciences health_lifescience   \n",
       "\n",
       "                                               essay   price  \n",
       "0  i fortunate enough use fairy tale stem kits cl...  725.05  "
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import pandas as pd\n",
    "data  = pd.read_csv('preprocessed_data.csv', nrows=50000)\n",
    "y = data['project_is_approved'].values\n",
    "X = data.drop(['project_is_approved'], axis=1)\n",
    "X.head(1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "WtVV13Fyb3dH"
   },
   "source": [
    "<h2>1.2 Splitting data into Train and cross validation(or test): Stratified Sampling</h2>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "i-PyprDNb3dI"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(33500, 8) (33500,)\n",
      "(16500, 8) (16500,)\n",
      "****************************************************************************************************\n"
     ]
    }
   ],
   "source": [
    "    \n",
    "# train test split\n",
    "from sklearn.model_selection import train_test_split\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33, stratify=y)\n",
    "\n",
    "print(X_train.shape , y_train.shape)\n",
    "print(X_test.shape, y_test.shape)\n",
    "print('**'*50)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "EnZwVNULb3dO"
   },
   "source": [
    "<h2>2. Make Data Model Ready</h2>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h3> 2.1 Encoding Textual Features using BAG OF WORDS</h3>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "-SczZKiab3dR"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " after encoding in bow the size of :\n",
      " train feature -- (33500, 5000) (33500,)\n",
      "test feature -- (16500, 5000) (16500,)\n",
      "****************************************************************************************************\n",
      " after encoding in bow the size of :\n",
      " train feature -- (33500, 51) (33500,)\n",
      "test feature -- (16500, 51) (16500,)\n",
      "the features are :  ['ak', 'al', 'ar', 'az', 'ca', 'co', 'ct', 'dc', 'de', 'fl', 'ga', 'hi', 'ia', 'id', 'il', 'in', 'ks', 'ky', 'la', 'ma', 'md', 'me', 'mi', 'mn', 'mo', 'ms', 'mt', 'nc', 'nd', 'ne', 'nh', 'nj', 'nm', 'nv', 'ny', 'oh', 'ok', 'or', 'pa', 'ri', 'sc', 'sd', 'tn', 'tx', 'ut', 'va', 'vt', 'wa', 'wi', 'wv', 'wy']\n",
      "****************************************************************************************************\n",
      " after encoding in bow the size of :\n",
      " train feature -- (33500, 5) (33500,)\n",
      "test feature -- (16500, 5) (16500,)\n",
      "the features are :  ['dr', 'mr', 'mrs', 'ms', 'teacher']\n",
      "****************************************************************************************************\n",
      " after encoding in bow the size of :\n",
      " train feature -- (33500, 4) (33500,)\n",
      "test feature -- (16500, 4) (16500,)\n",
      "the features are :  ['grades_3_5', 'grades_6_8', 'grades_9_12', 'grades_prek_2']\n",
      "****************************************************************************************************\n",
      " after encoding in bow the size of :\n",
      " train feature -- (33500, 9) (33500,)\n",
      "test feature -- (16500, 9) (16500,)\n",
      "the features are :  ['appliedlearning', 'care_hunger', 'health_sports', 'history_civics', 'literacy_language', 'math_science', 'music_arts', 'specialneeds', 'warmth']\n",
      "****************************************************************************************************\n",
      " after encoding in bow the size of :\n",
      " train feature -- (33500, 30) (33500,)\n",
      "test feature -- (16500, 30) (16500,)\n",
      "the features are :  ['appliedsciences', 'care_hunger', 'charactereducation', 'civics_government', 'college_careerprep', 'communityservice', 'earlydevelopment', 'economics', 'environmentalscience', 'esl', 'extracurricular', 'financialliteracy', 'foreignlanguages', 'gym_fitness', 'health_lifescience', 'health_wellness', 'history_geography', 'literacy', 'literature_writing', 'mathematics', 'music', 'nutritioneducation', 'other', 'parentinvolvement', 'performingarts', 'socialsciences', 'specialneeds', 'teamsports', 'visualarts', 'warmth']\n",
      "****************************************************************************************************\n"
     ]
    }
   ],
   "source": [
    "\n",
    "# encoding essay attribute using count vectorizer\n",
    "vectorizer = CountVectorizer(min_df=10,ngram_range=(1,4), max_features=5000)\n",
    "vectorizer.fit(X_train['essay'].values)\n",
    "essay_feature_bow=vectorizer.get_feature_names()\n",
    "X_train_essay_bow=vectorizer.transform(X_train['essay'].values)\n",
    "X_test_essay_bow=vectorizer.transform(X_test['essay'].values)\n",
    "\n",
    "print(\" after encoding in bow the size of :\")\n",
    "print(\" train feature --\",X_train_essay_bow.shape,y_train.shape)\n",
    "print(\"test feature --\",X_test_essay_bow.shape,y_test.shape)\n",
    "# print('the features are : ',essay_feature_bow)\n",
    "print('**'*50)\n",
    "\n",
    "# encoding school_state using count vectorizer\n",
    "vectorizer=CountVectorizer()\n",
    "vectorizer.fit(X_train['school_state'].values)\n",
    "school_state_feature_bow=vectorizer.get_feature_names()\n",
    "X_train_school_state_bow=vectorizer.transform(X_train['school_state'].values)\n",
    "X_test_school_state_bow=vectorizer.transform(X_test['school_state'].values)\n",
    "\n",
    "print(\" after encoding in bow the size of :\")\n",
    "print(\" train feature --\",X_train_school_state_bow.shape,y_train.shape)\n",
    "print(\"test feature --\",X_test_school_state_bow.shape,y_test.shape)\n",
    "print('the features are : ',school_state_feature_bow)\n",
    "print('**'*50)\n",
    "\n",
    "#encoding teacher_prefix using count vectorizer\n",
    "vectorizer=CountVectorizer()\n",
    "vectorizer.fit(X_train['teacher_prefix'].values)\n",
    "teacher_prefix_feature_bow=vectorizer.get_feature_names()\n",
    "X_train_teacher_prefix_bow=vectorizer.transform(X_train['teacher_prefix'].values)\n",
    "X_test_teacher_prefix_bow=vectorizer.transform(X_test['teacher_prefix'].values)\n",
    "\n",
    "print(\" after encoding in bow the size of :\")\n",
    "print(\" train feature --\",X_train_teacher_prefix_bow.shape,y_train.shape)\n",
    "print(\"test feature --\",X_test_teacher_prefix_bow.shape,y_test.shape)\n",
    "print('the features are : ',teacher_prefix_feature_bow)\n",
    "print('**'*50)\n",
    "\n",
    "#encoding project_grade_category using count vectorizer\n",
    "vectorizer=CountVectorizer()\n",
    "vectorizer.fit(X_train['project_grade_category'].values)\n",
    "project_grade_category_feature_bow=vectorizer.get_feature_names()\n",
    "X_train_project_grade_category_bow=vectorizer.transform(X_train['project_grade_category'].values)\n",
    "X_test_project_grade_category_bow=vectorizer.transform(X_test['project_grade_category'].values)\n",
    "\n",
    "print(\" after encoding in bow the size of :\")\n",
    "print(\" train feature --\",X_train_project_grade_category_bow.shape,y_train.shape)\n",
    "print(\"test feature --\",X_test_project_grade_category_bow.shape,y_test.shape)\n",
    "print('the features are : ',project_grade_category_feature_bow)\n",
    "print('**'*50)\n",
    "\n",
    "#encoding clean_categories using count vectorizer\n",
    "vectorizer=CountVectorizer()\n",
    "vectorizer.fit(X_train['clean_categories'].values)\n",
    "clean_categories_feature_bow=vectorizer.get_feature_names()\n",
    "X_train_clean_categories_bow=vectorizer.transform(X_train['clean_categories'].values)\n",
    "X_test_clean_categories_bow=vectorizer.transform(X_test['clean_categories'].values)\n",
    "\n",
    "print(\" after encoding in bow the size of :\")\n",
    "print(\" train feature --\",X_train_clean_categories_bow.shape,y_train.shape)\n",
    "print(\"test feature --\",X_test_clean_categories_bow.shape,y_test.shape)\n",
    "print('the features are : ',clean_categories_feature_bow)\n",
    "print('**'*50)\n",
    "\n",
    "#encoding clean_subcategories using count vectorizer\n",
    "vectorizer=CountVectorizer()\n",
    "vectorizer.fit(X_train['clean_subcategories'].values)\n",
    "clean_subcategories_feature_bow=vectorizer.get_feature_names()\n",
    "X_train_clean_subcategories_bow=vectorizer.transform(X_train['clean_subcategories'].values)\n",
    "X_test_clean_subcategories_bow=vectorizer.transform(X_test['clean_subcategories'].values)\n",
    "\n",
    "print(\" after encoding in bow the size of :\")\n",
    "print(\" train feature --\",X_train_clean_subcategories_bow.shape,y_train.shape)\n",
    "print(\"test feature --\",X_test_clean_subcategories_bow.shape,y_test.shape)\n",
    "print('the features are : ',clean_subcategories_feature_bow)\n",
    "print('**'*50)\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h3>2.2 Encoding Text features using TFIDF </h2>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " after encoding in bow the size of :\n",
      " train feature -- (33500, 5000) (33500,)\n",
      "test feature -- (16500, 5000) (16500,)\n",
      "****************************************************************************************************\n",
      " after encoding in bow the size of :\n",
      " train feature -- (33500, 51) (33500,)\n",
      "test feature -- (16500, 51) (16500,)\n",
      "the features are :  ['ak', 'al', 'ar', 'az', 'ca', 'co', 'ct', 'dc', 'de', 'fl', 'ga', 'hi', 'ia', 'id', 'il', 'in', 'ks', 'ky', 'la', 'ma', 'md', 'me', 'mi', 'mn', 'mo', 'ms', 'mt', 'nc', 'nd', 'ne', 'nh', 'nj', 'nm', 'nv', 'ny', 'oh', 'ok', 'or', 'pa', 'ri', 'sc', 'sd', 'tn', 'tx', 'ut', 'va', 'vt', 'wa', 'wi', 'wv', 'wy']\n",
      "****************************************************************************************************\n",
      " after encoding in bow the size of :\n",
      " train feature -- (33500, 5) (33500,)\n",
      "test feature -- (16500, 5) (16500,)\n",
      "the features are :  ['dr', 'mr', 'mrs', 'ms', 'teacher']\n",
      "****************************************************************************************************\n",
      " after encoding in bow the size of :\n",
      " train feature -- (33500, 4) (33500,)\n",
      "test feature -- (16500, 4) (16500,)\n",
      "the features are :  ['grades_3_5', 'grades_6_8', 'grades_9_12', 'grades_prek_2']\n",
      "****************************************************************************************************\n",
      " after encoding in bow the size of :\n",
      " train feature -- (33500, 9) (33500,)\n",
      "test feature -- (16500, 9) (16500,)\n",
      "the features are :  ['appliedlearning', 'care_hunger', 'health_sports', 'history_civics', 'literacy_language', 'math_science', 'music_arts', 'specialneeds', 'warmth']\n",
      "****************************************************************************************************\n",
      " after encoding in bow the size of :\n",
      " train feature -- (33500, 30) (33500,)\n",
      "test feature -- (16500, 30) (16500,)\n",
      "the features are :  ['appliedsciences', 'care_hunger', 'charactereducation', 'civics_government', 'college_careerprep', 'communityservice', 'earlydevelopment', 'economics', 'environmentalscience', 'esl', 'extracurricular', 'financialliteracy', 'foreignlanguages', 'gym_fitness', 'health_lifescience', 'health_wellness', 'history_geography', 'literacy', 'literature_writing', 'mathematics', 'music', 'nutritioneducation', 'other', 'parentinvolvement', 'performingarts', 'socialsciences', 'specialneeds', 'teamsports', 'visualarts', 'warmth']\n",
      "****************************************************************************************************\n"
     ]
    }
   ],
   "source": [
    "\n",
    "# encoding essay attribute using count vectorizer\n",
    "vectorizer = TfidfVectorizer(min_df=10,ngram_range=(1,4), max_features=5000)\n",
    "vectorizer.fit(X_train['essay'].values)\n",
    "essay_feature_TF=vectorizer.get_feature_names()\n",
    "X_train_essay_TF=vectorizer.transform(X_train['essay'].values)\n",
    "X_test_essay_TF=vectorizer.transform(X_test['essay'].values)\n",
    "\n",
    "print(\" after encoding in bow the size of :\")\n",
    "print(\" train feature --\",X_train_essay_TF.shape,y_train.shape)\n",
    "print(\"test feature --\",X_test_essay_TF.shape,y_test.shape)\n",
    "# print('the features are : ',essay_feature_bow)\n",
    "print('**'*50)\n",
    "\n",
    "# encoding school_state using count vectorizer\n",
    "vectorizer=TfidfVectorizer()\n",
    "vectorizer.fit(X_train['school_state'].values)\n",
    "school_state_feature_TF=vectorizer.get_feature_names()\n",
    "X_train_school_state_TF=vectorizer.transform(X_train['school_state'].values)\n",
    "X_test_school_state_TF=vectorizer.transform(X_test['school_state'].values)\n",
    "\n",
    "print(\" after encoding in bow the size of :\")\n",
    "print(\" train feature --\",X_train_school_state_TF.shape,y_train.shape)\n",
    "print(\"test feature --\",X_test_school_state_TF.shape,y_test.shape)\n",
    "print('the features are : ',school_state_feature_TF)\n",
    "print('**'*50)\n",
    "\n",
    "#encoding teacher_prefix using count vectorizer\n",
    "vectorizer=TfidfVectorizer()\n",
    "vectorizer.fit(X_train['teacher_prefix'].values)\n",
    "teacher_prefix_feature_TF=vectorizer.get_feature_names()\n",
    "X_train_teacher_prefix_TF=vectorizer.transform(X_train['teacher_prefix'].values)\n",
    "X_test_teacher_prefix_TF=vectorizer.transform(X_test['teacher_prefix'].values)\n",
    "\n",
    "print(\" after encoding in bow the size of :\")\n",
    "print(\" train feature --\",X_train_teacher_prefix_TF.shape,y_train.shape)\n",
    "print(\"test feature --\",X_test_teacher_prefix_TF.shape,y_test.shape)\n",
    "print('the features are : ',teacher_prefix_feature_TF)\n",
    "print('**'*50)\n",
    "\n",
    "#encoding project_grade_category using count vectorizer\n",
    "vectorizer=TfidfVectorizer()\n",
    "vectorizer.fit(X_train['project_grade_category'].values)\n",
    "project_grade_category_feature_TF=vectorizer.get_feature_names()\n",
    "X_train_project_grade_category_TF=vectorizer.transform(X_train['project_grade_category'].values)\n",
    "X_test_project_grade_category_TF=vectorizer.transform(X_test['project_grade_category'].values)\n",
    "\n",
    "print(\" after encoding in bow the size of :\")\n",
    "print(\" train feature --\",X_train_project_grade_category_TF.shape,y_train.shape)\n",
    "print(\"test feature --\",X_test_project_grade_category_TF.shape,y_test.shape)\n",
    "print('the features are : ',project_grade_category_feature_TF)\n",
    "print('**'*50)\n",
    "\n",
    "#encoding clean_categories using count vectorizer\n",
    "vectorizer=TfidfVectorizer()\n",
    "vectorizer.fit(X_train['clean_categories'].values)\n",
    "clean_categories_feature_TF=vectorizer.get_feature_names()\n",
    "X_train_clean_categories_TF=vectorizer.transform(X_train['clean_categories'].values)\n",
    "X_test_clean_categories_TF=vectorizer.transform(X_test['clean_categories'].values)\n",
    "\n",
    "print(\" after encoding in bow the size of :\")\n",
    "print(\" train feature --\",X_train_clean_categories_TF.shape,y_train.shape)\n",
    "print(\"test feature --\",X_test_clean_categories_TF.shape,y_test.shape)\n",
    "print('the features are : ',clean_categories_feature_TF)\n",
    "print('**'*50)\n",
    "\n",
    "#encoding clean_subcategories using count vectorizer\n",
    "vectorizer=TfidfVectorizer()\n",
    "vectorizer.fit(X_train['clean_subcategories'].values)\n",
    "clean_subcategories_feature_TF=vectorizer.get_feature_names()\n",
    "X_train_clean_subcategories_TF=vectorizer.transform(X_train['clean_subcategories'].values)\n",
    "X_test_clean_subcategories_TF=vectorizer.transform(X_test['clean_subcategories'].values)\n",
    "\n",
    "print(\" after encoding in bow the size of :\")\n",
    "print(\" train feature --\",X_train_clean_subcategories_TF.shape,y_train.shape)\n",
    "print(\"test feature --\",X_test_clean_subcategories_TF.shape,y_test.shape)\n",
    "print('the features are : ',clean_subcategories_feature_TF)\n",
    "print('**'*50)\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "bGy86kgHb3dX"
   },
   "source": [
    "<h3>2.3 Encoding numerical features using Normalizer</h2>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "EfXkofX1b3da"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " after encoding using normalizer the size of :\n",
      " train feature -- (33500, 1) (33500,)\n",
      "test feature -- (16500, 1) (16500,)\n",
      "****************************************************************************************************\n",
      " after encoding using normalizer the size of :\n",
      " train feature -- (33500, 1) (33500,)\n",
      "test feature -- (16500, 1) (16500,)\n",
      "****************************************************************************************************\n"
     ]
    }
   ],
   "source": [
    "normalizer = Normalizer()\n",
    "normalizer.fit(X_train['price'].values.reshape(-1,1))\n",
    "X_train_price_norm=normalizer.transform(X_train['price'].values.reshape(-1,1))\n",
    "X_test_price_norm=normalizer.transform(X_test['price'].values.reshape(-1,1))\n",
    "\n",
    "print(\" after encoding using normalizer the size of :\")\n",
    "print(\" train feature --\",X_train_price_norm.shape,y_train.shape)\n",
    "print(\"test feature --\",X_test_price_norm.shape,y_test.shape)\n",
    "print('**'*50)\n",
    "\n",
    "\n",
    "normalizer = Normalizer()\n",
    "normalizer.fit(X_train['teacher_number_of_previously_posted_projects'].values.reshape(-1,1))\n",
    "X_train_teacher_number_of_previously_posted_projects_norm=normalizer.transform(X_train['teacher_number_of_previously_posted_projects'].values.reshape(-1,1))\n",
    "X_test_teacher_number_of_previously_posted_projects_norm=normalizer.transform(X_test['teacher_number_of_previously_posted_projects'].values.reshape(-1,1))\n",
    "\n",
    "print(\" after encoding using normalizer the size of :\")\n",
    "print(\" train feature --\",X_train_teacher_number_of_previously_posted_projects_norm.shape,y_train.shape)\n",
    "print(\"test feature --\",X_test_teacher_number_of_previously_posted_projects_norm.shape,y_test.shape)\n",
    "print('**'*50)\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 2.4 Final Data Preparation- merging all the vectorized features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(33500, 5101) (33500,)\n",
      "(16500, 5101) (16500,)\n",
      "****************************************************************************************************\n",
      "(33500, 5101) (33500,)\n",
      "(16500, 5101) (16500,)\n",
      "****************************************************************************************************\n"
     ]
    }
   ],
   "source": [
    "# merge two sparse matrices: https://stackoverflow.com/a/19710648/4084039\n",
    "from scipy.sparse import hstack\n",
    "X_train_BOW = hstack((X_train_essay_bow, X_train_school_state_bow, X_train_teacher_prefix_bow, X_train_project_grade_category_bow,X_train_clean_categories_bow,X_train_clean_subcategories_bow, X_train_price_norm,X_train_teacher_number_of_previously_posted_projects_norm)).tocsr()\n",
    "X_test_BOW = hstack((X_test_essay_bow, X_test_school_state_bow, X_test_teacher_prefix_bow, X_test_project_grade_category_bow,X_test_clean_categories_bow,X_test_clean_subcategories_bow, X_test_price_norm,X_test_teacher_number_of_previously_posted_projects_norm)).tocsr()\n",
    "\n",
    "print(X_train_BOW.shape, y_train.shape)\n",
    "print(X_test_BOW.shape, y_test.shape)\n",
    "print('**'*50)\n",
    "\n",
    "X_train_TFIDF = hstack((X_train_essay_TF, X_train_school_state_TF, X_train_teacher_prefix_TF, X_train_project_grade_category_TF,X_train_clean_categories_TF,X_train_clean_subcategories_TF, X_train_price_norm,X_train_teacher_number_of_previously_posted_projects_norm)).tocsr()\n",
    "X_test_TFIDF = hstack((X_test_essay_TF, X_test_school_state_TF, X_test_teacher_prefix_TF, X_test_project_grade_category_TF,X_test_clean_categories_TF,X_test_clean_subcategories_TF, X_test_price_norm,X_test_teacher_number_of_previously_posted_projects_norm)).tocsr()\n",
    "\n",
    "print(X_train_TFIDF.shape, y_train.shape)\n",
    "print(X_test_TFIDF.shape, y_test.shape)\n",
    "print('**'*50)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "yYHPYadYb3dh"
   },
   "source": [
    "<h2>3 Appling NB on different kind of featurization as mentioned in the instructions</h2>\n",
    "\n",
    "<br>Apply NB on different kind of featurization as mentioned in the instructions"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "DNwqilFxb3di"
   },
   "source": [
    "## Training    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Hyper-parameter tuning on BOW features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "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>mean_fit_time</th>\n",
       "      <th>std_fit_time</th>\n",
       "      <th>mean_score_time</th>\n",
       "      <th>std_score_time</th>\n",
       "      <th>param_alpha</th>\n",
       "      <th>params</th>\n",
       "      <th>split0_test_score</th>\n",
       "      <th>split1_test_score</th>\n",
       "      <th>split2_test_score</th>\n",
       "      <th>split3_test_score</th>\n",
       "      <th>...</th>\n",
       "      <th>mean_test_score</th>\n",
       "      <th>std_test_score</th>\n",
       "      <th>rank_test_score</th>\n",
       "      <th>split0_train_score</th>\n",
       "      <th>split1_train_score</th>\n",
       "      <th>split2_train_score</th>\n",
       "      <th>split3_train_score</th>\n",
       "      <th>split4_train_score</th>\n",
       "      <th>mean_train_score</th>\n",
       "      <th>std_train_score</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.163707</td>\n",
       "      <td>0.009532</td>\n",
       "      <td>0.019389</td>\n",
       "      <td>0.001357</td>\n",
       "      <td>0.0001</td>\n",
       "      <td>{'alpha': 0.0001}</td>\n",
       "      <td>0.685899</td>\n",
       "      <td>0.695508</td>\n",
       "      <td>0.689933</td>\n",
       "      <td>0.683829</td>\n",
       "      <td>...</td>\n",
       "      <td>0.687440</td>\n",
       "      <td>0.004816</td>\n",
       "      <td>2</td>\n",
       "      <td>0.744240</td>\n",
       "      <td>0.742061</td>\n",
       "      <td>0.742930</td>\n",
       "      <td>0.744787</td>\n",
       "      <td>0.746388</td>\n",
       "      <td>0.744081</td>\n",
       "      <td>0.001500</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.161706</td>\n",
       "      <td>0.008836</td>\n",
       "      <td>0.017992</td>\n",
       "      <td>0.000632</td>\n",
       "      <td>0.001</td>\n",
       "      <td>{'alpha': 0.001}</td>\n",
       "      <td>0.686202</td>\n",
       "      <td>0.695502</td>\n",
       "      <td>0.689908</td>\n",
       "      <td>0.683733</td>\n",
       "      <td>...</td>\n",
       "      <td>0.687464</td>\n",
       "      <td>0.004821</td>\n",
       "      <td>1</td>\n",
       "      <td>0.744198</td>\n",
       "      <td>0.742022</td>\n",
       "      <td>0.742890</td>\n",
       "      <td>0.744765</td>\n",
       "      <td>0.746358</td>\n",
       "      <td>0.744047</td>\n",
       "      <td>0.001505</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.157109</td>\n",
       "      <td>0.001165</td>\n",
       "      <td>0.018191</td>\n",
       "      <td>0.000399</td>\n",
       "      <td>0.01</td>\n",
       "      <td>{'alpha': 0.01}</td>\n",
       "      <td>0.686314</td>\n",
       "      <td>0.695491</td>\n",
       "      <td>0.689865</td>\n",
       "      <td>0.683599</td>\n",
       "      <td>...</td>\n",
       "      <td>0.687433</td>\n",
       "      <td>0.004847</td>\n",
       "      <td>3</td>\n",
       "      <td>0.744126</td>\n",
       "      <td>0.741957</td>\n",
       "      <td>0.742830</td>\n",
       "      <td>0.744725</td>\n",
       "      <td>0.746307</td>\n",
       "      <td>0.743989</td>\n",
       "      <td>0.001510</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.157307</td>\n",
       "      <td>0.003196</td>\n",
       "      <td>0.018191</td>\n",
       "      <td>0.000976</td>\n",
       "      <td>0.1</td>\n",
       "      <td>{'alpha': 0.1}</td>\n",
       "      <td>0.686338</td>\n",
       "      <td>0.695404</td>\n",
       "      <td>0.689781</td>\n",
       "      <td>0.683352</td>\n",
       "      <td>...</td>\n",
       "      <td>0.687325</td>\n",
       "      <td>0.004881</td>\n",
       "      <td>4</td>\n",
       "      <td>0.743932</td>\n",
       "      <td>0.741780</td>\n",
       "      <td>0.742658</td>\n",
       "      <td>0.744588</td>\n",
       "      <td>0.746155</td>\n",
       "      <td>0.743823</td>\n",
       "      <td>0.001521</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0.158108</td>\n",
       "      <td>0.006172</td>\n",
       "      <td>0.018190</td>\n",
       "      <td>0.000400</td>\n",
       "      <td>1</td>\n",
       "      <td>{'alpha': 1}</td>\n",
       "      <td>0.685815</td>\n",
       "      <td>0.694649</td>\n",
       "      <td>0.689297</td>\n",
       "      <td>0.682411</td>\n",
       "      <td>...</td>\n",
       "      <td>0.686660</td>\n",
       "      <td>0.004902</td>\n",
       "      <td>5</td>\n",
       "      <td>0.742862</td>\n",
       "      <td>0.740668</td>\n",
       "      <td>0.741524</td>\n",
       "      <td>0.743596</td>\n",
       "      <td>0.745116</td>\n",
       "      <td>0.742753</td>\n",
       "      <td>0.001560</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>0.153309</td>\n",
       "      <td>0.000800</td>\n",
       "      <td>0.018990</td>\n",
       "      <td>0.000633</td>\n",
       "      <td>10</td>\n",
       "      <td>{'alpha': 10}</td>\n",
       "      <td>0.680188</td>\n",
       "      <td>0.687300</td>\n",
       "      <td>0.683953</td>\n",
       "      <td>0.674619</td>\n",
       "      <td>...</td>\n",
       "      <td>0.680297</td>\n",
       "      <td>0.004866</td>\n",
       "      <td>6</td>\n",
       "      <td>0.732906</td>\n",
       "      <td>0.730491</td>\n",
       "      <td>0.731273</td>\n",
       "      <td>0.734170</td>\n",
       "      <td>0.735287</td>\n",
       "      <td>0.732825</td>\n",
       "      <td>0.001774</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>0.162506</td>\n",
       "      <td>0.007028</td>\n",
       "      <td>0.018390</td>\n",
       "      <td>0.001019</td>\n",
       "      <td>100</td>\n",
       "      <td>{'alpha': 100}</td>\n",
       "      <td>0.630617</td>\n",
       "      <td>0.632113</td>\n",
       "      <td>0.623321</td>\n",
       "      <td>0.624077</td>\n",
       "      <td>...</td>\n",
       "      <td>0.627051</td>\n",
       "      <td>0.003600</td>\n",
       "      <td>7</td>\n",
       "      <td>0.653381</td>\n",
       "      <td>0.650276</td>\n",
       "      <td>0.654455</td>\n",
       "      <td>0.655525</td>\n",
       "      <td>0.655394</td>\n",
       "      <td>0.653806</td>\n",
       "      <td>0.001926</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>7</td>\n",
       "      <td>0.173699</td>\n",
       "      <td>0.002925</td>\n",
       "      <td>0.017389</td>\n",
       "      <td>0.000490</td>\n",
       "      <td>1000</td>\n",
       "      <td>{'alpha': 1000}</td>\n",
       "      <td>0.501331</td>\n",
       "      <td>0.499822</td>\n",
       "      <td>0.499933</td>\n",
       "      <td>0.500288</td>\n",
       "      <td>...</td>\n",
       "      <td>0.500333</td>\n",
       "      <td>0.000533</td>\n",
       "      <td>9</td>\n",
       "      <td>0.499994</td>\n",
       "      <td>0.500388</td>\n",
       "      <td>0.500455</td>\n",
       "      <td>0.500205</td>\n",
       "      <td>0.500294</td>\n",
       "      <td>0.500267</td>\n",
       "      <td>0.000161</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>8</td>\n",
       "      <td>0.165703</td>\n",
       "      <td>0.010974</td>\n",
       "      <td>0.016390</td>\n",
       "      <td>0.003381</td>\n",
       "      <td>10000</td>\n",
       "      <td>{'alpha': 10000}</td>\n",
       "      <td>0.506504</td>\n",
       "      <td>0.502863</td>\n",
       "      <td>0.500587</td>\n",
       "      <td>0.505758</td>\n",
       "      <td>...</td>\n",
       "      <td>0.504284</td>\n",
       "      <td>0.002228</td>\n",
       "      <td>8</td>\n",
       "      <td>0.504091</td>\n",
       "      <td>0.504776</td>\n",
       "      <td>0.505382</td>\n",
       "      <td>0.503943</td>\n",
       "      <td>0.503886</td>\n",
       "      <td>0.504416</td>\n",
       "      <td>0.000578</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>9 rows × 21 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "   mean_fit_time  std_fit_time  mean_score_time  std_score_time param_alpha  \\\n",
       "0       0.163707      0.009532         0.019389        0.001357      0.0001   \n",
       "1       0.161706      0.008836         0.017992        0.000632       0.001   \n",
       "2       0.157109      0.001165         0.018191        0.000399        0.01   \n",
       "3       0.157307      0.003196         0.018191        0.000976         0.1   \n",
       "4       0.158108      0.006172         0.018190        0.000400           1   \n",
       "5       0.153309      0.000800         0.018990        0.000633          10   \n",
       "6       0.162506      0.007028         0.018390        0.001019         100   \n",
       "7       0.173699      0.002925         0.017389        0.000490        1000   \n",
       "8       0.165703      0.010974         0.016390        0.003381       10000   \n",
       "\n",
       "              params  split0_test_score  split1_test_score  split2_test_score  \\\n",
       "0  {'alpha': 0.0001}           0.685899           0.695508           0.689933   \n",
       "1   {'alpha': 0.001}           0.686202           0.695502           0.689908   \n",
       "2    {'alpha': 0.01}           0.686314           0.695491           0.689865   \n",
       "3     {'alpha': 0.1}           0.686338           0.695404           0.689781   \n",
       "4       {'alpha': 1}           0.685815           0.694649           0.689297   \n",
       "5      {'alpha': 10}           0.680188           0.687300           0.683953   \n",
       "6     {'alpha': 100}           0.630617           0.632113           0.623321   \n",
       "7    {'alpha': 1000}           0.501331           0.499822           0.499933   \n",
       "8   {'alpha': 10000}           0.506504           0.502863           0.500587   \n",
       "\n",
       "   split3_test_score  ...  mean_test_score  std_test_score  rank_test_score  \\\n",
       "0           0.683829  ...         0.687440        0.004816                2   \n",
       "1           0.683733  ...         0.687464        0.004821                1   \n",
       "2           0.683599  ...         0.687433        0.004847                3   \n",
       "3           0.683352  ...         0.687325        0.004881                4   \n",
       "4           0.682411  ...         0.686660        0.004902                5   \n",
       "5           0.674619  ...         0.680297        0.004866                6   \n",
       "6           0.624077  ...         0.627051        0.003600                7   \n",
       "7           0.500288  ...         0.500333        0.000533                9   \n",
       "8           0.505758  ...         0.504284        0.002228                8   \n",
       "\n",
       "   split0_train_score  split1_train_score  split2_train_score  \\\n",
       "0            0.744240            0.742061            0.742930   \n",
       "1            0.744198            0.742022            0.742890   \n",
       "2            0.744126            0.741957            0.742830   \n",
       "3            0.743932            0.741780            0.742658   \n",
       "4            0.742862            0.740668            0.741524   \n",
       "5            0.732906            0.730491            0.731273   \n",
       "6            0.653381            0.650276            0.654455   \n",
       "7            0.499994            0.500388            0.500455   \n",
       "8            0.504091            0.504776            0.505382   \n",
       "\n",
       "   split3_train_score  split4_train_score  mean_train_score  std_train_score  \n",
       "0            0.744787            0.746388          0.744081         0.001500  \n",
       "1            0.744765            0.746358          0.744047         0.001505  \n",
       "2            0.744725            0.746307          0.743989         0.001510  \n",
       "3            0.744588            0.746155          0.743823         0.001521  \n",
       "4            0.743596            0.745116          0.742753         0.001560  \n",
       "5            0.734170            0.735287          0.732825         0.001774  \n",
       "6            0.655525            0.655394          0.653806         0.001926  \n",
       "7            0.500205            0.500294          0.500267         0.000161  \n",
       "8            0.503943            0.503886          0.504416         0.000578  \n",
       "\n",
       "[9 rows x 21 columns]"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.naive_bayes import MultinomialNB\n",
    "from sklearn.model_selection import GridSearchCV\n",
    "# from scipy.stats import randint as sp_randint\n",
    "\n",
    "model = MultinomialNB()\n",
    "parameters = {'alpha':[0.0001, 0.001, 0.01, 0.1,1,10,100,1000,10000]}   # values changed to interval(10^[-4] to 10^4) as suggested\n",
    "clf = GridSearchCV(model, parameters, cv=5, scoring='roc_auc',return_train_score=True,n_jobs=-1)\n",
    "clf.fit(X_train_BOW, y_train)\n",
    "\n",
    "results = pd.DataFrame.from_dict(clf.cv_results_)\n",
    "results = results.sort_values(['param_alpha'])\n",
    "train_auc= results['mean_train_score']\n",
    "train_auc_std= results['std_train_score']\n",
    "cv_auc = results['mean_test_score'] \n",
    "cv_auc_std= results['std_test_score']\n",
    "K =  results['param_alpha']\n",
    "\n",
    "plt.plot(K, train_auc, label='Train AUC')\n",
    "# this code is copied from here: https://stackoverflow.com/a/48803361/4084039\n",
    "# plt.gca().fill_between(K, train_auc - train_auc_std,train_auc + train_auc_std,alpha=0.2,color='darkblue')\n",
    "\n",
    "plt.plot(K, cv_auc, label='CV AUC')\n",
    "# this code is copied from here: https://stackoverflow.com/a/48803361/4084039\n",
    "# plt.gca().fill_between(K, cv_auc - cv_auc_std,cv_auc + cv_auc_std,alpha=0.2,color='darkorange')\n",
    "\n",
    "plt.scatter(K, train_auc, label='Train AUC points')\n",
    "plt.scatter(K, cv_auc, label='CV AUC points')\n",
    "\n",
    "plt.legend()\n",
    "plt.xlabel(\"Alfa Hyperparameter\")\n",
    "plt.ylabel(\"AUC\")\n",
    "plt.title(\"Hyper parameter Vs AUC plot\")\n",
    "plt.grid()\n",
    "plt.show()\n",
    "\n",
    "results\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "MultinomialNB(alpha=0.001, class_prior=None, fit_prior=True)\n",
      "the best value of hyper-parameter alpha by Gridsearch on BOW Vectorization is :-  0.001\n"
     ]
    }
   ],
   "source": [
    "print(clf.best_estimator_)\n",
    "best_alpha_bow = clf.best_estimator_.alpha\n",
    "print('the best value of hyper-parameter alpha by Gridsearch on BOW Vectorization is :- ',best_alpha_bow)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Hyper parameter tuning on TFIDF features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAY4AAAEWCAYAAABxMXBSAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjEsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy8QZhcZAAAgAElEQVR4nOzde3xU1bn4/88zk0ACBBCwUS7KRapyM0DwUgUCeAOrgFpIWlrpUZAerUetHKGnX6HUKtr+SvXUc7wdxXpJQJSLFaXeghYVQQkg1wAihItgBCSShFye3x97zziZTJKZJJOQzPN+vfYre6+99tprZSDPrLX3XltUFWOMMSZcnsaugDHGmKbFAocxxpiIWOAwxhgTEQscxhhjImKBwxhjTEQscBhjjImIBQ5jTJMgIrtF5PLGroexwGFqEOo/q4hMFpF/NVadmgMR6S4iKiJx9VRegogcFZGRIfbNE5FFtSx3tlvPC0OkvxAiv4rIOQHbV4nI+yJyXEQOi8hKEbmuNnWJoM5pIpIXzXPEOgscpkkQR9T+vUa7/PoWHHBUtQhYAPwiKJ8XyACeq8U5BPg58A1wUy2OvxF4Gfg70BVIBu4Dro20LHNqaTL/UcypSUSmi8grQWn/LSJ/ddezReRBEflERI6JyFIR6RCQ92IR+dD9trxeRNIC9mWLyB9FZBVwAugZ4vy7RWSmiGwWkSMi8qyIJLj7ThORf7jfdI+4612rK19EfikiW9xvyLtE5NaA/Gkikici/ykih0TkgIiME5ExIrJdRL4Rkd8G5PeIyAwR2Ski+SKyMKDt77s/j4pIgYhc4h7zb+75j4jIChE5O6A8FZHbRCQXyA3xcTwH3CAirQLSrsL5f/6GW8a9IrLPbd82ERkVohyfoUBn4D+AdBFpUU3eCtyg8xfgD6r6tKoeU9VyVV2pqlOqOGa2iCwSkQVu/T4TkQuqyNtSRP4qIvvd5a9uWmu3rZ3d32uBiHQOt94mTKpqiy1VLsBu4PKgtMnAv9z1M4HvgPbudhxwCBjsbmcD+4B+QGvgFeAFd18XIB8Yg/PH7Qp3+/SAY/cAfd1y46uo3+dAN6ADsAq4393XEbgBaAUk4Xz7XRJwbKXygWuAXoAAw3ECyiA3fxpQivOtOR6YAhwGXnLL7wsUAT3d/HcCH+N8224JPAFkuvu6AwrEBdRnHLADON+tz++ADwP2K/CW287EKj6v7cCkgO1M4K/u+rnAXqBzQB16VfPZ/x+w0G1rPnB9wL7Zvs8x6BgFzgHOc9d7RPBvbTZQAtzonvMe4Avf507Av0Vgjvu7/QFwOvAhTpDyfU55jf1/pzkvjV4BW07txf3PWgAcDVhO4AYON88bwBR3/cfA5oB92cDcgO0+wEnAC9wLPB90vhXATQHHzgmjftMCtscAO6vImwIcCapbTeUvAf7DXU8DCgGvu53k/nG8KCD/p8A4d30LMCpg35nuH8Y4QgeON4CbA7Y97u/6bHdbgZE11Pd3wD/d9bbu8QPd7XNwgvrlhAjCQeW0Ar4NaMsTwNKA/bOpPnBc6q4nRPBvbTbwcVD7DwBDAz5rX+DYCYwJyHsVsDvgc7LAEcXFhqpMOMapanvfAvx70P7ngEnu+iTg+aD9ewPWv8T5NtkJOBv4iTtMdVREjgKX4fyBDXVsVYLL7wwgIq1E5AkR+VJEvsUZHmrvjvuHLF9ERovIx+6w01GcQNQpIEu+qpa564Xuz68C9hcCbdz1s4HFAW3bApThjPWHcjbwSED+b3B6Pl2qqm8IfwdGiEgXnG/uO1R1HYCq7sDpBc0GDolIVjXDOONxelfL3e0XgdEicrq7XYrzOfqJiG+7BKeHAhU/y3D426eq5UAe7ucZpDPOZ+3zZRX5TBRY4DD1YQkwQET64fQ4Xgza3y1g/SycPyxf4/yReD4wKKlqa1WdG5A/nOmbg8vf767/Bmd45iJVbQsMc9MlVPki0hJnKO3PQLIbJJcH5Y/EXmB0UPsSVHUfodu1F7g1KH+iqn4Yqr6hqOoe4APgZzgXtv8etP8lVb0MJ0gp8FAVRd2EEwD3iMhBnGG+eJwL7eAM8XUPOqYHTmDcB2xz23NDdfUNwf9ZujcrdOX7zzPQfrcNPoGfu035HWUWOEydqXNHzyKcsf5P3D9egSaJSB/3ou0cYJH7rf0F4Fpxbtn0inNLaVrgBeww3SYiXd0Lz7/FubsInKGkQpwL0B2AWTWU0wLnWsRhoFRERgNXRliXQI8Df/Rd4BaR00VkrLvvMFBOxQv+jwMzRaSvm7+diPykFud9DrgdZ7jIH8RF5FwRGekGyCKc301Z8MFub2UUzpeAFHe5ACfI+O6uehM4V0R+LiLx7u/3AZzPtlSdMaO7gf/n3nDQ1r1Z4DIRebKaug8WkevFuWvsTqAY51pGsEzgd+7vtBPOdSff7cFfAR1FpF2NvylTKxY4TH15DuhP5WEq3LT5wEEgAbgDQFX3AmNx/tgfxvmGOp3I/12+BPwT2OUu97vpfwUScXo3H+P8sauSqh5367YQOAL8FFgWYV0CPeIe/08ROe7W4SL3XCeAPwKr3KGpi1V1Mc4f5yx3aO1zYHQtzrsIOA14R1UPBKS3BObi/D4O4lxY/m3lw/k5kKOq/1TVg74FeBS3Z6mqh3CG8W7FuW7yOXAM+JWvEFVdBEwE/g2nN/AVzmeztJq6L3WPOeLW43pVLQmR735gLbAB2Ah85qahqltxAssu93drQ1j1TJwvBsbUjYicBWwFzlDVbwPSs3Euoj4dpfPuBm5R1bejUb5pOCIyGzhHVSfVlNc0LutxmDpzx6LvBrICg4Yxpnmql+kOTOxyH7j6CueulqsbuTrGmAZgQ1XGGGMiYkNVxhhjIhITQ1WdOnXS7t2717mc7777jtatW9e9Qk1ELLU3ltoK1t7mrD7b+umnn36tqqcHp8dE4OjevTtr166tcznZ2dmkpaXVvUJNRCy1N5baCtbe5qw+2yoiX4ZKt6EqY4wxEbHAYYwxJiIWOIwxxkQkJq5xGGPqrqSkhLy8PIqKihq7KhFr164dW7ZsaexqNIjatDUhIYGuXbsSHx9fc2YscBhjwpSXl0dSUhLdu3fHecFf03H8+HGSkpIauxoNItK2qir5+fnk5eXRo0ePsI6xoSpjTFiKioro2LFjkwsapnoiQseOHSPqSVrgMMaEzYJG8xTp52qBwxhjTEQscIThZ099RPcZr7Nx3zG6z3idnz31UWNXyZiYk5+fT0pKCikpKZxxxhl06dLFv33y5MmwyvjlL3/Jtm3bIj73Nddcw9ChQyukTZo0iSVLlvi3S0tLad++vX9769atjB49mt69e3P++eeTnp7OoUOHIj73qcgujtfgZ099xKqd31RIW7XzG3721Ee8OOWSRqqVMbGnY8eO5OTkADB79mzatGnDPffcUyGPqqKqeDyhvxM/++yzEZ83Pz+fjRs3kpCQwJ49ezjrrLNqPKawsJAf//jHPProo4wZMwaAd955h/z8fH7wgx9EXIdTjfU4ahAYNFYfkpDpxpjGs2PHDvr168e0adMYNGgQBw4cYOrUqaSmptK3b1/mzJnjz3vZZZeRk5Pj7x3MmDGDCy64gEsuuaTK3sCiRYsYN24cEydOZMGCBSHzBHv++ecZNmyYP2gAjBo1ivPPP79ujT1FWI8jArnfWpw1BuD3r21i8/76fWdXn85tmXVt31odu3nzZp599lkef/xxAObOnUuHDh0oLS1lxIgRjB49miFDhlQ45tixYwwfPpy5c+dy991388wzzzBjxoxKZWdmZvLggw/Srl07Jk2axPTp02usz+eff87gwYNr1ZamwP4SGmOavF69elUIDJmZmQwaNIhBgwaxZcsWtm7dWumYxMRERo92Xuk+ePBgdu/eXSnPvn372LNnDxdffDF9+vShrKzMX1aoO5Fi5a4z63EYYyJW255BtAROI56bm8sjjzzCJ598Qvv27Zk0aVLIZxRatGjhX/d6vZSWllbKs2DBAvLz8/0Pxh07doysrCxmz55Nx44dOXLkiD/vN998Q6dOnQDo27cvq1evrrf2nWqsx1GDLu0T/euBL0sMTDfGnDq+/fZbkpKSaNu2LQcOHGDFihW1LiszM5O3336b3bt3s3v3bj755BMyMzMBSEtLIysri5KSEgDmz5/PiBEjAPj5z3/OypUrefPNN/1lLV++nM2bN9ehZacOCxw1mH7VuSTGeyukJcZ7mX7VuY1UI2NMdQYNGkSfPn3o168fU6ZM4dJLL61VOTt37uTgwYOkpqb603r37k3Lli359NNPGTduHBdddBGDBw8mJSWFNWvW8OCDDwLQqlUrXnvtNebNm0fv3r3p06cPL7zwAqefXumdSE1STLxzPDU1VevyIqcl6/Yx45UNtI0rIz6hFdOvOpdxA7vUYw1PTfbym+arNu3dsmVLk70ryOaqqlmoz1dEPlXV1OC8Ue1xiMjVIrJNRHaISKXbFURknojkuMt2ETnqpqeIyEcisklENojIxIBj5ovIFwHHpUSzDQDjBnbh0nM6kdjCy6oZI2MiaBhjTFWidnFcRLzAY8AVQB6wRkSWqap/kE9V7wrI/2tgoLt5AviFquaKSGfgUxFZoapH3f3TVXVRtOoeSozcLGGMMTWKZo/jQmCHqu5S1ZNAFjC2mvwZQCaAqm5X1Vx3fT9wCGj0wcEYGNUzxpgaRfN23C7A3oDtPOCiUBlF5GygB/BuiH0XAi2AnQHJfxSR+4B3gBmqWhziuKnAVIDk5GSys7Nr1wpXfn4R5eVldS6nKSkoKIiZ9sZSW6F27W3Xrh3Hjx+PToWirKysrMnWPVK1bWtRUVHY/yaiGThCDe5U9Z09HVikqmUVChA5E3geuElVy93kmcBBnGDyJHAvMIcgqvqku5/U1FSt64XPF75cy9eFh+0CajMVS22F2l8cb6oXmO3ieM0SEhIYOHBgzRmJ7lBVHtAtYLsrsL+KvOm4w1Q+ItIWeB34nap+7EtX1QPqKAaexRkSizq7xmGMMY5oBo41QG8R6SEiLXCCw7LgTCJyLnAa8FFAWgtgMfB3VX05KP+Z7k8BxgGfR60FriXr9vHB9sMUlZRx6dx3WbJuX7RPaYwJ4eDBg6Snp9OrVy/69OnDmDFj2L59Oz169Kg0Xfqdd97Jww8/HLKcefPmkZCQwLFjx/xp8+fP5/bbb6+QLy0tDd+t/AUFBdx666306tWLvn37MmzYsGb9dHh1ohY4VLUUuB1YAWwBFqrqJhGZIyLXBWTNALK04gMlE4BhwOQQt92+KCIbgY1AJ+D+aLUBnKAx89WNFJU6I2X7jhYy89WNFjyMaWCqyvjx40lLS2Pnzp1s3ryZBx54gK+++or09HSysrL8ecvLy1m0aBETJ04MWVZmZiZDhgxh8eLFYZ//lltuoUOHDuTm5rJp0ybmz5/P119/Xed2NUVRnatKVZcDy4PS7gvanh3iuBeAF6ooc2Q9VrFGf1qxjcKSCpdeKCwp408rttnzHMY0oPfee4/4+HimTZvmT0tJcb5PtmvXjokTJzJr1iwA3n//fbp3787ZZ59dqZydO3dSUFDAn/70Jx544AEmT55c47l37tzJ6tWrefHFF/3v+ujZsyc9e/ash5Y1PTbJYQ32Hy30r2sV6cbEnDdmwMGN9VvmGf1h9Nwqd1c3VfmAAQPweDysX7+eCy64gKysLDIyMkLmzczMJCMjg6FDh7Jt2zYOHTpU48uVNm3aREpKCl6vt9p8scLmqqpB58DJDLWKdGNMo8vIyCArK4vS0lKWLl3KT37yk5D5srKySE9Px+PxcP311/Pyy85l1KqmRI+VqdIjYT2OGky/6lxmvrqRwpIyf9ywSQ5NzKumZxAtffv2ZdGiqieMyMjI4Morr2T48OEMGDAgZC9iw4YN5ObmcsUVVwBw8uRJevbsyW233VZpmnT4fqr09u3bs379esrLy6t8LW0ssd9ADcYN7MKD1/cnId75VXVpn8iD1/e36xvGNLCRI0dSXFzMU0895U9bs2YNK1euBJyXOXXs2JEZM2ZUO0w1e/Zs/zTp+/fvZ9++fXz55ZcMGTKEVatWcfDgQQDWrl1LcXEx3bp1o1evXqSmpjJr1ix89/Hk5uaydOnSKLf61GSBIwzjBnZh5Hk/oGWcxyY5NKaRiAiLFy/mrbfe8t8SO3v2bDp37uzPk5GRwdatWxk/fnzIMrKysirtGz9+PFlZWSQnJ/PII48wZswYUlJSuPPOO8nMzPT3MJ5++mkOHjzIOeecQ//+/ZkyZUqFc8cSG6oKk4R8EN4Y05A6d+7MwoULq9x/1113cdddd1W5/4svvqiU9pe//MW/PnbsWMaODT2lXtu2bSv0dmKZ9TgiYHMcGmOMBY7wiQUOY4wBCxxhE7DIYYwxWOAIm4hY3DDGGCxwhM0ujRtjjMMCR5js4VFjjHFY4IiADVUZ03jy8/NJSUkhJSWFM844gy5duvi3T548GVYZv/zlLytNvx6Oa665hqFDh1ZImzRpEkuWLPFvl5aW0r59e//21q1bGT16NL179+b8888nPT2dQ4cORXzuYFdddVWNb/h75pln/A8yRoM9xxEmwd45bkxj6tixIzk5OQDMnj2bNm3acM8991TIo6qoapXTgjz77LMRnzc/P5+NGzeSkJDAnj17OOuss2o8prCwkB//+Mc8+uijjBkzBoB33nmH/Pz8GidUrMmKFStqzPPMM88waNAgzjjjjDqdqyrW4wiTTXRmTGSWrNvHpXPfpceM16P6ArQdO3bQr18/pk2bxqBBgzhw4ABTp04lNTWVvn37MmfO92+Wvuyyy8jJyfH3DmbMmMEFF1zAJZdcUmVvYNGiRYwbN46JEyeyYMGCsOr0/PPPM2zYMH/QABg1ahTnn39+hXxvv/02I0aMYNy4cfTp04fbbrvNP6XJCy+8QP/+/enXrx+//e1v/cd07dqVo0eP+tt9880307dvX0aPHk1RURGvvPIKOTk5TJw40d8bmz59On369GHAgAHce++9Yf9uqxLVwCEiV4vINhHZISIzQuyfF/Cipu0icjRg300ikusuNwWkDxaRjW6Zj0oD/UUXbKjKmHD5XoC272ghSvRfgLZ582Zuvvlm1q1bR5cuXZg7dy5r165l/fr1vPXWW2zdurXSMceOHWP48OGsX7+eSy65hGeeeSZk2b5p2DMyMsjMzAyZJ1h1U8AHW716NX/961/ZuHEjW7ZsYenSpeTl5fG73/2O9957j3Xr1rFq1Sr+8Y9/VDp227Zt3HnnnWzatInExESWLFnCDTfcQEpKCgsWLCAnJ4cjR46wfPlyNm3axIYNG5g5c2ZY9apO1AKHiHiBx4DRQB8gQ0T6BOZR1btUNUVVU4D/Bl51j+0AzAIuwnmn+CwROc097H+BqUBvd7k6Wm2oQGyoyphwVfcCtGjo1asXQ4YM8W9nZmYyaNAgBg0axJYtW0IGjsTEREaPHg3A4MGD2b17d6U8+/btY8+ePVx88cX06dOHsrIyf1mhvrPW5nvsxRdfTPfu3fF6vaSnp/Ovf/2L1atXM3LkSDp16kR8fDw//elPef/99ysd65s3q7o2dOjQAY/Hw5QpU1i8eDGtW7eOuI7BotnjuBDYoaq7VPUkkAWEngTGkQH4wvlVwFuq+o2qHgHeAq523zfeVlU/cl81+3ec945Hnc1VZUz4qnrRWbRegBb4xzA3N5dHHnmEd999lw0bNnD11VdTVFRU6ZgWLVr4171eL6WlpZXyLFiwgPz8fHr06EH37t3Zs2eP/xW1wdOw+6ZgB2cK+E8//TSsugcHGxFBw/yW2rJlyxrbEB8fz9q1axk3bhyvvPIK11xzTVhlVyeaF8e7AHsDtvNwehCViMjZQA/g3WqO7eIueSHSQ5U5FadnQnJyMtnZ2RE3INDBg8WUa3mdy2lKCgoKYqa9sdRWqF1727VrV+PdPD5ntG3JgW+LQ6aHW0Z1iouLiY+P5/jx4xQUFFBeXu4v98CBA7Ru3RoRITc3lzfffJOhQ4dy/PhxysrK+O677/x5fT8LCwspKSmpVLcXXniBpUuX+oedduzYwYQJE/jNb37DRRddxNNPP83YsWOJj4/niSee4LLLLuP48eOMGzeOuXPn8uqrr/rf/bFixQrOPvtszjvvPH/5J06c4OOPP2bz5s107tyZzMxMbr31Vvr168c999zD7t27adeuHS+++CK//vWvOX78OKoast3FxcWUlpZSVlZGYmIiX331FcePH+f48eMUFxczfPhw+vbty5AhQ0J+BkVFRWH/m4hm4Aj1Fb2qMJoOLFJVX9+2qmPDLlNVnwSeBEhNTdW0tLRqK1uTfxxez5Zv9lHXcpqS7OzsmGlvLLUVatfeLVu2kJSUFFbee0ef738Bmk9ivJd7R58fdhnVadmyJS1btiQpKYk2bdrg8Xj85Q4dOpR+/fpxySWX0LNnTy677DL/fq/XS+vWrf15fT8TExOJj4+vULedO3dy+PBhhg8f7u8VDBw4kMTERLZv305GRgZbtmwhLS0Nj8dD7969efzxx0lKSiIpKYnXX3+du+66i3vvvZf4+HhSUlJ45JFHKpyjVatW/OhHP2LWrFls2rSJtLQ0Jk6ciIjwhz/8gWuvvRZV5dprr/W/0VBESEpKori4uEK7W7Zsidfrxev1MmXKFO644w4SExNZtmwZEydOpLi4mPLycubNmxfyM0hISGDgwIHhfQC+29fqewEuAVYEbM8EZlaRdx3wo4DtDOCJgO0n3LQzga1V5atqGTx4sNbVPQtzdOCs1+tcTlPy3nvvNXYVGkwstVW1du3dvHlzRPkXf5anP3rwHe1+7z/0Rw++o4s/y4v4nPXl22+/bbRzV+ett97SsWPH1muZtW1rqM8XWKsh/qZGs8exBugtIj2AfTi9ip8GZxKRc4HTgI8CklcADwRcEL8SJ+h8IyLHReRiYDXwC5yL6lEndnHcmIiMG9jFXnrWTEUtcKhqqYjcjhMEvMAzqrpJRObgRLFlbtYMIMuNbr5jvxGRP+AEH4A5qvqNu/4rYD6QCLzhLlEn2CSHxpj6dfnll3P55Zc3djUiFtUnx1V1ObA8KO2+oO3ZVRz7DFDpxmpVXQv0q79ahsee/zPGGIc9OW6MMSYiFjjCJPYGQGOMASxwREDs4rgxxmCBI2xi7441ptEdPHiQ9PR0evXqRZ8+fRgzZgzbt2+nR48elaZLv/POO3n44YdDljNv3jwSEhI4duyYP23+/PncfvvtFfKlpaWxdu1awHlo8tZbb6VXr1707duXYcOGsXr16jq36b777uPtt9+uNk92djYffvhhnc9VXyxwhMkmOTSmcakq48ePJy0tjZ07d7J582YeeOABvvrqK9LT0/1TgQCUl5ezaNEiJk6cGLKszMxMhgwZwuLFi8M+/y233EKHDh3Izc1l06ZNzJ8/n6+//rrO7ZozZ06Nd1ZZ4Gii7K4qYyK0YSHM6wez2zs/NyysU3Hvvfce8fHxTJs2zZ+WkpLC0KFDycjIqBA43n//fbp3787ZZ59dqZydO3dSUFDA/fffH/Zstzt37mT16tXcf//9/nd99OzZM+S8T23atOE3v/kNgwYNYtSoURw+fBiAnJwcLr74YgYMGMD48eP981xNnjyZRYsWAdC9e3dmzZrFoEGD6N+/P1u3bmX37t08/vjjzJs3j5SUFD744ANefvll+vXrxwUXXMCwYcPC/A3WHwsckbAuhzHh2bAQXrsDju0F1Pn52h11Ch7VTVU+YMAAPB4P69evByArK4uMjIyQeX3TpA8dOpRt27aF9Va+TZs2kZKSgtfrrTHvd999x6BBg/jss88YPnw4v//97wH4xS9+wUMPPcSGDRvo37+/Pz1Yp06d+Oyzz/jVr37Fn//8Z7p37860adO46667yMnJYejQocyZM4cVK1awfv16li1bFrKcaLLAESZ7ANCYCLwzB0qCZsItKXTSo8TX6ygtLWXp0qX+uZ2CZWVlkZ6ejsfj4frrr+fll18Gqp4SPdKp0j0ej3+IbNKkSfzrX//i2LFjHD16lOHDhwNw0003hZwmHeD6668Hqp4mHeDSSy9l8uTJPPXUU5SVlYXME0326tgw2e24xkTgWF5k6WHo27evf0gnlIyMDK688kqGDx/OgAEDQr6idcOGDeTm5vpnrD158iQ9e/bktttuqzRNOnw/VXr79u1Zv3495eXlVb6WtiqRBh7fVOlVTZMO8Pjjj7N69Wpef/11UlJSyMnJoWPHjhGdpy6sxxEmu8RhTATadY0sPQwjR46kuLiYp556yp+2Zs0aVq5cCTgvc+rYsSMzZsyodphq9uzZ7N69m927d7N//3727dvHl19+yZAhQ1i1ahUHDx4EYO3atRQXF9OtWzd69epFamoqs2bN8r8rIzc3l6VLl1Y6h+/CPMBLL73EZZddRrt27TjttNP44IMPAOfVsr7eRziSkpIqTIW+c+dOLrroIubMmUOnTp3Yu3dvNUfXPwscYbJ3jhsTgVH3QXxixbT4RCe9lkSExYsX89Zbb/lviZ09ezadO3f258nIyGDr1q2MHz8+ZBlZWVmV9o0fP56srCySk5N55JFHGDNmDCkpKdx5551kZmb6exhPP/00Bw8e9L91b8qUKRXO7dO6dWs2bdrE4MGDeffdd7nvPqfNzz33HNOnT2fAgAHk5OT408Nx7bXXsnjxYv/F8enTp/vfRz5s2DAuuOCCsMuqD6Ix8FRbamqq+u7Frq3Zyzax8JPdbL6/7m/Paipi6R0VsdRWqP37OM4///zwD9iw0LmmcSzP6WmMug8GTIisovXk+PHj9fIekHC0adOGgoKCBjlXKLVta6jPV0Q+VdXU4Lx2jSMCzT/EGlOPBkxotEBhosuGqsJkI1XGmHA0Zm+joVjgCJPYXFXGEAtD27Eo0s/VAkeYrMdhYl1CQgL5+fkWPJoZVSU/P5+EhISwj4nqNQ4RuRp4BOcNgE+r6twQeSYAs3EuIaxX1Z+KyAhgXkC284B0VV0iIvOB4YBvdrLJqpoTvVa49Yz2CYw5xXXt2pW8vDz/FBpNSVFRUUR/GJuy2rQ1ISGBrl3Dv1mZdOwAACAASURBVFU6aoFDRLzAY8AVQB6wRkSWqermgDy9gZnApap6RER+AKCq7wEpbp4OwA7gnwHFT1fVqp8EihL7nmViWXx8PD169GjsatRKdnY2AwcObOxqNIiGaGs0h6ouBHao6i5VPQlkAWOD8kwBHlPVIwCqGmrSmBuBN1T1RBTrWiN7ctwYYxzRHKrqAgQ+zpgHXBSU54cAIrIKZzhrtqq+GZQnHfhLUNofReQ+4B1ghqoWB59cRKYCUwGSk5PJzs6uZTMce/eeRFXrXE5TUlBQEDPtjaW2grW3OWuItkYzcIS6LBD8pT0O6A2kAV2BD0Skn6oeBRCRM4H+wIqAY2YCB4EWwJPAvUClmdNU9Ul3P6mpqVqnh7ueu46P9nYGvYa07LHQYzjc1PAzUja0WHooLpbaCtbe5qwh2hrNoao8oFvAdldgf4g8S1W1RFW/ALbhBBKfCcBiVS3xJajqAXUUA8/iDIlFz3PXwRcriaOERIqctC9WOunGGBODohk41gC9RaSHiLTAGXIK/pq+BBgBICKdcIaudgXszwAqvGnF7YUgzuRR44DPo1J7ny+cCdSu967ihfg/Vko3xphYE7WhKlUtFZHbcYaZvMAzqrpJROYAa1V1mbvvShHZDJTh3C2VDyAi3XF6LMF/oV8UkdNxhsJygGk0AMVuyTXGGIjycxyquhxYHpR2X8C6Ane7S/Cxu3EusAenj6z3ihpjjAmbPTleE6/zUhWBiu8AdNONMSbWWOCoydi/gXgq3g4mHifdGGNikAWOmgyYAOOfAE+8s92um7Nt00UbY2KUBY5wDJiAnHa2M1R11+cWNIwxMc0CR5gUsbuqjDEGCxzh2bAQjux21uf1c7aNMSZGWeCoyYaF8NodUF7iDFUd2+tsW/AwxsQoCxw1eWcOlBTi3JCrzlsASwqddGOMiUEWOGpyLM+/6oQOqZRujDGxxAJHTdp9/1asCg8Atgv/bVnGGNOcWOCoyaj7ID7RnatKndARn+ikG2NMDLLAUZMBE+DaR8HbwhmqaneWs23PchhjYpQFjnAMmAAde+OhHP2PHAsaxpiYZoEjTOpeE1d78bgxJsZZ4Aibx73GYZHDGBPboho4RORqEdkmIjtEZEYVeSaIyGYR2SQiLwWkl4lIjrssC0jvISKrRSRXRBa4bxeMPgm6q8oYY2JU1AKHiHiBx4DRQB8gQ0T6BOXpDcwELlXVvsCdAbsLVTXFXQJf8P0QME9VewNHgJuj1YaKnLmqbKjKGBProtnjuBDYoaq7VPUkkAWMDcozBXhMVY8AqOqh6gp03zM+EljkJj2H897xqFPxWI/DGGOI7qtjuwB7A7bzgIuC8vwQQERW4byXfLaqvunuSxCRtUApMFdVlwAdgaOqWhpQZqXXy7plTgWmAiQnJ5OdnV2nxiQdL6AVysr336elNzbmyS0oKKjz762piKW2grW3OWuItkYzcIT66xr8lT0O6A2kAV2BD0Skn6oeBc5S1f0i0hN4V0Q2At+GUaaTqPok8CRAamqqpqWl1aoRPtu2PoV8pwwdOpRWLaL6qvZTRnZ2NnX9vTUVsdRWsPY2Zw3R1mgOVeUB3QK2uwL7Q+RZqqolqvoFsA0nkKCq+92fu4BsYCDwNdBeROKqKTNKnDho1ziMMbEumoFjDdDbvQuqBZAOLAvKswQYASAinXCGrnaJyGki0jIg/VJgs6oq8B5wo3v8TcDSKLahArvGYYwxUQwc7nWI24EVwBZgoapuEpE5IuK7S2oFkC8im3ECwnRVzQfOB9aKyHo3fa6qbnaPuRe4W0R24Fzz+L9otaECke/nqjLGmBgW1cF6VV0OLA9Kuy9gXYG73SUwz4dA/yrK3IVzx1aDUjzu7bgWOowxsa3KHoeIXCUiN4ZI/5mIXBHdap2CRJy5qhq7HsYY08iqG6r6PbAyRPo7QAy+/s4eADTGGKg+cLRS1cPBiap6EGgdvSqdotxrHMYYE+uqCxwJAbe9+olIPJAYvSqdmtR9cazFDmNMrKsucLwKPCUi/t6Fu/64uy/GuENVFjmMMTGuusDxO+Ar4EsR+VREPgN2A4fdfbHFdzuuxQ1jTIyr8nZc9zmMGSLye+AcN3mHqhY2SM1ONSJ4rL9hjDHV3o57vYhcjzMtem+c4JEqIkkNVblTxZJ1+8jZewxBGfPIByxZt6+xq2SMMY2mugcArw2R1gEYICI3q+q7UarTKWXJun3MfHUj/6mKeJWD3xYx89WNAIwbGHJiXmOMadaqG6r6Zah0ETkbWEjlKdKbpT+t2EZhSRnEfT/db2FJGX9asc0ChzEmJkU8V5WqfgnER6Eup6T9R51LOv7bcYPSjTEm1kQcOETkPKA4CnU5JXVu7zyyUu7ejhucbowxsabKoSoReY3Kj7t1AM4EJkWzUqeS6Vedy8xXN6LqzFUFkBjvZfpV5zZyzYwxpnFUd3H8z0HbCnyDEzwmAR9Fq1KnEt91jG+XeZFyOL1NS/7rmvPt+oYxJmZVd3HcP8GhiKQAPwUmAF8Ar0S/aqeOcQO7sG97MrJZ+dtPB3JRz46NXSVjjGk01T3H8UMRuU9EtgB/A/YCoqojVPVv4RQuIleLyDYR2SEiM6rIM0FENovIJhF5yU1LEZGP3LQNIjIxIP98EflCRHLcJSWiFteS1+tFUOcOK2OMiWHVDVVtBT4ArlXVHQAicle4BYuIF3gMuALn3eJrRGRZwJv8EJHewEzgUlU9IiI/cHedAH6hqrki0hn4VERWqOpRd/90VV0Ubl3qQ5zXuTheVFLekKc1xphTTnV3Vd0AHATeE5GnRGQUVLixqCYX4kxRsktVTwJZwNigPFOAx1T1CICqHnJ/blfVXHd9P3AIOD2Cc9c7rzcOUIpLrcdhjIltVQYOVV2sqhOB84Bs4C4gWUT+V0SuDKPsLjjDWz55blqgHwI/FJFVIvKxiFwdXIiIXAi0AHYGJP/RHcKaJyItw6hLncV5PXhQimyoyhgT42p857iqfge8CLwoIh2AnwAzgH/WcGio3knw7b1xOPNgpQFdgQ9EpJ9vSEpEzgSeB25SVd8Y0UycnlAL4EngXkK8kVBEpgJTAZKTk8nOzq6hutXrsn8/PVA2bN5G8ne76lRWU1FQUFDn31tTEUttBWtvc9YQba0xcARS1W+AJ9ylJnlAt4DtrsD+EHk+VtUS4AsR2YYTSNaISFvgdeB3qvpxQB0OuKvFIvIscE8VdX0SJ7CQmpqqaWlpYVS5aiUnVyJ5SrfuPUkb3qtOZTUV2dnZ1PX31lTEUlvB2tucNURbI35yPAJrgN4i0kNEWgDpwLKgPEuAEQAi0gln6GqXm38x8HdVfTnwALcXgogIMA74PIpt8IvzetyL4zZUZYyJbRH1OCKhqqUicjuwAvACz6jqJhGZA6xV1WXuvitFZDNQhnO3VL6ITAKGAR1FZLJb5GRVzcEZMjsdZygsB5gWrTYEEvEgonZXlTEm5kUtcACo6nJgeVDafQHrCtztLoF5XgBeqKLMkfVf0zCI0zkrOlnaKKc3xphTRTSHqpoXca71F5dY4DDGxDYLHGFzAkeRBQ5jTIyzwBEu8QUOuzhujIltFjjC5huqssBhjIltFjjC5T7OaENVxphYZ4EjbG6Po9RuxzXGxDYLHOFyb8c9ebKkkStijDGNywJHuHy349rsuMaYGGeBI2z2HIcxxoAFjvCJ3VVljDFggSMCNlRljDFggSN8bo+jpLSM8vLg14oYY0zssMARLveuKg9qt+QaY2KaBY6w+V5oqBTadQ5jTAyzwBEud6jKXuZkjIl1FjjC5gscaoHDGBPToho4RORqEdkmIjtEZEYVeSaIyGYR2SQiLwWk3yQiue5yU0D6YBHZ6Jb5qPsK2ejasBCyH3TOj1K0+c2on9IYY05VUQscIuIFHgNGA32ADBHpE5SnNzATuFRV+wJ3uukdgFnARcCFwCwROc097H+BqUBvd7k6Wm0AnKDx2h1QdNSpM1C4cp6TbowxMSiaPY4LgR2quktVTwJZwNigPFOAx1T1CICqHnLTrwLeUtVv3H1vAVeLyJlAW1X9yH3t7N+BcVFsA7wzB0oK/ZselOKScifdGGNiUDTfOd4F2BuwnYfTgwj0QwARWQV4gdmq+mYVx3Zxl7wQ6ZWIyFScngnJyclkZ2fXrhVn3AJnwBnH1nHewSUkyEk+6fpLTrY9AbUts4koKCio/e+tiYmltoK1tzlriLZGM3CEuvYQ/ORcHM5wUxrQFfhARPpVc2w4ZTqJqk8CTwKkpqZqWlpaWJWuZN7tcOz7GNaGQn6Y9yppHQ5Axue1K7OJyM7Opta/tyYmltoK1t7mrCHaGs2hqjygW8B2V2B/iDxLVbVEVb8AtuEEkqqOzXPXqyuzfo26D+IT/ZtJnKDQ28ZJN8aYGBTNwLEG6C0iPUSkBZAOLAvKswQYASAinXCGrnYBK4ArReQ096L4lcAKVT0AHBeRi927qX4BLI1iG2DABLj2UWj9AwDaSCFF/X/upBtjTAyKWuBQ1VLgdpwgsAVYqKqbRGSOiFznZlsB5IvIZuA9YLqq5qvqN8AfcILPGmCOmwbwK+BpYAewE3gjWm3wGzABJr8OQBKFFCUPjPopjTHmVBXNaxyo6nJgeVDafQHrCtztLsHHPgM8EyJ9LdCv3itbk5ZJgNvjsAcAjTExzJ4cD1dCWwDaygkLHMaYmGaBI1zxrVA8tPcUWeAwxsQ0CxzhEqE0LpF2niKKSmxadWNM7LLAEYEyb2vaeopsWnVjTEyzwBGB0rhE2tk1DmNMjLPAEYEybyv3riobqjLGxC4LHBEojWtFawopLrUehzEmdlngiIAvcBSetMBhjIldFjgiUOZtRevy7yiyHocxJoZZ4IhAaVwiiXrCrnEYY2KaBY4IlHlb0UKLOVlc3NhVMcaYRmOBIwKlca0AiCstaOSaGGNM47HAEYEyrxM44ku/a+SaGGNM47HAEQFfjyPeehzGmBhmgSMCZV7nTYCJ5ScoKbML5MaY2BTVwCEiV4vINhHZISIzQuyfLCKHRSTHXW5x00cEpOWISJGIjHP3zReRLwL2pUSzDYFK41oD9k4OY0xsi9qLnETECzwGXIHzrvA1IrJMVTcHZV2gqrcHJqjqe0CKW04HnLf9/TMgy3RVXRStulelNM7pcSTh3JKblNDQNTDGmMYXzR7HhcAOVd2lqieBLGBsLcq5EXhDVU/Ua+1qwXdxPMl6HMaYGBbNwNEF2BuwneemBbtBRDaIyCIR6RZifzqQGZT2R/eYeSLSsp7qWyP/UBUWOIwxsSua7xyXEGkatP0akKmqxSIyDXgOGOkvQORMoD+wIuCYmcBBoAXwJHAvMKfSyUWmAlMBkpOTyc7OrnVDfApOnKQcD22kkFUff8K+dt46l3kqKygoqJffW1MQS20Fa29z1hBtjWbgyAMCexBdgf2BGVQ1P2DzKeChoDImAItVtSTgmAPuarGIPAvcE+rkqvokTmAhNTVV09LSatGEirKzsylrkURS6Qn6XjCQId071LnMU1l2djb18XtrCmKprWDtbc4aoq3RHKpaA/QWkR4i0gJnyGlZYAa3R+FzHbAlqIwMgoapfMeIiADjgM/rud7VKo9vQ5LYDLnGmNgVtR6HqpaKyO04w0xe4BlV3SQic4C1qroMuENErgNKgW+Ayb7jRaQ7To9lZVDRL4rI6ThDYTnAtGi1IZTylm3tGocxJqZFc6gKVV0OLA9Kuy9gfSbONYtQx+4mxMV0VR1ZOXcDaplEG74jv9QeADTGxCZ7cjxSLZNIsveOG2NimAWOCHkSbKjKGBPbLHBEyJPYzh4ANMbENAscEfImJPmnHDHGmFhkgSNCnoR2JEgJJ4uLGrsqxhjTKCxwRKplEgDlxccbuSLGGNM4LHBEKqGt89MChzEmRlngiJTb4/AUf9vIFTHGmMZhgSNC/9pbDMAnW7/k0rnvsmTdvkaukTHGNCwLHBE4WljCIx8cBJy3AO47WsjMVzda8DDGxBQLHBH46lgRX5c6r/1rQyEAhSVl/GnFtsasljHGNCgLHBE4WVZOgTqvj20r37+QcP/RwsaqkjHGNDgLHBFo4fVwHCdw+HocAEkJcRSX2pPkxpjYYIEjAsntEpD4RErUSxtxAodH4NuiUi7/y0pe33AA1eCXHBpjTPNigSMC7RPjefD6AZyQRJIopEv7RP4yIYXnb76Q1i3iuO2lz7jx8Y9Yt+dIY1fVGGOixgJHhMZ5V9FOCrkp7p+sankH47yrGNr7dF6/Yyhzr+/Pl/knGP8/H3JH5jryjpyouUBjjGlioho4RORqEdkmIjtEZEaI/ZNF5LCI5LjLLQH7ygLSlwWk9xCR1SKSKyIL3NfSNozCI/DaHaDu9Yxje53tDQvxeoT0C88ie3oat484hxWbDjLy/1vJw29u5XhRSfXlGmNMExK1wCEiXuAxYDTQB8gQkT4hsi5Q1RR3eTogvTAg/bqA9IeAearaGzgC3BytNlRy/ACUBN1BVVII78zxb7ZpGcc9V53Le/ekcU3/M/mf7J2M+HM2L67+ktIym1HXGNP0RbPHcSGwQ1V3qepJIAsYW5cCRUSAkcAiN+k5YFydahmJspOh04/thaN7KiR1bp/IvIkpLLv9Unp2asN/Lf6cMY9+wMrthxugosYYEz0SrbuARORG4GpVvcXd/jlwkareHpBnMvAgcBjYDtylqnvdfaVADlAKzFXVJSLSCfhYVc9x83QD3lDVfiHOPxWYCpCcnDw4Kyurzm0qOHKYNkXOU+JxZSc4/fgmkr/dQPtCJ2gcbdePr5KHc/j0H1Ea38Z/nKry6VdlLNx+kkMnlH6dvKSf24KuSaf2JaaCggLatGlTc8ZmIJbaCtbe5qw+2zpixIhPVTU1OD2ageMnwFVBgeNCVf11QJ6OQIGqFovINGCCqo5093VW1f0i0hN4FxgFfAt8FBQ4lqtq/+rqkpqaqmvXrq1zm7LfWEzaZ7+qOFwVnwgj/x+cPAEbsiB/B3hbwg+vggvS4ZwrIM65DHOytJy/f7SbR9/JpaC4lIlDzuLuK37I6Ukt61y3aMjOziYtLa2xq9EgYqmtYO1tzuqzrSISMnDE1UvpoeUB3QK2uwL7AzOoan7A5lM41y98+/a7P3eJSDYwEHgFaC8icapaGqrMqEo8Da591LmmcSwP2nWFUffBgAnO/mH3wP51sGEBbFwEW5Y5x/S9HgZMpEW3C7llaE9uGNSVR9/N5fmPvmRZzj7+fcQ53HxZDxLivQ3WFGOMqa1oBo41QG8R6QHsA9KBnwZmEJEzVfWAu3kdsMVNPw044fZEOgGXAg+rqorIe8CNONdMbgKWRrENlQ2Y8H2gCCYCXQY5y5X3w65sWJ8FOS/B2v+D07rDgImc1n8Cs67ty88vPpsH39jKn1Zs46XVe/jPq8/l2gGd8XikIVtkjDERidogu9sjuB1YgRMQFqrqJhGZIyK+u6TuEJFNIrIeuAOY7KafD6x109/Ducax2d13L3C3iOwAOgL/F6021Ik3HnpfATf+H0zPhXGPO4Fj5cPwt8Hw1Eh67nqJp27oTuaUi2nfKp7/yMph/P9+yNrd3zR27Y0xpkrR7HGgqsuB5UFp9wWszwRmhjjuQyDkdQtV3YVzx1bT0TIJUjKc5dv98PkrsH4BvDEdVszkknMu57URE1hSeAEPvb2bGx//iDH9z2DQWafx7Krd7D9aSOf2iUy/6lzGDezS2K0xxsS4qAYOE0LbzvCjXzvLV5uc6yEbXsaz/U2ub5HEdef9mKXlw/ivdfDmxv1M8LzHT+JXcqQgiaxXruDA0Ylc0fcM4jyC113iPILH/else/B4cH4KOHcxG2NM/bDA0ZiS+8IVc2DULPhyFaxfQNzmpdxwMpNR8W34zNOLh0vTySoZ5T/k7RXbeCjC938EBphK6yJ4vU6w8W97hDivUHC8kL9t+dC/7RHfcR68bmAKFbgqnsPJ6/V4Qp8/MPiJcx5fXv/5vE69/IHRXxdPiLpVLD8wiPr2WTA1pm4scJwKPF7oMcxZrvkzbHuDli/fykjvekZ616MKiqAI5Qi0aIXioVy8qHgqrJf71vGtuz/xUo5QJl5320NZ8M8yD2UIZXgoUw+FZSV4jydQph5K3XxlKpSqx01z1ktVKmyXuHlKy4USFYrdtLLAc2mI8wesl1aRr6r1ynmlyvIVT6XApWWlJPzrLX+wqTqoVQ5yXgkIjEFBrrogWrGXWFUPsnLe74P999vf1+X74Br8RcCf1+OhsFQpPFlW4XymaVuybh9/WrGN9G7H+a+570Z1aNsCx6kmPhH6XU/LRf/mTxIBQQHFA8igSc58WeVlAT/Lg7arSy8P2C6tIr2MwvLvSIxvAeUBebQMKP9+3Z8eYjqVU/j5xnK8lIvHDbZeygTQOMrLPJSX+wKtG3D8wTcoWOn3wagM8QcuX4Atw+MEWUIF24A87rnK3IB5IjDo1RBgy4LqUnG/VBtgn3xn/ffBVDyoeEG8iMeDeryIeBFPHOLxgseLuEvgusfd9nq9AYHSEzrIhQxqnko9xlA92CqDaJU92IpBeu/xcrZ/dbyKLwKekMG4KQXTJev28a/F/8MCstghv2b8if/mr4vTgX+PSvCwwHGKKko8g1aFByqlFyaeSavRcxukDqsjeZBINcLgFU5Qi1J6eSkeLcMTkJa390u6dj4zjDLCbUtpjXXTGuosoYJxQ1KgzF3C4AtM5XiddakcsMoRSt39Zf4gJgHrwcFQqu1tFofRM/X1QN9c/VxkwRgPBARUPB7wxDlf3zy+NC8i7rYvqMr3AdYJsnEBQTcOj9dJ9/iCr9fJ4/F68Xji3EAch9cbesjWHyi9Hn/vdtk/FnNt+UbWyTlsPdaGwfIdc3iSh1+PY9zA39f7Pw0LHKeoVqPnULr018SVFfnTSr0JtBo9p5qjGpGI/z9SU7QjO5uuDfxkcY3fZ8MOxqURBrUy1ues44L+/eo10Hu1DG+tAvr35Wp5GVpe6gTV8rKKP93ecXUBV9zer/i3v9/nhK06zpRRHvQzispVQgSz0L3Ii9RDmcdZ/+GBeeSTRHfPIW45+QJggSN2DJjgfDgBT6nHBT6lbpq/KAbjI3uA3mn1Xm5dCWEE1FrwT8OhWg/BshbptSjD1ytWdykvL0XLylAtQ8vKKHeDLOVllG9eBu510G9a9ST5hHMDTWdPfrW/l9qywHEqq+4pdWNM5ETAG0dT+tPnC6bVXTI88dB5/qHtT8+awNnbZgHukHcU6nQKX740xhgTjlaj51DqTaiQFs2h7aYTdo0xxoQWOLQN0K5bVIe2LXAYY0xz4Bvazs6GjM+jeiobqjLGGBMRCxzGGGMiYoHDGGNMRCxwGGOMiUhUA4eIXC0i20Rkh4jMCLF/sogcFpEcd/G9nzxFRD5yX/K0QUQmBhwzX0S+CDgmJZptMMYYU1HU7qoSES/wGHAFzvvH14jIsoA3+fksUNXbg9JOAL9Q1VwR6Qx8KiIrVPWou3+6qi6KVt2NMcZULZo9jguBHaq6S1VP4rwjfGw4B6rqdlXNddf3A4eA06NWU2OMMWET1TpO+lVVwSI3Alerqm/46efARYG9CxGZDDwIHAa2A3ep6t6gci4EngP6qmq5iMwHLgGKgXeAGapaHOL8U4GpAMnJyYOzsrLq3KaCggLatGlT53Kailhqbyy1Fay9zVl9tnXEiBGfqmpqcHo0A8dPgKuCAseFqvrrgDwdgQJVLRaRacAEVR0ZsP9MIBu4SVU/Dkg7CLQAngR2qmq1z9WLyGHgy3poVifg63oop6mIpfbGUlvB2tuc1Wdbz1bVSqM90XxyPA/oFrDdFdgfmEFVA6dufAp4yLchIm2B14Hf+YKGe4zvJRXFIvIscE9NFQnV8NoQkbWhom9zFUvtjaW2grW3OWuItkbzGscaoLeI9BCRFkA6sCwwg9t78LkO2OKmtwAWA39X1ZdDHSPOS6PHAdF9tt4YY0wFUetxqGqpiNwOrAC8wDOquklE5gBrVXUZcIeIXAeUAt8Ak93DJwDDgI7udRCAyaqaA7woIqfjzDScA0yLVhuMMcZUFrVrHM2RiExV1Scbux4NJZbaG0ttBWtvc9YQbbXAYYwxJiI25YgxxpiIWOAwxhgTEQscYapp3q2mQES6ich7IrLFnQfsP9z0DiLylojkuj9Pc9NFRB5127xBRAYFlHWTmz9XRG5qrDbVRES8IrJORP7hbvcQkdVuvRe4d/AhIi3d7R3u/u4BZcx007eJyFWN05KaiUh7EVkkIlvdz/iSZv7Z3uX+O/5cRDJFJKE5fb4i8oyIHBKRzwPS6u3zFJHBIrLRPeZR907V8KiqLTUsOHeF7QR64jx4uB7o09j1qkU7zgQGuetJOE/r9wEexnkCH2AG8JC7PgZ4A+cOtouB1W56B2CX+/M0d/20xm5fFW2+G3gJ+Ie7vRBId9cfB37lrv878Li7no4zhxru72c90BLo4f478DZ2u6po63PALe56C6B9c/1sgS7AF0BiwOc6uTl9vjh3lg4CPg9Iq7fPE/gEZxYOcY8dHXbdGvuX0xQW95e7ImB7JjCzsetVD+1aijMJ5TbgTDftTGCbu/4EkBGQf5u7PwN4IiC9Qr5TZcF56PQdYCTwD/c/yNdAXPDninPb+CXuepybT4I/68B8p9ICtHX/kEpQenP9bLsAe90/iHHu53tVc/t8ge5BgaNePk9339aA9Ar5alpsqCo8vn+kPnluWpPldtUHAquBZHWfyHd//sDNVlW7m8rv46/AfwLl7nZH4KiqlrrbgfX2t8ndf8zN31Ta2hNnzrdn3aG5p0WkNc30s1XVfcCfgT3AAZzP61Oa7+frU1+fZxd3PTg9LBY4whNq7K/J3scsIm2AV4A7VfXb6rKGSNNq0k8ZIvJj4JCqfhqYHCKr1rDvlG+rTqBiQAAABilJREFUKw5nWON/VXUg8B3OUEZVmnR73bH9sTjDS52B1sDoEFmby+dbk0jbV6d2W+AIT43zbjUVIhKPEzReVNVX3eSv5PupXM7EmcYeqm53U/h9XApcJyK7cab0H4nTA2kvIr4ZEwLr7W+Tu78dzmwGTaGt4NQzT1VXu9uLcAJJc/xsAS4HvlDVw6paArwK/Ijm+/n61NfnmeeuB6eHxQJHeGqcd6sp+P/bO7cQraoojv/+mjhZYTUYPWSNhiY0mQ8SXmEg8CkiwSiz1BKtoAwhgrIeIokk0BexArXpQpIpSfSgWaaYIWXiBdPxFuSD9RASankpVw9rfXn8HGc8OjV+n+sHmzl7nX3fw7f2PnvvtWPXxGJgl5nNK7z6DKjstpiCr31U5JNjx8YI4PeYHq8Gxkm6IUZ+40J22WBmL5rZLWbWhPfXWjObBHwNTIhg1XWttMGECG8hfzh25QwABuGLipcVZvYLcFDSHSG6F/iROuzb4GdghKQ+8X9dqW9d9m+BLunPeHdE0ohov8mFtDqnuxd/asXhuxb24LsuZnd3eS6yDmPw6eh23M7X1qhXI76IvDf+3hjhhd/iuB/YAQwvpPUEsC/c491dt07q3cKZXVUD8R+GfcAnQO+QN4R/X7wfWIg/O9qgjRI7T7qhnsOAzdG/K/FdNHXbt8CrwG7c0OkH+M6ouulfYCm+fnMKnyFM68r+BIZH2+0HFlC1saIjlyZHkiRJklLkp6okSZKkFKk4kiRJklKk4kiSJElKkYojSZIkKUUqjiRJkqQUqTiSmkTSeEkmaUhB1lRlSXRpWAqddYFpTpW0oEq2TtLwrit57RDt+Uh3lyO5/EjFkdQqE4Fv8MN95yDpZmCUmQ01s/n/a8lKUDjlfLHxe3ZVWdqhCSilOP7j8iSXCak4kpojbG2Nxg9Etas4gC+AmyRtlTRW0nRJ30vaJmmFpD4l85wmaX7BP13SvBiV75b0XsxullfSjvsO1kv6QdLqgqmIdZJel7QeeE5Sq6S3JW2QtCfsbFVG/BskbQk3KuQt8ntVPsIPeyFpZeSzU9KMQjmPSpob776UdE/kf0DS/RGmp6Q3o322S3oyor8BjI02nHW+cO2VJ6lzuvt0ZLp0ZR3wKLA4nr/lzB0jTYQJas41R91YeJ4DPNtOulNxC7NbC+4ofsL2GvyEba9CvndFPgaMDvkS4HmgV4TpF/KHgCXxvA5YWMi3FViFD+QG4aeEG4A+QEOEGQRsjucW3IjhgEIalRPEV+OngRvDb8RpaOBTXKH2Au4GtoZ8BvByPPfGT58PoHDi/gLCnVWedPXtLmmanCTdxETcYCG4AcOJwJZO4jRLmoNfbnQt57e/9LGZPVPxSFoHYGbHJK0F7pO0C1cgO+Tm6Q+a2caI8iEwE1cEzcAaNwVET9x8xL/5VOW7zMxOA3slHQCG4PdrLJA0DPgbGFwI/52Z/VTwz5Q0Pp7744rmN+BklAV8NnDCzE5J2oErPXD7RUMlVWw89Y34J6vK2FG46vIkdUwqjqSmkNSIW7ptlmT4D7JJeqGTqK3AA2a2TdJUfJRclkXAS7h9pHcL8mq7PRWz1TvNbOR50jrWTpxq/yzgV3x20AM43l58SS24tdiRZvZHKLuGeH3KzCppnwZOAJjZ6cL6ivAZ2FnKNNI9S9RBuOr6JHVMrnEktcYE4H0zu83MmsysPz4yH9NJvOuAQ3Kz8pMuJmNzk+X98QXjpYVXt0qqKIjKon0b0K8il9RL0p0dJP+gpB6SbscN9bXhI/pDMRN5DFeS7dEXOBxKYwh+dWgZVgNPR9sgabD8EqgjeLt1Fi65wkjFkdQaE/Fv9UVW0Pnun1fw2w7X4DOGi2UZsNHMDhdku4ApkrbjV5m+ZWYncSU3V9I2fL1kVAfptgHr8bufnzKz48DCSHcT/pnqfKP6VcBVkf9rwKaSdVqEmyTfEtuZ38G/RmwH/ooNBbM6CJdcYaR13CQpgaTPgflm9lX4m/AF5OZLSLM10ljeFWVMkv+anHEkyQUg6XpJe4A/K0ojSa5UcsaRJEmSlCJnHEmSJEkpUnEkSZIkpUjFkSRJkpQiFUeSJElSilQcSZIkSSn+AXY9bO/lyyg9AAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "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>mean_fit_time</th>\n",
       "      <th>std_fit_time</th>\n",
       "      <th>mean_score_time</th>\n",
       "      <th>std_score_time</th>\n",
       "      <th>param_alpha</th>\n",
       "      <th>params</th>\n",
       "      <th>split0_test_score</th>\n",
       "      <th>split1_test_score</th>\n",
       "      <th>split2_test_score</th>\n",
       "      <th>split3_test_score</th>\n",
       "      <th>...</th>\n",
       "      <th>mean_test_score</th>\n",
       "      <th>std_test_score</th>\n",
       "      <th>rank_test_score</th>\n",
       "      <th>split0_train_score</th>\n",
       "      <th>split1_train_score</th>\n",
       "      <th>split2_train_score</th>\n",
       "      <th>split3_train_score</th>\n",
       "      <th>split4_train_score</th>\n",
       "      <th>mean_train_score</th>\n",
       "      <th>std_train_score</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.161874</td>\n",
       "      <td>0.013547</td>\n",
       "      <td>0.019491</td>\n",
       "      <td>2.794829e-03</td>\n",
       "      <td>0.0001</td>\n",
       "      <td>{'alpha': 0.0001}</td>\n",
       "      <td>0.661381</td>\n",
       "      <td>0.680464</td>\n",
       "      <td>0.673421</td>\n",
       "      <td>0.670404</td>\n",
       "      <td>...</td>\n",
       "      <td>0.670958</td>\n",
       "      <td>0.006194</td>\n",
       "      <td>1</td>\n",
       "      <td>0.732870</td>\n",
       "      <td>0.731353</td>\n",
       "      <td>0.734320</td>\n",
       "      <td>0.734554</td>\n",
       "      <td>0.736486</td>\n",
       "      <td>0.733917</td>\n",
       "      <td>0.001723</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.179296</td>\n",
       "      <td>0.010377</td>\n",
       "      <td>0.017990</td>\n",
       "      <td>5.560829e-07</td>\n",
       "      <td>0.001</td>\n",
       "      <td>{'alpha': 0.001}</td>\n",
       "      <td>0.661365</td>\n",
       "      <td>0.680452</td>\n",
       "      <td>0.673414</td>\n",
       "      <td>0.670395</td>\n",
       "      <td>...</td>\n",
       "      <td>0.670947</td>\n",
       "      <td>0.006196</td>\n",
       "      <td>2</td>\n",
       "      <td>0.732856</td>\n",
       "      <td>0.731339</td>\n",
       "      <td>0.734305</td>\n",
       "      <td>0.734540</td>\n",
       "      <td>0.736471</td>\n",
       "      <td>0.733902</td>\n",
       "      <td>0.001722</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.159305</td>\n",
       "      <td>0.003260</td>\n",
       "      <td>0.017592</td>\n",
       "      <td>4.869431e-04</td>\n",
       "      <td>0.01</td>\n",
       "      <td>{'alpha': 0.01}</td>\n",
       "      <td>0.661253</td>\n",
       "      <td>0.680336</td>\n",
       "      <td>0.673355</td>\n",
       "      <td>0.670276</td>\n",
       "      <td>...</td>\n",
       "      <td>0.670852</td>\n",
       "      <td>0.006196</td>\n",
       "      <td>3</td>\n",
       "      <td>0.732716</td>\n",
       "      <td>0.731196</td>\n",
       "      <td>0.734158</td>\n",
       "      <td>0.734407</td>\n",
       "      <td>0.736332</td>\n",
       "      <td>0.733762</td>\n",
       "      <td>0.001724</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.167902</td>\n",
       "      <td>0.009095</td>\n",
       "      <td>0.019190</td>\n",
       "      <td>1.720601e-03</td>\n",
       "      <td>0.1</td>\n",
       "      <td>{'alpha': 0.1}</td>\n",
       "      <td>0.660663</td>\n",
       "      <td>0.679221</td>\n",
       "      <td>0.672648</td>\n",
       "      <td>0.669182</td>\n",
       "      <td>...</td>\n",
       "      <td>0.669995</td>\n",
       "      <td>0.006048</td>\n",
       "      <td>4</td>\n",
       "      <td>0.731362</td>\n",
       "      <td>0.729790</td>\n",
       "      <td>0.732698</td>\n",
       "      <td>0.733074</td>\n",
       "      <td>0.734974</td>\n",
       "      <td>0.732380</td>\n",
       "      <td>0.001735</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0.161309</td>\n",
       "      <td>0.006678</td>\n",
       "      <td>0.018188</td>\n",
       "      <td>9.787721e-04</td>\n",
       "      <td>1</td>\n",
       "      <td>{'alpha': 1}</td>\n",
       "      <td>0.651999</td>\n",
       "      <td>0.668533</td>\n",
       "      <td>0.665159</td>\n",
       "      <td>0.657880</td>\n",
       "      <td>...</td>\n",
       "      <td>0.660784</td>\n",
       "      <td>0.005744</td>\n",
       "      <td>5</td>\n",
       "      <td>0.717862</td>\n",
       "      <td>0.716038</td>\n",
       "      <td>0.718325</td>\n",
       "      <td>0.719814</td>\n",
       "      <td>0.721426</td>\n",
       "      <td>0.718693</td>\n",
       "      <td>0.001822</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>0.167703</td>\n",
       "      <td>0.019167</td>\n",
       "      <td>0.019787</td>\n",
       "      <td>3.652249e-03</td>\n",
       "      <td>10</td>\n",
       "      <td>{'alpha': 10}</td>\n",
       "      <td>0.594915</td>\n",
       "      <td>0.602411</td>\n",
       "      <td>0.606176</td>\n",
       "      <td>0.592003</td>\n",
       "      <td>...</td>\n",
       "      <td>0.600202</td>\n",
       "      <td>0.005725</td>\n",
       "      <td>6</td>\n",
       "      <td>0.634694</td>\n",
       "      <td>0.631277</td>\n",
       "      <td>0.631562</td>\n",
       "      <td>0.636071</td>\n",
       "      <td>0.634182</td>\n",
       "      <td>0.633557</td>\n",
       "      <td>0.001854</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>0.193887</td>\n",
       "      <td>0.006159</td>\n",
       "      <td>0.023188</td>\n",
       "      <td>2.314960e-03</td>\n",
       "      <td>100</td>\n",
       "      <td>{'alpha': 100}</td>\n",
       "      <td>0.547873</td>\n",
       "      <td>0.547446</td>\n",
       "      <td>0.549319</td>\n",
       "      <td>0.545836</td>\n",
       "      <td>...</td>\n",
       "      <td>0.550889</td>\n",
       "      <td>0.006635</td>\n",
       "      <td>7</td>\n",
       "      <td>0.559569</td>\n",
       "      <td>0.557829</td>\n",
       "      <td>0.557272</td>\n",
       "      <td>0.559055</td>\n",
       "      <td>0.554544</td>\n",
       "      <td>0.557654</td>\n",
       "      <td>0.001759</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>7</td>\n",
       "      <td>0.176098</td>\n",
       "      <td>0.010063</td>\n",
       "      <td>0.019390</td>\n",
       "      <td>1.355722e-03</td>\n",
       "      <td>1000</td>\n",
       "      <td>{'alpha': 1000}</td>\n",
       "      <td>0.540736</td>\n",
       "      <td>0.535326</td>\n",
       "      <td>0.535413</td>\n",
       "      <td>0.539907</td>\n",
       "      <td>...</td>\n",
       "      <td>0.540978</td>\n",
       "      <td>0.006650</td>\n",
       "      <td>8</td>\n",
       "      <td>0.542960</td>\n",
       "      <td>0.543433</td>\n",
       "      <td>0.543193</td>\n",
       "      <td>0.542570</td>\n",
       "      <td>0.539028</td>\n",
       "      <td>0.542237</td>\n",
       "      <td>0.001630</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>8</td>\n",
       "      <td>0.170901</td>\n",
       "      <td>0.002530</td>\n",
       "      <td>0.019188</td>\n",
       "      <td>1.166033e-03</td>\n",
       "      <td>10000</td>\n",
       "      <td>{'alpha': 10000}</td>\n",
       "      <td>0.538314</td>\n",
       "      <td>0.531779</td>\n",
       "      <td>0.532987</td>\n",
       "      <td>0.538095</td>\n",
       "      <td>...</td>\n",
       "      <td>0.537740</td>\n",
       "      <td>0.005555</td>\n",
       "      <td>9</td>\n",
       "      <td>0.538577</td>\n",
       "      <td>0.539600</td>\n",
       "      <td>0.539255</td>\n",
       "      <td>0.538215</td>\n",
       "      <td>0.535601</td>\n",
       "      <td>0.538249</td>\n",
       "      <td>0.001411</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>9 rows × 21 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "   mean_fit_time  std_fit_time  mean_score_time  std_score_time param_alpha  \\\n",
       "0       0.161874      0.013547         0.019491    2.794829e-03      0.0001   \n",
       "1       0.179296      0.010377         0.017990    5.560829e-07       0.001   \n",
       "2       0.159305      0.003260         0.017592    4.869431e-04        0.01   \n",
       "3       0.167902      0.009095         0.019190    1.720601e-03         0.1   \n",
       "4       0.161309      0.006678         0.018188    9.787721e-04           1   \n",
       "5       0.167703      0.019167         0.019787    3.652249e-03          10   \n",
       "6       0.193887      0.006159         0.023188    2.314960e-03         100   \n",
       "7       0.176098      0.010063         0.019390    1.355722e-03        1000   \n",
       "8       0.170901      0.002530         0.019188    1.166033e-03       10000   \n",
       "\n",
       "              params  split0_test_score  split1_test_score  split2_test_score  \\\n",
       "0  {'alpha': 0.0001}           0.661381           0.680464           0.673421   \n",
       "1   {'alpha': 0.001}           0.661365           0.680452           0.673414   \n",
       "2    {'alpha': 0.01}           0.661253           0.680336           0.673355   \n",
       "3     {'alpha': 0.1}           0.660663           0.679221           0.672648   \n",
       "4       {'alpha': 1}           0.651999           0.668533           0.665159   \n",
       "5      {'alpha': 10}           0.594915           0.602411           0.606176   \n",
       "6     {'alpha': 100}           0.547873           0.547446           0.549319   \n",
       "7    {'alpha': 1000}           0.540736           0.535326           0.535413   \n",
       "8   {'alpha': 10000}           0.538314           0.531779           0.532987   \n",
       "\n",
       "   split3_test_score  ...  mean_test_score  std_test_score  rank_test_score  \\\n",
       "0           0.670404  ...         0.670958        0.006194                1   \n",
       "1           0.670395  ...         0.670947        0.006196                2   \n",
       "2           0.670276  ...         0.670852        0.006196                3   \n",
       "3           0.669182  ...         0.669995        0.006048                4   \n",
       "4           0.657880  ...         0.660784        0.005744                5   \n",
       "5           0.592003  ...         0.600202        0.005725                6   \n",
       "6           0.545836  ...         0.550889        0.006635                7   \n",
       "7           0.539907  ...         0.540978        0.006650                8   \n",
       "8           0.538095  ...         0.537740        0.005555                9   \n",
       "\n",
       "   split0_train_score  split1_train_score  split2_train_score  \\\n",
       "0            0.732870            0.731353            0.734320   \n",
       "1            0.732856            0.731339            0.734305   \n",
       "2            0.732716            0.731196            0.734158   \n",
       "3            0.731362            0.729790            0.732698   \n",
       "4            0.717862            0.716038            0.718325   \n",
       "5            0.634694            0.631277            0.631562   \n",
       "6            0.559569            0.557829            0.557272   \n",
       "7            0.542960            0.543433            0.543193   \n",
       "8            0.538577            0.539600            0.539255   \n",
       "\n",
       "   split3_train_score  split4_train_score  mean_train_score  std_train_score  \n",
       "0            0.734554            0.736486          0.733917         0.001723  \n",
       "1            0.734540            0.736471          0.733902         0.001722  \n",
       "2            0.734407            0.736332          0.733762         0.001724  \n",
       "3            0.733074            0.734974          0.732380         0.001735  \n",
       "4            0.719814            0.721426          0.718693         0.001822  \n",
       "5            0.636071            0.634182          0.633557         0.001854  \n",
       "6            0.559055            0.554544          0.557654         0.001759  \n",
       "7            0.542570            0.539028          0.542237         0.001630  \n",
       "8            0.538215            0.535601          0.538249         0.001411  \n",
       "\n",
       "[9 rows x 21 columns]"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model = MultinomialNB()\n",
    "parameters = {'alpha':[0.0001, 0.001, 0.01, 0.1,1,10,100,1000,10000]}  # values changed to interval(10^[-4] to 10^4) as suggested\n",
    "clf = GridSearchCV(model, parameters, cv=5, scoring='roc_auc',return_train_score=True,n_jobs=-1)\n",
    "clf.fit(X_train_TFIDF, y_train)\n",
    "\n",
    "results = pd.DataFrame.from_dict(clf.cv_results_)\n",
    "results = results.sort_values(['param_alpha'])\n",
    "train_auc= results['mean_train_score']\n",
    "train_auc_std= results['std_train_score']\n",
    "cv_auc = results['mean_test_score'] \n",
    "cv_auc_std= results['std_test_score']\n",
    "K =  results['param_alpha']\n",
    "\n",
    "plt.plot(K, train_auc, label='Train AUC')\n",
    "# this code is copied from here: https://stackoverflow.com/a/48803361/4084039\n",
    "# plt.gca().fill_between(K, train_auc - train_auc_std,train_auc + train_auc_std,alpha=0.2,color='darkblue')\n",
    "\n",
    "plt.plot(K, cv_auc, label='CV AUC')\n",
    "# this code is copied from here: https://stackoverflow.com/a/48803361/4084039\n",
    "# plt.gca().fill_between(K, cv_auc - cv_auc_std,cv_auc + cv_auc_std,alpha=0.2,color='darkorange')\n",
    "\n",
    "plt.scatter(K, train_auc, label='Train AUC points')\n",
    "plt.scatter(K, cv_auc, label='CV AUC points')\n",
    "\n",
    "plt.legend()\n",
    "plt.xlabel(\"Alfa Hyperparameter\")\n",
    "plt.ylabel(\"AUC\")\n",
    "plt.title(\"Hyper parameter Vs AUC plot\")\n",
    "plt.grid()\n",
    "plt.show()\n",
    "\n",
    "results\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "MultinomialNB(alpha=0.0001, class_prior=None, fit_prior=True)\n",
      "the best value of hyper-parameter alpha by Gridsearch on TFIDF Vectorization is :-  0.0001\n"
     ]
    }
   ],
   "source": [
    "print(clf.best_estimator_)\n",
    "best_alpha_tfidf = clf.best_estimator_.alpha\n",
    "print('the best value of hyper-parameter alpha by Gridsearch on TFIDF Vectorization is :- ',best_alpha_tfidf)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Testing "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "def batch_predict(clf, data):\n",
    "    # roc_auc_score(y_true, y_score) the 2nd parameter should be probability estimates of the positive class\n",
    "    # not the predicted outputs\n",
    "\n",
    "    y_data_pred = []\n",
    "    tr_loop = data.shape[0] - data.shape[0]%1000\n",
    "    # consider you X_tr shape is 49041, then your tr_loop will be 49041 - 49041%1000 = 49000\n",
    "    # in this for loop we will iterate unti the last 1000 multiplier\n",
    "    for i in range(0, tr_loop, 1000):\n",
    "        y_data_pred.extend(clf.predict_proba(data[i:i+1000])[:,1])\n",
    "    # we will be predicting for the last data points\n",
    "    if data.shape[0]%1000 !=0:\n",
    "        y_data_pred.extend(clf.predict_proba(data[tr_loop:])[:,1])\n",
    "    \n",
    "    return y_data_pred"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.metrics import roc_curve, auc\n",
    "\n",
    "model = MultinomialNB(alpha=best_alpha_bow)\n",
    "model.fit(X_train_BOW, y_train)\n",
    "# roc_auc_score(y_true, y_score) the 2nd parameter should be probability estimates of the positive class\n",
    "# not the predicted outputs\n",
    "\n",
    "y_train_pred_BOW = batch_predict(model, X_train_BOW)    \n",
    "y_test_pred_BOW = batch_predict(model, X_test_BOW)\n",
    "\n",
    "train_fpr_BOW, train_tpr_BOW, train_thresholds_BOW = roc_curve(y_train, y_train_pred_BOW)\n",
    "test_fpr_BOW, test_tpr_BOW, test_thresholds_BOW = roc_curve(y_test, y_test_pred_BOW)\n",
    "\n",
    "plt.plot(train_fpr_BOW, train_tpr_BOW, label=\"train AUC =\"+str(auc(train_fpr_BOW, train_tpr_BOW)))\n",
    "plt.plot(test_fpr_BOW, test_tpr_BOW, label=\"test AUC =\"+str(auc(test_fpr_BOW, test_tpr_BOW)))\n",
    "AUC_BOW=auc(test_fpr_BOW, test_tpr_BOW)\n",
    "plt.legend()\n",
    "plt.xlabel(\"TFR\")\n",
    "plt.ylabel(\"TPR\")\n",
    "plt.title(\"ROC PLOTS\")\n",
    "plt.grid()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.metrics import roc_curve, auc\n",
    "\n",
    "model = MultinomialNB(alpha=best_alpha_tfidf)\n",
    "model.fit(X_train_TFIDF, y_train)\n",
    "# roc_auc_score(y_true, y_score) the 2nd parameter should be probability estimates of the positive class\n",
    "# not the predicted outputs\n",
    "\n",
    "y_train_pred_TFIDF = batch_predict(model, X_train_TFIDF)    \n",
    "y_test_pred_TFIDF = batch_predict(model, X_test_TFIDF)\n",
    "\n",
    "train_fpr_TFIDF, train_tpr_TFIDF, train_thresholds_TFIDF = roc_curve(y_train, y_train_pred_TFIDF)\n",
    "test_fpr_TFIDF, test_tpr_TFIDF, test_thresholds_TFIDF = roc_curve(y_test, y_test_pred_TFIDF)\n",
    "\n",
    "plt.plot(train_fpr_TFIDF, train_tpr_TFIDF, label=\"train AUC =\"+str(auc(train_fpr_TFIDF, train_tpr_TFIDF)))\n",
    "plt.plot(test_fpr_TFIDF, test_tpr_TFIDF, label=\"test AUC =\"+str(auc(test_fpr_TFIDF, test_tpr_TFIDF)))\n",
    "AUC_TFIDF=auc(test_fpr_TFIDF, test_tpr_TFIDF)     \n",
    "plt.legend()\n",
    "plt.xlabel(\"TFR\")\n",
    "plt.ylabel(\"TPR\")\n",
    "plt.title(\"ROC PLOTS\")\n",
    "plt.grid()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "# we are writing our own function for predict, with defined thresould\n",
    "# we will pick a threshold that will give the least fpr\n",
    "def find_best_threshold(threshould, fpr, tpr):\n",
    "    t = threshould[np.argmax(tpr*(1-fpr))]\n",
    "    # (tpr*(1-fpr)) will be maximum if your fpr is very low and tpr is very high\n",
    "#     print(\"the maximum value of tpr*(1-fpr)\", max(tpr*(1-fpr)), \"for threshold\", np.round(t,3))\n",
    "    return t\n",
    "\n",
    "def predict_with_best_t(proba, threshould):\n",
    "    predictions = []\n",
    "    for i in proba:\n",
    "        if i>=threshould:\n",
    "            predictions.append(1)\n",
    "        else:\n",
    "            predictions.append(0)\n",
    "    return predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_CF_MAT(X):\n",
    "    from sklearn.metrics import confusion_matrix\n",
    "    import seaborn as sns\n",
    "    import pandas as pd\n",
    "    \n",
    "    df = pd.DataFrame(X, index = [0,1],columns = [0,1])\n",
    "    print(df)\n",
    "    print('\\n')\n",
    "    sns.cubehelix_palette(8)\n",
    "    sns.set(font_scale=1.5)\n",
    "    df.index.name = 'Actual'\n",
    "    df.columns.name = 'Predicted'\n",
    "    labels =  np.array([[{'TN':X[0][0]},{'FP':X[0][1]}],[{'FN':X[1][0]},{'TP':X[1][1]}]])\n",
    "    fig, ax = plt.subplots()\n",
    "    ax = sns.heatmap(df, annot = labels,fmt='',annot_kws={\"size\": 16},cbar=False,linecolor='white',linewidths=.75,cmap=\"PuBu\")\n",
    "    plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " The confusion matrix of model with BOW vectorization\n",
      "************************************************************\n",
      "      0     1\n",
      "0  1629  1013\n",
      "1  4574  9284\n",
      "\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "****************************************************************************************************\n",
      "**************************************************************************************************** \n",
      "\n",
      "\n",
      " The confusion matrix of model with TFIDF vectorization\n",
      "************************************************************\n",
      "      0     1\n",
      "0  1615  1027\n",
      "1  4723  9135\n",
      "\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "\n",
    "print (\" The confusion matrix of model with BOW vectorization\")\n",
    "print(\"**\"*30)\n",
    "best_t = find_best_threshold(train_thresholds_BOW, train_fpr_BOW, train_tpr_BOW)\n",
    "CF_MAT_BOW=confusion_matrix(y_test, predict_with_best_t(y_test_pred_BOW, best_t))\n",
    "plot_CF_MAT(CF_MAT_BOW)\n",
    "\n",
    "print(\"**\"*50)\n",
    "print(\"**\"*50,\"\\n\\n\")\n",
    "\n",
    "\n",
    "print (\" The confusion matrix of model with TFIDF vectorization\")\n",
    "print(\"**\"*30)\n",
    "best_t = find_best_threshold(train_thresholds_TFIDF, train_fpr_TFIDF, train_tpr_TFIDF)\n",
    "CF_MAT_TFIDF=confusion_matrix(y_test, predict_with_best_t(y_test_pred_TFIDF, best_t))\n",
    "plot_CF_MAT(CF_MAT_TFIDF)\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "def extract_feature_names(n,d,f1,f2,f3,f4,f5,f6,prob_val):\n",
    "    print('The top ',n,' features are --')\n",
    "    print('**'*50)\n",
    "    for i in range(n):\n",
    "        if d[i] < 5000:\n",
    "            print(i+1 ,\": feature belongs to the ESSAY feature in the data with\",prob_val[d[i]],'probability, and the feature name after_vectorising is,',f1[d[i]])\n",
    "        elif d[i] >= 5000 and d[i]<5051:\n",
    "            print(i+1 ,\": feature belongs to the SCHOOL_STATE feature in the data with\",prob_val[d[i]],'probability, and the feature name after_vectorising is,',f2[d[i]-5000])\n",
    "        elif d[i] >= 5051 and d[i]<5056:\n",
    "            print(i+1 ,\": feature belongs to the TEACHER_PREFIX feature in the data with\",prob_val[d[i]],'probability, and the feature name after_vectorising is,',f3[d[i]-5051])\n",
    "        elif d[i] >= 5056 and d[i]<5060:\n",
    "            print(i+1 ,\": feature belongs to the PROJECT_GRADE_CATEGORY feature in the data with\",prob_val[d[i]],'probability, and the feature name after_vectorising is,',f4[d[i]-5056])\n",
    "        elif d[i] >= 5060 and d[i]<5069:\n",
    "            print(i+1 ,\": feature belongs to the CLEAN_CATEGORIES feature in the data with\",prob_val[d[i]],'probability, and the feature name after_vectorising is,',f5[d[i]-5060])\n",
    "        elif d[i] >= 5069 and d[i]<5099:\n",
    "            print(i+1 ,\": feature belongs to the CLEAN_SUB_CATEGORIES feature in the data with\",prob_val[d[i]],'probability, and the feature name after_vectorising is,',f6[d[i]-5069])\n",
    "        elif d[i] == 5099:\n",
    "            print(i+1 ,\": feature belongs to the PRICE feature in the data with\",prob_val[d[i]],'probability ')\n",
    "        else:\n",
    "            print(i+1 ,\": feature belongs to the PRICE feature in the data with\",prob_val[d[i]],'probability ')\n",
    "        \n",
    "    print('**'*50)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The top  20  features are --\n",
      "****************************************************************************************************\n",
      "1 : feature belongs to the CLEAN_SUB_CATEGORIES feature in the data with 13.464030908513056 probability, and the feature name after_vectorising is, warmth\n",
      "2 : feature belongs to the CLEAN_SUB_CATEGORIES feature in the data with 13.464030908513056 probability, and the feature name after_vectorising is, care_hunger\n",
      "3 : feature belongs to the CLEAN_CATEGORIES feature in the data with 13.460873361739761 probability, and the feature name after_vectorising is, warmth\n",
      "4 : feature belongs to the CLEAN_CATEGORIES feature in the data with 13.460873361739761 probability, and the feature name after_vectorising is, care_hunger\n",
      "5 : feature belongs to the TEACHER_PREFIX feature in the data with 11.99895007024821 probability, and the feature name after_vectorising is, dr\n",
      "6 : feature belongs to the ESSAY feature in the data with 10.703444812833702 probability, and the feature name after_vectorising is, school day by\n",
      "7 : feature belongs to the SCHOOL_STATE feature in the data with 10.69969132802797 probability, and the feature name after_vectorising is, wy\n",
      "8 : feature belongs to the ESSAY feature in the data with 10.681571760357583 probability, and the feature name after_vectorising is, earliest learners continue path\n",
      "9 : feature belongs to the ESSAY feature in the data with 10.68120527801569 probability, and the feature name after_vectorising is, however certainly control experience\n",
      "10 : feature belongs to the ESSAY feature in the data with 10.678242773939184 probability, and the feature name after_vectorising is, us see\n",
      "11 : feature belongs to the ESSAY feature in the data with 10.672279782568404 probability, and the feature name after_vectorising is, even earliest learners continue\n",
      "12 : feature belongs to the ESSAY feature in the data with 10.665913967902972 probability, and the feature name after_vectorising is, prevent getting\n",
      "13 : feature belongs to the ESSAY feature in the data with 10.66044223794927 probability, and the feature name after_vectorising is, lives however certainly control\n",
      "14 : feature belongs to the ESSAY feature in the data with 10.658868370405722 probability, and the feature name after_vectorising is, earliest learners continue\n",
      "15 : feature belongs to the ESSAY feature in the data with 10.654963752348786 probability, and the feature name after_vectorising is, hopeful inspire even\n",
      "16 : feature belongs to the ESSAY feature in the data with 10.652232201183946 probability, and the feature name after_vectorising is, inspire even earliest learners\n",
      "17 : feature belongs to the ESSAY feature in the data with 10.652232201183946 probability, and the feature name after_vectorising is, even earliest learners\n",
      "18 : feature belongs to the ESSAY feature in the data with 10.64755067957362 probability, and the feature name after_vectorising is, certainly control experience school\n",
      "19 : feature belongs to the ESSAY feature in the data with 10.647290495480494 probability, and the feature name after_vectorising is, creative positive way hopeful\n",
      "20 : feature belongs to the ESSAY feature in the data with 10.645836884673486 probability, and the feature name after_vectorising is, inspire even earliest\n",
      "****************************************************************************************************\n"
     ]
    }
   ],
   "source": [
    "# code is taken from here (https://stackoverflow.com/questions/36265930/how-to-sort-in-descending-order-with-numpy)\n",
    "from numpy import argsort\n",
    "prob_val = abs(model.feature_log_prob_[1]) # Probabilities related to the class label 1\n",
    "feature_index=np.argsort(-prob_val)  # -prob_val for descending order\n",
    "extract_feature_names(20,feature_index,essay_feature_TF,school_state_feature_TF,teacher_prefix_feature_TF,project_grade_category_feature_TF,clean_categories_feature_TF,clean_subcategories_feature_TF,prob_val)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "lS9hIHdib3dp"
   },
   "source": [
    "<h1>3. Summary</h1>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+---------------+---------------+------------------------+-----------------+--------------------+\n",
      "| Vectorization |     Model     | Hyper-parameter tuning | Hyper-parameter |     AUC_SCORE      |\n",
      "+---------------+---------------+------------------------+-----------------+--------------------+\n",
      "|      BOW      | MultinomialNB |      GridSearchCV      |      0.001      | 0.6913981615627919 |\n",
      "|     TFIDF     | MultinomialNB |      GridSearchCV      |      0.0001     | 0.6782618669583531 |\n",
      "+---------------+---------------+------------------------+-----------------+--------------------+\n"
     ]
    }
   ],
   "source": [
    "from prettytable import PrettyTable\n",
    "\n",
    "x = PrettyTable([\"Vectorization\",\"Model\", \"Hyper-parameter tuning\", \"Hyper-parameter\" ,\"AUC_SCORE\"])\n",
    "x.add_row([\" BOW \", \"MultinomialNB\",\"GridSearchCV\", best_alpha_bow, AUC_BOW])\n",
    "x.add_row([\" TFIDF \",\"MultinomialNB\", \"GridSearchCV\", best_alpha_tfidf, AUC_TFIDF])\n",
    "print(x)"
   ]
  }
 ],
 "metadata": {
  "colab": {
   "collapsed_sections": [],
   "name": "6_Assignment_NB_Instructions.ipynb",
   "provenance": [],
   "version": "0.3.2"
  },
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}