Group2_Part_1_CategoryClassification_Final.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Q&A GOOGLE QUEST PROJECT \n",
    "**Text Analytics | Prof Evan**\n",
    "\n",
    "**Author:** Tram Ngoc Le\n",
    "\n",
    "## Part I: Q&A Category Classification\n",
    "\n",
    "In Q&A Category Classification part, Google pairs of questions and answers are classified into **5 topics**, including **life/arts, culture, science, technology**, and **stackoverflow** by application of supervised machine learning technique along with text analytics. \n",
    "\n",
    "The process of the methodology implementation are set up as below:\n",
    "\n",
    "- **Step 1. Preliminary steps**  - Set up packages | Load Text | Initial exploratory | Text Preprocessing\n",
    "- **Step 2. Text Preprocessing** - Stopwords Removal, Stemming, Lowercase, ...etc\n",
    "- **Step 3. Feature Engineering**  - TFIDF features\n",
    "\n",
    "- **Step 4. Oversampling** - Handle unbalanced class problem with TFIDF features\n",
    "- **Step 5. Model Building** - SVM | Naive Bayes | Random Forest | Decision Tree \n",
    "- **Step 6. Model Selection** - Model Comparison & Selection - Metrics: Accuracy rate on training data\n",
    "- **Step 7. Model Evaluation** - Metrics: Accuracy, F1 Score, Precision Score, Recall Score\n",
    "- **Step 8. Conclusion**\n",
    "- **Appendix** - Evaluate other features + models that were tried before the best features and best models are created \n",
    "- **Reference**\n",
    "**Note**: *Questions and Answers are categorized seperately*\n",
    "# ___\n",
    "## Step 1: Preliminary Steps\n",
    "- Setup required packages/libraries "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "# Set up\n",
    "\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import nltk\n",
    "import re\n",
    "import gensim\n",
    "import seaborn as sns\n",
    "%matplotlib inline\n",
    "import matplotlib\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.model_selection import cross_val_score\n",
    "# SMOTE\n",
    "import imblearn\n",
    "# print(imblearn.__version__)\n",
    "from imblearn.over_sampling import SMOTE #use for oversampling \n",
    "from collections import Counter"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- Load Data & Examize Training vs Testing\n",
    "- The Kaggle provide both training and testing data for CATEGORY CLASSIFICATION\n",
    "- The partition between **training and testing** is **93% - 7%** of the data, respectively"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load Data\n",
    "train = pd.read_csv(\"train.csv\")\n",
    "test = pd.read_csv(\"test.csv\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "df_train = train\n",
    "df_test = test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Observations for trainning: 6079 \n",
      "Observations for testing: 476\n",
      "Train-Test: 93 - 7\n"
     ]
    }
   ],
   "source": [
    "# Training vs. Testing Ratio\n",
    "print(\"Observations for trainning:\", len(train),\"\\nObservations for testing:\", len(test))\n",
    "print(\"Train-Test:\", round((len(train)/(len(train)+len(test)))*100),\"-\",round((len(test)/(len(train)+len(test)))*100))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- Preprocess Data | Adding category_id to the data set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Preprocess data\n",
    "# Reduce the dimension of the original dataset\n",
    "train = train[['question_body','answer','category']]\n",
    "test = test[['question_body','answer','category']]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ID to Category: {0: 'LIFE_ARTS', 1: 'CULTURE', 2: 'SCIENCE', 3: 'STACKOVERFLOW', 4: 'TECHNOLOGY'}\n",
      "Category to ID: {'LIFE_ARTS': 0, 'CULTURE': 1, 'SCIENCE': 2, 'STACKOVERFLOW': 3, 'TECHNOLOGY': 4}\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\Jennie\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py:3: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
      "  This is separate from the ipykernel package so we can avoid doing imports until\n"
     ]
    },
    {
     "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>question_body</th>\n",
       "      <th>answer</th>\n",
       "      <th>category</th>\n",
       "      <th>category_id</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>After playing around with macro photography on...</td>\n",
       "      <td>I just got extension tubes, so here's the skin...</td>\n",
       "      <td>LIFE_ARTS</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>I am trying to understand what kinds of places...</td>\n",
       "      <td>It might be helpful to look into the definitio...</td>\n",
       "      <td>CULTURE</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>I'm working on a PCB that has through-hole com...</td>\n",
       "      <td>Do you even need grooves?  We make several pro...</td>\n",
       "      <td>SCIENCE</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>An affidavit, from what i understand, is basic...</td>\n",
       "      <td>Sending an \"affidavit\" it is a dispute between...</td>\n",
       "      <td>CULTURE</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>I am trying to make a binary image. I want mor...</td>\n",
       "      <td>Check out Image Trace in Adobe Illustrator. \\n...</td>\n",
       "      <td>LIFE_ARTS</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                       question_body  \\\n",
       "0  After playing around with macro photography on...   \n",
       "1  I am trying to understand what kinds of places...   \n",
       "2  I'm working on a PCB that has through-hole com...   \n",
       "3  An affidavit, from what i understand, is basic...   \n",
       "4  I am trying to make a binary image. I want mor...   \n",
       "\n",
       "                                              answer   category  category_id  \n",
       "0  I just got extension tubes, so here's the skin...  LIFE_ARTS            0  \n",
       "1  It might be helpful to look into the definitio...    CULTURE            1  \n",
       "2  Do you even need grooves?  We make several pro...    SCIENCE            2  \n",
       "3  Sending an \"affidavit\" it is a dispute between...    CULTURE            1  \n",
       "4  Check out Image Trace in Adobe Illustrator. \\n...  LIFE_ARTS            0  "
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Setting category id for the target\n",
    "# Assign ID to 5 values in category \n",
    "train['category_id']=train['category'].factorize()[0] \n",
    "# Set category to id and id to category\n",
    "category_id_df = train[['category','category_id']].drop_duplicates().sort_values('category_id')\n",
    "# Set category to id\n",
    "category_to_id = dict(category_id_df.values)\n",
    "# Set id to category\n",
    "id_to_category = dict(category_id_df[['category_id','category']].values)\n",
    "print(\"ID to Category:\", id_to_category)\n",
    "print(\"Category to ID:\", category_to_id)\n",
    "train.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\Jennie\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py:15: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
      "  from ipykernel import kernelapp as app\n"
     ]
    },
    {
     "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>question_body</th>\n",
       "      <th>answer</th>\n",
       "      <th>category</th>\n",
       "      <th>category_id</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>see questions/information online get rid corps...</td>\n",
       "      <td>consequence leaving corps anywhere . leave mar...</td>\n",
       "      <td>CULTURE</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>new wordpress . issue feature image . need add...</td>\n",
       "      <td>think possible custom field . add custom field...</td>\n",
       "      <td>TECHNOLOGY</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>experiment started bot game , toggled invincib...</td>\n",
       "      <td>armour screenshots . suggests affected large a...</td>\n",
       "      <td>CULTURE</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>used raspberry pi torrent-server quite . weste...</td>\n",
       "      <td>western digital hard drive disappearing shown ...</td>\n",
       "      <td>TECHNOLOGY</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>bought delhi-london return flight 4 saudi airl...</td>\n",
       "      <td>called two person work saudia ( ticketing agen...</td>\n",
       "      <td>CULTURE</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                       question_body  \\\n",
       "0  see questions/information online get rid corps...   \n",
       "1  new wordpress . issue feature image . need add...   \n",
       "2  experiment started bot game , toggled invincib...   \n",
       "3  used raspberry pi torrent-server quite . weste...   \n",
       "4  bought delhi-london return flight 4 saudi airl...   \n",
       "\n",
       "                                              answer    category  category_id  \n",
       "0  consequence leaving corps anywhere . leave mar...     CULTURE            1  \n",
       "1  think possible custom field . add custom field...  TECHNOLOGY            4  \n",
       "2  armour screenshots . suggests affected large a...     CULTURE            1  \n",
       "3  western digital hard drive disappearing shown ...  TECHNOLOGY            4  \n",
       "4  called two person work saudia ( ticketing agen...     CULTURE            1  "
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "category_id_df\n",
    "A = []\n",
    "for i in test.category:\n",
    "    if i == \"LIFE_ARTS\":\n",
    "        category_id = 0\n",
    "    elif i == \"CULTURE\":\n",
    "        category_id = 1\n",
    "    elif i == \"SCIENCE\":\n",
    "        category_id = 2\n",
    "    elif i == \"STACKOVERFLOW\":\n",
    "        category_id = 3\n",
    "    else:\n",
    "        category_id = 4\n",
    "    A.append(category_id)\n",
    "test['category_id'] = A\n",
    "test.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- Initial Data Exploratory on Multi-class Target"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Observation Distribution in Training Data (%):\n",
      "TECHNOLOGY       40.0\n",
      "STACKOVERFLOW    21.0\n",
      "CULTURE          16.0\n",
      "SCIENCE          12.0\n",
      "LIFE_ARTS        12.0\n",
      "Name: category, dtype: float64\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "Text(0.5, 1.0, 'Observations Distribution of Classes')"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Number of observations\n",
    "print(\"Observation Distribution in Training Data (%):\")\n",
    "print(round((train.category.value_counts()/len(train.category))*100))\n",
    "# Create a bar chat of distribution of observations for each class\n",
    "plt.figure(figsize=(10,6))\n",
    "sns.countplot(x=\"category\", data=train, palette=\"Set2\", saturation = 1).set_title(\"Observations Distribution of Classes\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### **`Problem - Imbalanced Classes`**\n",
    "\n",
    "- Based on the **Category** distribution graph, one possible problem has emerged, which is **imbalanced classes**. Number of observations for **Technology** is **extremely higher** than other clasees. When encountering such problem, there are absolutely difficulties with standard algorithms since they can be more biased towards the majority class. Furthermore, it can affect the accuracy rate. \n",
    "- However, one **solution** here is **oversampling** the training features before fitting the model. This methodology will be implemented after feature engineering\n",
    "\n",
    "# ___\n",
    "## Step 2: Text Preprocessing\n",
    "- Stopwords Removal\n",
    "- Stemming & Tokenization\n",
    "- Lowercase\n",
    "- Filter out non-character/non-alphabet"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "from nltk.corpus import stopwords\n",
    "def preprocessing_text(text):\n",
    "    # Stopwords Removal\n",
    "    my_stopwords = stopwords.words('english')\n",
    "    # Stemming\n",
    "    wnl = nltk.WordNetLemmatizer()\n",
    "    tag_remove = re.compile(r'<[^>]+>')\n",
    "    text = tag_remove.sub('',text)\n",
    "    # tokenization\n",
    "    tokens = nltk.word_tokenize(text)\n",
    "    # alphabet / lowercase / stopwords removal\n",
    "    words = [w.lower() for w in tokens if w.isalpha if w.lower() not in my_stopwords]\n",
    "    stem = [wnl.lemmatize(w) for w in words]\n",
    "    clean_text = ' '.join(w for w in stem)\n",
    "    clean_text = clean_text.replace(\"``\",\"\")\n",
    "    return clean_text"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\Jennie\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py:2: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
      "  \n",
      "C:\\Users\\Jennie\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py:4: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
      "  after removing the cwd from sys.path.\n",
      "C:\\Users\\Jennie\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py:6: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
      "  \n",
      "C:\\Users\\Jennie\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py:8: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
      "  \n"
     ]
    }
   ],
   "source": [
    "# Preprocess ##question for training data\n",
    "train['question_body'] = train['question_body'].apply(lambda x:preprocessing_text(x))\n",
    "# Preprocess ##answer for training data\n",
    "train['answer'] = train['answer'].apply(lambda x:preprocessing_text(x))\n",
    "# Preprocess ##question for testing data\n",
    "test['question_body'] = test['question_body'].apply(lambda x:preprocessing_text(x))\n",
    "# Preprocess ##answer for testing data\n",
    "test['answer'] = test['answer'].apply(lambda x:preprocessing_text(x))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_question, test_question, train_labels, test_labels = train.question_body.tolist(), test.question_body.tolist(), train.category_id.tolist(), test.category_id.tolist()\n",
    "train_answer, test_answer, train_labels, test_labels = train.answer.tolist(), test.answer.tolist(), train.category_id.tolist(), test.category_id.tolist()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "\"trying understand kind place spam value p 231 refer 5th edition main book shadowrun . per p 15 , sprawl plex , plex  metropolitan complex , short metroplex '' . per google metroplex  large metropolitan area , especially one aggregation two city '' . city downtown sprawl downtown would tend similar density , reason sprawl ( includes suburb ? ) higher spam zone noise rating ( p 231 ) . similarly , 'd think downtown dense noisy ( e.g . office building street vendor ) commercial district , e.g . outdoor mall . noise rating make think thinking incorrectly . better way thinking ?\""
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_question[1]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# ___\n",
    "## Step 3: Feature Engineering\n",
    "\n",
    "- There are 3 features were created during the process: BOW, TFIDF and Word2Vec. This part demonstrate the **best** feature **TFIDF** that help to achieve the highest accuracy. \n",
    "- For information about **BOW** and **Word2Vec**, visit **Step 9: Appendix**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "# TFIDF features\n",
    "from sklearn.feature_extraction.text import TfidfVectorizer\n",
    "\n",
    "tfidf_vectorizer = TfidfVectorizer(sublinear_tf=True, # use logarithmic form for frequency\n",
    "                                    min_df=1, # The minimum numbers of document a word must be present to be kept\n",
    "                                   norm='l2', # Ensure all features vectors have a eulicidean norm of 1\n",
    "                                   encoding='latin-1',\n",
    "                                   smooth_idf=True,\n",
    "                                   use_idf=True, \n",
    "                                   ngram_range=(1,2)) #unigram and bigram\n",
    "\n",
    "# QUESTION - TFIDF\n",
    "tfidf_train_question = tfidf_vectorizer.fit_transform(train_question)\n",
    "tfidf_test_question = tfidf_vectorizer.transform(test_question)\n",
    "\n",
    "# ANSWER - TFIDF\n",
    "tfidf_train_answer = tfidf_vectorizer.fit_transform(train_answer)\n",
    "tfidf_test_answer = tfidf_vectorizer.transform(test_answer) "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Find the terms that are the most correlated with each topic [Question]**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "# TOPIC: 'CULTURE':\n",
      "  . Most correlated unigrams:\n",
      ". australia\n",
      ". extradition\n",
      ". infected\n",
      ". facelet\n",
      ". brush\n",
      "  . Most correlated bigrams:\n",
      ". code andrew\n",
      ". data matasano\n",
      ". anything using\n",
      ". invalidoperationexception change\n",
      ". mysql get\n",
      "# TOPIC: 'LIFE_ARTS':\n",
      "  . Most correlated unigrams:\n",
      ". cours\n",
      ". coil\n",
      ". date\n",
      ". fridge\n",
      ". accountant\n",
      "  . Most correlated bigrams:\n",
      ". ask rate\n",
      ". display tag\n",
      ". harder applying\n",
      ". library nautilus\n",
      ". energy generally\n",
      "# TOPIC: 'SCIENCE':\n",
      "  . Most correlated unigrams:\n",
      ". extensibility\n",
      ". downsize\n",
      ". destiny\n",
      ". divs\n",
      ". logo\n",
      "  . Most correlated bigrams:\n",
      ". hearing work\n",
      ". cnt select\n",
      ". hand account\n",
      ". could update\n",
      ". expression local\n",
      "# TOPIC: 'STACKOVERFLOW':\n",
      "  . Most correlated unigrams:\n",
      ". fk_storenum_previousfinished\n",
      ". condensor\n",
      ". affixed\n",
      ". ledonms\n",
      ". backup_dir\n",
      "  . Most correlated bigrams:\n",
      ". basically kurtosis\n",
      ". level amp\n",
      ". debate conclusion\n",
      ". decorative piece\n",
      ". exact specific\n",
      "# TOPIC: 'TECHNOLOGY':\n",
      "  . Most correlated unigrams:\n",
      ". datacolumn\n",
      ". drupal_add_css\n",
      ". chbr\n",
      ". erratum\n",
      ". multiplexing\n",
      "  . Most correlated bigrams:\n",
      ". involved judgement\n",
      ". flash frequency\n",
      ". dogflap page\n",
      ". kernel img\n",
      ". martin smith\n"
     ]
    }
   ],
   "source": [
    "from sklearn.feature_selection import chi2\n",
    "features = tfidf_train_question\n",
    "labels = train.category_id\n",
    "N = 5 # extract the 5 terms that are most relating to each topic\n",
    "for category, category_id in sorted(category_to_id.items()):\n",
    "    features_chi = chi2(features, labels == category_id)\n",
    "    index = np.argsort(features_chi[0])\n",
    "    feature_names = np.array(tfidf_vectorizer.get_feature_names())[index]\n",
    "    unigrams = [w for w in feature_names if len(w.split(' ')) == 1]\n",
    "    bigrams = [w for w in feature_names if len(w.split(' ')) == 2]\n",
    "    print(\"# TOPIC: '{}':\".format(category))\n",
    "    print(\"  . Most correlated unigrams:\\n. {}\".format('\\n. '.join(unigrams[-N:])))\n",
    "    print(\"  . Most correlated bigrams:\\n. {}\".format('\\n. '.join(bigrams[-N:])))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "# TOPIC: 'CULTURE':\n",
      "  . Most correlated unigrams:\n",
      ". bike\n",
      ". god\n",
      ". english\n",
      ". game\n",
      ". word\n",
      "  . Most correlated bigrams:\n",
      ". move something\n",
      ". native speaker\n",
      ". work ability\n",
      ". present tense\n",
      ". could say\n",
      "# TOPIC: 'LIFE_ARTS':\n",
      "  . Most correlated unigrams:\n",
      ". phd\n",
      ". tax\n",
      ". camera\n",
      ". student\n",
      ". lens\n",
      "  . Most correlated bigrams:\n",
      ". credit score\n",
      ". white balance\n",
      ". zoom lens\n",
      ". star trek\n",
      ". focal length\n",
      "# TOPIC: 'SCIENCE':\n",
      "  . Most correlated unigrams:\n",
      ". times\n",
      ". electron\n",
      ". energy\n",
      ". mathbb\n",
      ". frac\n",
      "  . Most correlated bigrams:\n",
      ". black hole\n",
      ". angular momentum\n",
      ". potential energy\n",
      ". in mathbb\n",
      ". frac frac\n",
      "# TOPIC: 'STACKOVERFLOW':\n",
      "  . Most correlated unigrams:\n",
      ". void\n",
      ". string\n",
      ". return\n",
      ". gt\n",
      ". lt\n",
      "  . Most correlated bigrams:\n",
      ". public void\n",
      ". public class\n",
      ". div gt\n",
      ". lt div\n",
      ". gt lt\n",
      "# TOPIC: 'TECHNOLOGY':\n",
      "  . Most correlated unigrams:\n",
      ". usepackage\n",
      ". sudo\n",
      ". install\n",
      ". ubuntu\n",
      ". server\n",
      "  . Most correlated bigrams:\n",
      ". ip address\n",
      ". sudo apt\n",
      ". apt get\n",
      ". end document\n",
      ". begin document\n"
     ]
    }
   ],
   "source": [
    "from sklearn.feature_selection import chi2\n",
    "features = tfidf_train_answer\n",
    "labels = train.category_id\n",
    "N = 5 # extract the 5 terms that are most relating to each topic\n",
    "for category, category_id in sorted(category_to_id.items()):\n",
    "    features_chi = chi2(features, labels == category_id)\n",
    "    index = np.argsort(features_chi[0])\n",
    "    feature_names = np.array(tfidf_vectorizer.get_feature_names())[index]\n",
    "    unigrams = [w for w in feature_names if len(w.split(' ')) == 1]\n",
    "    bigrams = [w for w in feature_names if len(w.split(' ')) == 2]\n",
    "    print(\"# TOPIC: '{}':\".format(category))\n",
    "    print(\"  . Most correlated unigrams:\\n. {}\".format('\\n. '.join(unigrams[-N:])))\n",
    "    print(\"  . Most correlated bigrams:\\n. {}\".format('\\n. '.join(bigrams[-N:])))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# ___\n",
    "## Step 4: Over Sampling\n",
    "\n",
    "- The method of over sampling are used with `SMOTE` - This will randomly oversample obervations for each category until they are equally distributed among the train data. \n",
    "\n",
    "- Oversampling is only implemented on training data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Setup of packages needed\n",
    "# SMOTE\n",
    "import imblearn\n",
    "# print(imblearn.__version__)\n",
    "from imblearn.over_sampling import SMOTE #use for oversampling\n",
    "from collections import Counter\n",
    "# X: input for oversampling\n",
    "# y: target label for oversampling\n",
    "oversample = SMOTE()\n",
    "X_train_question, y_train_question = oversample.fit_resample(tfidf_train_question, train_labels)\n",
    "X_train_answer, y_train_answer = oversample.fit_resample(tfidf_train_answer, train_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Counter({0: 2441, 1: 2441, 2: 2441, 3: 2441, 4: 2441})"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# The distribution of observations in those dataset\n",
    "counter = Counter(y_train_question)\n",
    "plt.figure(figsize=(8,5))\n",
    "plt.bar(counter.keys(), counter.values())\n",
    "counter"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Counter({0: 2441, 1: 2441, 2: 2441, 3: 2441, 4: 2441})"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "counter = Counter(y_train_answer)\n",
    "plt.figure(figsize=(8,5))\n",
    "plt.bar(counter.keys(), counter.values())\n",
    "counter"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "After oversampling, the observations in all category are equal to one another."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# ___\n",
    "## Step 5: Model Building\n",
    "\n",
    "- Define matrix for evaluation: Accuracy, Precision, Recall, F1 Score\n",
    "- Define how to train and evaluate classifiers\n",
    "- Building the models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Define matrix for evaluation\n",
    "from sklearn import metrics\n",
    "# define a function to evaluate our classification models based on four metrics\n",
    "def get_metrics(true_labels, predicted_labels): \n",
    "    print('Accuracy:', np.round(metrics.accuracy_score(true_labels,\n",
    "                                                        predicted_labels),5)) \n",
    "    print('Precision:', np.round(metrics.precision_score(true_labels,\n",
    "                                                         predicted_labels,\n",
    "                                                         average='weighted',\n",
    "                                                         zero_division=1),5)) \n",
    "    # average parameter is required for multiclass target multiple labels\n",
    "    print('Recall:', np.round(metrics.recall_score(true_labels, \n",
    "                                                   predicted_labels,\n",
    "                                                   average='weighted',\n",
    "                                                   zero_division=1),5)) \n",
    "    # average parameter is required for multiclass target multiple labels\n",
    "    print('F1 Score:',np.round(metrics.f1_score(true_labels,\n",
    "                                               predicted_labels,\n",
    "                                               average='weighted',\n",
    "                                               zero_division=1),5)) \n",
    "    # average parameter is required for multiclass target multiple labels\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Define a function that trains the model, performs predictions and evaluates the predictions\n",
    "def train_predict_evaluate_model(classifier,\n",
    "                                train_features, train_labels,\n",
    "                                 test_features, test_labels):\n",
    "    # Build model\n",
    "    classifier.fit(train_features, train_labels)\n",
    "    # Predict using model\n",
    "    predictions = classifier.predict(test_features)\n",
    "    # Evaluate model prediction performance\n",
    "    get_metrics(true_labels=test_labels,\n",
    "               predicted_labels=predictions)\n",
    "    return predictions"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Model Building for Multi-class Classification**\n",
    "- Naive Bayes\n",
    "- Support Vector Machine (SDGC)\n",
    "- Decision Tree\n",
    "- Random Forest"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Model Building\n",
    "from sklearn.naive_bayes import MultinomialNB # Naive Bayes \n",
    "from sklearn.linear_model import SGDClassifier # Support Vector Machine (SDG)\n",
    "from sklearn.tree import DecisionTreeClassifier # Decision Tree\n",
    "from sklearn.ensemble import RandomForestClassifier #RandomForestClassifier\n",
    "\n",
    "mnb = MultinomialNB() # Naive Bayes\n",
    "svm = SGDClassifier(loss='hinge', max_iter=100) # SVM SDG\n",
    "dct = DecisionTreeClassifier(criterion = \"entropy\", # Decision Tree\n",
    "                             # set criterion as entropy for the information gain \n",
    "                             #(measure the quality of the split)\n",
    "                             random_state = 0)\n",
    "rf = RandomForestClassifier(n_estimators=200, max_depth=3, random_state=0) # Random Forest"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# ___\n",
    "## Step 6: Model Selection\n",
    "- Fit and evaluate Naive Bayes, SVM, Decision Tree & Random Forest with TFIDF\n",
    "- Evaluate the models based on the accuracy rate from the training set\n",
    "- Models accuracy visualization\n",
    "\n",
    "### **`Note`**\n",
    "\n",
    "*This section **only features the models selection regarding TFIDF** - which are the **best models** built with the best feature regarding the category classification project.*\n",
    "\n",
    "*For further details on **other models selections** , please refers to **Step 9: Appendix**, which includes all the models and features we have tried before come up with the best model.*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Set up\n",
    "models = [mnb, svm, dct, rf]\n",
    "cross_validation = 5 # 5-fold cross validation\n",
    "cv_tfidf = pd.DataFrame(index=range(cross_validation*len(models)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### A. Question - Category Classification"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 648x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "model name\n",
       "DecisionTreeClassifier    0.891602\n",
       "MultinomialNB             0.957149\n",
       "RandomForestClassifier    0.738304\n",
       "SGDClassifier             0.980008\n",
       "Name: accuracy, dtype: float64"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "entry = []\n",
    "for model in models:\n",
    "    model_name = model.__class__.__name__\n",
    "    accuracies = cross_val_score(model,X_train_question,y_train_question,scoring='accuracy',cv=cross_validation)\n",
    "    for fold_index, accuracy in enumerate(accuracies):\n",
    "        entry.append((model_name,fold_index,accuracy))\n",
    "    cv_tfidf=pd.DataFrame(entry, columns=['model name','fold_idx','accuracy'])\n",
    "\n",
    "plt.figure(figsize=(9,6))\n",
    "sns.boxplot(x='model name',y='accuracy',data=cv_tfidf)\n",
    "sns.stripplot(x='model name',y='accuracy',data=cv_tfidf,\n",
    "             size=9, jitter=True, edgecolor=\"gray\",linewidth=3)\n",
    "\n",
    "plt.show()\n",
    "cv_tfidf.groupby('model name').accuracy.mean() # The accuracy mean of those models"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### **`Discussion | Question Category Classification | Model Selection`**\n",
    "- **Support Vector Machine** (SGDClassifier) performs the **best** with the highest accuracy rate of approximately **98%** on training data.\n",
    "- **Naive Bayes** performs **slightly less accurate** comparing to **Support Vector Machine** by **3%** on training data. \n",
    "- **Random Forest** and **Decision Tree** do not perform as well as other models on training data.\n",
    "#### => CONCLUSION: SUPPORT VECTOR MACHINE - 98% OF ACCURACY ON TRAINING\n",
    "\n",
    "### B. Answer - Category Classification"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAikAAAFzCAYAAAD7bpkSAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjEsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy8QZhcZAAAgAElEQVR4nO3deZhcZZn+8e9T1XvS2ZskkoQsHfaEAAFEILKICzqC4LA4o6KDjjNiWAb3hUVFUX46RmAYVMB1UBRCBGQHA7ImJJAFQhoSkk6AdPb03lX1/P44p6uru6uT7qSr63T3/bmuXDlbnXpq6aq73vOe85q7IyIiIhI1sXwXICIiIpKNQoqIiIhEkkKKiIiIRJJCioiIiESSQoqIiIhEkkKKiIiIRFJBvgvoqTFjxvjkyZPzXYaIiIj0gsWLF29294ps6/pdSJk8eTKLFi3KdxkiIiLSC8zsza7W6XCPiIiIRJJCioiIiESSQoqIiIhEkkKKiIiIRJJCioiIiESSQoqIiIhEkkKKiIiIRJJCioiIiESSQoqIiIhEkkKKiIiIRFK/uyy+yEBVV1fHli1bKC4upqKiglis/W8Id6empoaGhgZGjRpFeXl5nioVEekbCikiebZixQruuOMOnnrqKZLJJADjxo3jzDPP5JxzziEWi3HPPfcwf/581q9fD4CZcfzxx3P++ecza9asfJYvIpIz5u75rqFHZs+e7RpgUAaKe++9l+uvv55UKpV1/bRp0yguLmblypVd7uOLX/wi5513Xq5KFBHJKTNb7O6zs61TnxSRPFm6dGmngDKupJnSeDI9//rrr7cLKMWxFONLmtrt58Ybb+Qf//hH7gsWEeljOtwjkie/+93v0gFl+tAGvnHoOqYNbaQpady1YQz/U/WudttfOPltzp9UQ1lBivX1RfzglUks3zEEgN/+9reccMIJff4YRERySS0pInmwefNmnn/+eQAM55rD1zJtaCMAxXHngkk1fGDc1vT2Y0ua+OzUdygrCELNxLJmrjl8LYUWzK9cuZK1a9f27YMQEckxtaTIgDNv3jyqqqpyeh/V1dU0NDTs9e1bO8gCVA5tYP+y5k7bnLzfDh58exQApfHOfVbGFCeYMaKOF7cFZ/l87nOfo6Bg7/6kS0tLmTBhwl7dtrsqKyuZO3duTu9DRAYWhRQZcKqqqnht+YtMGprc88Z7KVkfI5W0vb590F89aMjclSjAHazD7nY0xzO2z35fO1ra/oS9pYFUYu/qSbbspHHtW3t3425YVxvf80YiIh0opMiANGlokm/Nrs13GV1qSsIXF46gMWm83VjEk5uHMadiZ3p9IgV3bRiTnt/WEqcpaRTH287Ge3HbEF6vLQWgwJyfnbiT8qJonq33vUVD812CiPRDCikieVAch5PGN/FwdQkA311xAOdNquE9o3dS01TIHesqeG1XWXr7nS2FfPHFSi6YVMO7Spt4fks5/7duv/T6Y8c2RzagiIjsLYUUkTz56JRGFtUUsa0pRlMqxm/WjuU3a8d2uf1ru8q4esUB2Vc6WQ8ZiYj0Zzq7RyRPRhY73zhqF+PKsvedMZyzpjRQXpD9Qm+Znn6nmKfeKurtEkVE8kotKSJ5NH5Iih++eyeLawp56q0itjTGKIzBwSMTnLp/Exvr4sxPBL8l4jgThzQSNxhT1MLJ+23n2S3l/L1mJAB/W1fMieOb1ZoiIgOGQopInhXE4LixLRw3tqXTurvfKElP//OkGv6zsv0ZOHMqdvLsluE0pWKsqy3grfoY7xqy55YXEZH+QCFFBpzq6mrqdsUHxBklmafuHjmi89lK5YVJKoc2sGJncOXZG5YNoawgeh1o39wVZ0h1db7LEJF+Rn1SRCIslnHoZm19Saf1iRRsaGjrixIjegFFRGRvqSVFBpwJEybQmHgr0tdJ6a75bxTz5zeCU5H/vH4MHxy3jZFFbVdsu3N9BdtbCgEYUZTiqmNrKYjgT4/vLRpKSY6vaCsiA49CikgELd9SwN1rSli1vTC9rKapiE89dxBnjN/KfiXNPLN5GM9vHZZef9qEpkgGFBGRvaWQIhIxT2wo4levlOF0Pk1nR0tBu4u4tZo2LMEZBzT2RXkiIn1GIUUkQtbujLcLKHFzZg6vo6apgOqGzn1SDOeY/Vq46NA6ijU8jogMMAopEeXuvPjii9x33328+eabxONxDjjgAEaODK6JUVJSwlFHHcURRxyB6cIYA8YD64vTAWX60AaunbmGsSXBqcnLt5dyxUvTqE8GaaRyWAv/cXg9Y8t0yrGIDEwKKRG0fft2vv3tb/PSSy+1W/7qq6+2m7/99tuZMmUKV1xxBTNmzOjLEiUHEil47p22M3WuOHh9OqAAHD6igS9Me4ufvBZ0QH1jZwFFcZ3NIyIDl0JKxDQ1NfGVr3ylUyDpypo1a7jsssu4/vrrmTVrVo6r6z/W1fa/66QkUtCSClpRhhYkOWRYQ6dtjhm1Kz2dwrjq+XIqSqPfkrKuNs6B+S5CRPodhZSIue+++9IBxYHasbOorzgUS7Yw9O0llG2rSm+bsjgxT9Lc3Mx3v/td7rjjDgoLC7vY8+BRWVmZ7xL2SjKZhGXLAKhLxNjWXNDudGNof00UgJ2pEiZOPqzPatxbB9J/XxcRyR+FlIi555570tM7Js1h54Tj0/ONI6Yw+rUFDNnyanq+eFc18UQjNTU1LFy4kNNOO63Pa46auXPn5vw+5s2bR1VV1Z437IF4PE5ZWRn19fU4xm/W7sfc6RvTY/EkUvC7N9uPktzS0kIqlSIW27dzjysrK/vkeRMR6YkBH1Jy8WXSUXV1NQ0NnZvme8rdqa+vD6Yxdo07qv0GZux61+x0SCmqr6F23FEMr34agO9973tcf/31+1RDaWkpE3J80S19IXZtzJgxrFu3DoC/VFewtq6E943dTmMyxoKNo1hTV9rpNuo4LSID1YAPKVVVVSxZtpJU2aic3Yc11mOpzoPD9Zi3XRnDYwV4vKjTJqmCjC8pT9I8pO2aGYlkitrG5n0qYVez807T2/u0j92J1W/N2b77Uq5CVjKZ5Oyzz2bbtm0ALN5WzuJt5V1uf9hhh/Hzn/88J7WIiOTbgA8pAKmyUTQe+pF8l7Fn7pQt/h2WbCKWaqFk+xs0jpzWbpOyLa+lpxMlIyneuaFtvmI6zVPn9Fm5e6Nk5b35LiHS4vE4V1xxBd/85jezro/hpDIu8vbRj360r0oTEelzuoh2lJiRqGjrXDjq9Qcp3rEO3MFTlNWsZFh4aAegadhEhr6zND2fHDWlT8uV3DjxxBM58cQTs67LDCgzZ87k9NNP76uyRET63KBoSelPWsYeTsGmVVgqQUHzLsau+D8SRUOxVJJ4oq3fSypWwLDqp9NfWamS4SSHawC3gcDMuPLKK7n22mt5/PHHs24ze/Zsrr76agoK9CcsIgOXPuEixkvKaTzwfZS89giWCk4/LWjuPJpvLNV2aqrHi2mcfhqoA+WAUVxczNVXX83555/PggULWL16Ne7OlClT+MhHPqIrDYvIoKCQEkGp4RNoOPwsCt9aRsGWKiyVBILrpnT8WkoMn0DzAe/GS0f0eZ2Se4cccgiHHHJIvssQEckLhZSI8tIRNE89iebJx2NNdWCGF5YQ37ERa66DWAHJYePwkuH5LlVERCQnFFKiLlaAl7YFkeSoyfmrRUREpA/p7B4RERGJJIUUERERiSQd7ulr7sRqNxGr3wJAqnQkqfJxOjNHRESkA4WUPhTfvp6idc8Ta9jWbnmqZDjNE44mOXpqnioTERGJHoWUPlJQs5qiN/7e6RRigFjjDkqqHqOppZ7EuMP7vDYREZEoUp+UPmCNOyla82Tb1WFjhdSNOYS6isNIZQwiWPTms8TqtuSnSBERkYhRS0ofKHznFcxTADSXjmHTYeeRKhoKQKylgYpX7qS49i0MKHhnReQHCRQREekLaknpA/Fta9LT2w94bzqgAKQKS9k2+dT0fMHWNYiIiMggaEmprq4mVr+DkpX35q0Ga6pLTzcN6zwIYHP5/m3bJlsoWfHX3Z/t444lGiGVAAyPF0C8uF+cIRSr30J1dWLPG4qIyKA34ENKJJiBOwBFdZtoGj6p3erC+pr0tO9uP+5YSz3WVIt12NIthheWQiqJJZuC+7MYXlCCF5VBTC+1iIj0LwP+m2vChAm801RA46EfyVsNRWueonDTqwAMX7eQmkPPw+OFwcpUkhFv/j29bXLUFJqmn5Z9P28+S+Hby7OuM08FY/pk8hTWUo8nGmiechKJigP3/cHso5KV9zJhwrh8lyEiIv1ATkOKmX0Q+BkQB37p7j/ssP4A4FagAtgK/Ku7V+eypnxoGXsoBZtexYCSXRsYv+QW6ioOA4sxpGYlBU07MrY9LOs+Yjs2tAsoieJh1I8+GEslKdu8kniiocv7N3eK3liIF5aSHDGx1x6XiIhILuUspJhZHLgROB2oBl4wswXuvjJjs+uB37j7r83sVOAHwCdzVVO+eNkoWibMpqh6EQAFzbUM3/Bcp+2ax88gNSx7K0Ph2yvS0w0jp1Fz0McgFgdg+6STGLviDxTVbQKgbsyhbJ32QYp3rmfE2scpatiMAYXrF5EcPqFf9F0RERHJZUvKsUCVu78BYGZ3AGcCmSHlUOCycPpxYH4uConVb81rx9lWqeJhWPMuzDv0J8Hw4qHEd20inq1Od2K176Rnt00+JR1QALygmB0TT6Li1b8AUFT7Nh4vpHHkVGqGVDD+xVuIpRLE67dQsnw+tB5qyoNY/VZAh3tERGTPchlS9gfWZ8xXA8d12OYl4ByCQ0IfA8rNbLS799oVzSorK3trV12qrq6moaHrwy1pBeDxUpLJJKlUcN0UM6OgoACzFHht1p6z7k59OJ2KFZIoHd1pm+ahY9PTsURjejpZVE7jiKmUbX0NgNLELgqt65BSWlrKhAmdz0DqPeP65DUREZH+L5chJdsxhY5fwVcAN5jZhcBCYAPQ6fxUM/s88HmASZMmdVy9W3Pnzu3R9ntj3rx5VFVV5Wz/7s5LL70EQCzVQmFdDS1DKtptU7RrY3o6VVjS5b7Gjh3L6NGdQ06rysrKPnnORERE9iSXIaUayOylOQHYmLmBu28EzgYws6HAOe6+gw7c/RbgFoDZs2fv9izdfOiLL/Urr7ySxx9/HICRax+l5uCz8fCS+rHmOkasezK9bf3og9LTseZaSra/kZ7/1re+xSGHHJLzekVERPZVLkPKC8B0M5tC0EJyPvCJzA3MbAyw1d1TwNcJzvSRLM4555x0SCnZ8SbvWnwz9aMPJJZKULplNbFUMxA0VTUOfRekkhTvXM/ItY8RSwWNU9OnT+fggw/O10MQERHpkZyFFHdPmNnFwIMEpyDf6u4rzOwaYJG7LwBOBn5gZk5wuOeLuaqnv5s5cyaf+cxnuO222wCIJxoof+elTtsZMDbsQNtuuRkXXXQRpjN7RESknzD3yB092a3Zs2f7okWL8l1G3tx7773cdttt1NTUtFs+bNgwdu7cmfU28XicK664gg9/+MN9UaKIiEi3mdlid5+ddZ1CSv+TSCRYtGgRb775JmbGQQcdxMyZM1m+fDl33XUXzz33HHV1dYwYMYJTTjmFj33sYxxwwAH5LltERKQThZRByN11aEdERCJvdyEl1tfFSN9QQBERkf5OIUVEREQiSSFFREREIkkhRURERCJJIUVEREQiSSFFREREIkkhRURERCJJIUVEREQiSSFFREREIkkhRURERCJJIUVEREQiSSFFREREIkkhRURERCJJIUVEREQiSSFFREREIkkhRURERCJJIUVEREQiSSFFREREIkkhRURERCJJIUVEREQiSSFFREREIkkhRURERCJJIUVEREQiSSFFREREIkkhRURERCJJIUVEREQiSSFFREREIkkhRURERCJJIUVEREQiSSFFREREIkkhRURERCJJIUVEREQiSSFFREREIkkhRURERCJJIUVEREQiSSFFREREIkkhRURERCJJIUVEREQiSSFFREREIkkhRURERCJJIUVEREQiSSFFREREIkkhRURERCJJIUVEREQiSSFFREREIkkhRURERCJJIUVEREQiSSFFREREIkkhRURERCJJIUVEREQiSSFFREREIkkhRURERCJJIUVEREQiSSFFREREIkkhRURERCIppyHFzD5oZqvMrMrMvpZl/SQze9zMlpjZy2Z2Ri7rERERkf4jZyHFzOLAjcCHgEOBC8zs0A6bfQv4k7sfCZwP3JSrekRERKR/yWVLyrFAlbu/4e7NwB3AmR22cWBYOD0c2JjDekRERKQfyWVI2R9YnzFfHS7LdBXwr2ZWDdwPfCnbjszs82a2yMwW1dTU5KJWERERiZhchhTLssw7zF8A3O7uE4AzgN+aWaea3P0Wd5/t7rMrKipyUKqIiIhETS5DSjUwMWN+Ap0P5/wb8CcAd38GKAHG5LAmERER6SdyGVJeAKab2RQzKyLoGLugwzbrgNMAzOwQgpCi4zkiIiKSu5Di7gngYuBB4BWCs3hWmNk1ZvbRcLP/Aj5nZi8B/wdc6O4dDwmJiIjIIFSQy527+/0EHWIzl30nY3olcEIuaxAREZH+SVecFRERkUhSSBEREZFIUkgRERGRSFJIERERkUhSSBEREZFIUkgRERGRSFJIERERkUhSSBEREZFIUkgRERGRSFJIERERkUhSSBEREZFI6tbYPWb2F+BW4G/unsptSSIiA19LSwtPPfUUy5Yto7m5mXHjxnH66aczduzYfJcmEhndHWDwf4DPAPPM7E7gdnd/NXdliYgMXAsXLuSnP/0pW7Zsabf8l7/8Je9///u5/PLLKSkpyVN1ItHRrcM97v6Iu/8LcBSwFnjYzJ42s8+YWWEuCxQRGUgeffRRvv3tb3cKKACpVIoHHniAr371qzQ3N+ehOpFo6XafFDMbDVwIXAQsAX5GEFoezkllIiIDzK5du/jRj36EuwOQKknROLOR+uPqaRnfkt5uyZIlzJ8/P19likRGt0KKmd0FPAmUAf/k7h919z+6+5eAobksUERkoHjggQdoaGgAIFmeZOc5O2k4poGmw5uoPaOWhlkN6W3vvvtuUil1AZTBrbt9Um5w98eyrXD32b1Yj4hIZM2bN4+qqqq9vv3rr7+enm6c0YiXeLv1jUc0UrKiBGsxNmzYwBe+8IWc9U2prKxk7ty5Odm3SG/p7uGeQ8xsROuMmY00s//MUU0iIgNSZstIakSWVpKCoIUl2/Yig5G1Hhvd7UZmS919VodlS9z9yJxV1oXZs2f7okWL+vpuRUT22Te/+U2efPJJABqObKDxqMZ2663BGH7HcCxlANx55506JVkGPDNb3NVRme62pMTMzDJ2GAeKeqM4EZHB4tRTT01PlywvIb4p3rayBcr+UZYOKIcddpgCigx63e2T8iDwJzO7GXDgC8ADOatKRGQA2bZtG08//TRbtmxh2LBh7Ny5E2sxhv11GC3jWvBSp2BjAbGmtt+N5557bh4rFomG7oaUrwL/DvwHYMBDwC9zVZSIyEBQW1vLDTfcwMMPP0xLS0vWbQrf7nypqYqKCk4++eQcVycSfd0KKeGl8P8n/CciIntQW1vLJZdcwurVq7t9m/HjxxOPxxk9ejQZR9hFBq3ujt0zHfgBcCiQPh/O3afmqC4RkX7t5ptvbhdQEhUJEmMTxLfGKdzY1noyZcoUzj//fMaPH8/MmTO59NJL81GuSCR193DPbcCVwE+BUwjG8VHMFxHJYteuXTzwQFu3vfrj62k6tCk9X7C+gKEPDcUw1qxZw8EHH8yUKVPyUapIpHX37J5Sd3+U4JTlN939KuDUPdxGRGRQev7559Nj7yRGJWg6pKnd+sTEBC2T2/qoLFy4sE/rE+kvutuS0mhmMWC1mV0MbAD2y11ZIiI9s69Xg+1NNTU16elkRTJru3OiIkHR2uBKDgsWLGDx4sUA6UNEA+FqsLqqreyr7oaUSwnG7ZkLfJfgkM+nc1WUiEhPVVVVsWTFEhix521zrhEsTCbxLfGsmxRsbvv43VS3iU0bNgUz4UVml2xYktMSc257vguQgWCPISW8cNu57v5loJagP4qISPSMgNTJEbiUfBPE7o1hKaNgcwFFq4tont6cXl3wVgGFa9s6z6ZOSEUjXPWi2BPd7U0g0rU9hhR3T5rZ0WZm3p1r6IuIDHbF4JMcWxu0pgxZOISi1UVtZ/esK0y3tPgYH3ABRaS3dPdwzxLgHjO7E6hrXejud+WkKhGRfs6PcHyrYzuDMFL4ViGFb7W/cJsXO6ljItDyI3lRW1vLG2+8QSqVYv/996eioiLfJUVOd0PKKGAL7c/ocUAhRUQkm6Lg0JO9aFi1pVtOWvl+TuroFAzNU32SN2+99Ra//vWveeSRR9JngQEcd9xxfPKTn2TmzJl5rC5aunvFWfVDERHpqWLw4x2vd2yDQRNQCD7eYdgebpsAmoPt6XzlfOmnVq9ezeWXX86OHTs6rXvuuedYtGgRX//613n/+9+fh+qip7tXnL2NoOWkHXf/bK9XJCIy0JSBT+9ml75NEHstBm/Rrt9KanoK9keX0ezHGhoa+NrXvtYuoIxNJChxZ11BAW5GMpnkBz/4AVOnTqWysjKP1UZDdw/33JsxXQJ8DNjY++WIiAxettyIvdL5rBjbbMQ3x0kdkMKPcQWVfurRRx9NX0OnJJXiCzt2cnBLCwa8E49z8/BhbCwoIJlM8qc//YlvfOMb+S04Arp1jpi7/yXj3++Bc4HDc1uaiMjgYWs6B5RUSQrPaMSOvRnDliuh9FcPPfRQevojdfUcEgYUgLHJJJ/auSu9/rHHHmvXX2Ww6m5LSkfTgUm9WYiIyKDlYCvbwkfLu1qof089qeEprM4oXVRKcVUxALba8INd/VT6SG9eyXjlypXp6RlZAsiURIIhqRR1sRjNzc3MnTuXoqKiXrnvVv3tKsDd7ZOyi/Z9Ut4GvpqTikREBptNYPVBSEkVp6h9X206hPgQp35OPQWbC4hvj2NJw9YbPlWXrepvYrG2lrKt8Rjjksl26+vNaDDLuv1g1d2ze8pzXYiIyL6orq6GHf30SqcNbZMtE1s6t5IYNE9tpvTF0mB2pWHrIn7YZztUe3W+q9hne9vqsG3bNp5++mm2bdtGeXk5xx57LHfccQd33303AA+UlXFQ8w4yB014qKyUVBhSpk2bxk033bSv5fd73W1J+RjwmLvvCOdHACe7+/xcFiciMthYS/bwYc0RDyUCBBdou+GGG3jooYdIJBLp5WbGrFmz0vOrioq4buQI5jQ0UuopXiguYUlJcXr9WWed1ad1R1V3+6Rc6e53t864+3YzuxJQSBGRSJgwYQI1VhONsXt6ajvEHw5+UxeuLyS2M0ZqWMbjaIaiqra+CalZKZjQ10X2TOyJGBP2j3iRvay2tpZLLrkkPZJ1JndnyZIllJWVUV9fD8CbhYX8trBz56IjjjiCM844I+f19gfdbRfNtt3edroVEZFMI8BHBX1MLGWU319O0aoiYjtiFL5RyLB7hxFrDD6GvcThXfksVrpy8803twsoU1paeF99PQdndJKtr69n/PjxxOPZR8c+6aSTuO666yjMEl4Go+4GjUVm9hPgRoIOtF8CFuesKhGRQSZ1RIrYEzHMjVhdjCFPDcm6nR/h3f95KX1m165dPPDAA+n5T+zaxXsbGtPzy4uK+PmI4UBwWfyf/exnvPzyy7z66qskk0mqqqoYPXo03//+9/u89ijrbkj5EvBt4I/h/EPAt3JSkYjIYDQGUiekiD0bwxKd+5+4OX6U45N0Vk8UPffcc+nrmkxoSTAnI6AAHN7czKzGJpaG/U6WLVvGpz/96fT6/nRacF/q7tk9dcDXclyLiMjgNh5SZ6SwtcGghJlj/fhUh7J8FxgtvXkNk33VeiVZgKmJlqwXBZ6aaGEpQUi55557WLRoUXpd62GigRBWevNaLN09u+dh4J/dfXs4PxK4w90/0CtViIhIoBj8IMcPUovJnlRVVfHq0qWMy3chBHmy1dqC7F+tawva+pm01NSwPSPYtB7B2750aQ6q6ztv9/L+unu4Z0xrQAFw921mtl8v1yIiItIj44B/i8BgRnU41xMMXr2usJBnSoo5vrEtuqwqLGRJcdsZWp8Axkeg7t72q85jEe+T7oaUlJlNcvd1AGY2mSyjIouIiAxGQzBm4rwYzt8+bBjPlDRT2dJCdUEBLxcV4eGF2iYzMANKLnQ3pHwTeMrM/h7OzwE+n5uSRERE+p8PAOuB1oM4q4qKWNVh7J2hwMf6uK7+rLujID8AzAZWEZzh81+0u5CziIjI4FaGcRFwOGRtJ5lG8Ot+lFpRuq27HWcvAi4huMbhUuDdwDPAqbkrTUREpH8pwzgP2IHzClALlAIHAhUKJz3W3cM9lwDHAM+6+ylmdjBwde7KEhER6b+GY7w730UMAN29bmGjuzcCmFmxu78KHJS7skRERGSw625LSnU48vF84GEz2wZszF1ZIiIiMth194qzrZ2RrzKzx4HhwAO7uYmIiIjsgyTOq8AaoAUYBsxkcPVt6fFIxu7+9z1vJSIiInvrNZx7gJ0dlj8BHILzMaB0EISVnI6laWYfNLNVZlZlZp3G/jGzn5rZ0vDfa2a2Pdt+REREBotXcX5H54DS6hXgNqBpEFxTtcctKd1lZnHgRuB0oBp4wcwWuPvK1m3c/bKM7b8EHJmrekRERKKuBedu2i7pPiSV4sSGRkankiwvKuLl4mCAwreAhQRfsANZzkIKcCxQ5e5vAJjZHcCZwMoutr8AuDKH9YiIiETaMqA+nB6WTPKtbdsYngoiy3sbGnmstJQ/lg8FYDFwCk7BAD7sk8uQsj/BFYJbVQPHZdvQzA4ApgCP5bAeEREZQKqrq9lF7w9ql081GdOnNjSkA0qrkxsaeLCslO3xOHXA/wIlEXr8bwG11dW9tr9c9knJFu26eibPB/7s7smsOzL7vJktMrNFNTU12TYRERHp91IZ0+MTnb8SY8B+ybbl0YknuZHLlpRqYGLG/AS6vrbK+cAXu9qRu98C3AIwe/bsgf6aiIhIN0yYMJq3iZIAAB2NSURBVIHtmzfzbwPocMd8nMXh9KtFhcxqbm63vsGMNwsK0/MfB8ZF6PH/CmfEhAm9tr9ctqS8AEw3sylmVkQQRBZ03MjMDgJGEowFJCIiMmgdnjH9ZGkpK4vaAkkz8IfyoTTFglBSAYzt0+r6Xs5aUtw9YWYXAw8CceBWd19hZtcAi9y9NbBcANzh7mohERGRQW0qQfB4B0iY8bMRIzigpYXRyRSrigqpi7W1LZwAWIRaUXIhl4d7cPf7gfs7LPtOh/mrclmDiIhIfxHDuADnlwQjKAO8WVjIm4XttzsaOKqPa8uHnF7MTURERHpmNMa/AzMIDkNkGg58mOB6HgO9FQVy3JIiIiIiPTcC41ygFmcdbWP3HEDQ2jJYKKSIiIhE1FCMQ/NdRB7pcI+IiIhEkkKKiIiIRJJCioiIiESSQoqIiIhEkkKKiIiIRJJCioiIiESSQoqIiIhEkkKKiIiIRJJCioiIiESSQoqIiIhEkkKKiIiIRJJCioiIiESSQoqIiIhEkkKKiIiIRJJCioiIiESSQoqIiIhEkkKKiIiIRJJCioiIiESSQoqIiIhEkkKKiIiIRJJCioiIiESSQoqIiIhEkkKKiIiIRJJCioiIiESSQoqIiIhEkkKKiIiIRJJCioiIiESSQoqIiIhEkkKKiIiIRFJBvgsQERk0WsDWGWwFHCgHn+xQmu/CRKJJIUVEJNccbLVhyw1LWvtVKxyf6vgsV9u2SAf6kxARyTF7xYi9FOsUUADMjdjrMWLPxoLWFRFJU0uKiEgu7QRb0RZOksOTNB3YBAVQ9EYRBe8EH8O2wbBqwycqqYi0UkgREckhe90wgpCSGJtg14d2QTxY13RIE2X/KKN4VXGwbZVCikgmhRQRGTi2Q+yJiB3F3to22TCrIR1QADBoOKqhLaRsNmJ/jYEBQ/u0yt63Hdg/30VIf6eQIiIDQmVlZb5LyGrF9hW0JFsASA1PdVrvZY4XOtYStLYMiQ3BzJi+//Q+rbPX7R/d10T6D4UUERkQ5s6dm+8Ssvrc5z7HqlWrACjYWEDzQc3t1sdr4umAUlRUxPTp0zEz5s2b1+e1ikRNxNpFRUQGltNOOy09Xbq4lNjWto9dqzfKni5Lz5988smYdT4DSGSwUkgREcmhM844g/LycgBiDTGG3T2MofcOZeiDQxn+x+EUbA7P7jHjn//5n/NZqkjkKKSIiOTQsGHDuOaaayguDjvHYhS+U0hhdSGWams1ueyyyzjooIPyVaZIJCmkiIjk2NFHH81NN93Ecccd1+lwzsEHH8x1113HWWedlafqRKJLHWdFRPrA9OnT+fGPf8zbb7/Na6+9RiqVYuLEiUybNi3fpYlElkKKiEgfGjduHOPGjct3GSL9gg73iIiISCQppIiIiEgkKaSIiIhIJCmkiIiISCQppIiIiEgkKaSIiIhIJCmkiIiISCQppIiIiEgkKaSIiIhIJCmkiIiISCQppIiIiEgk5TSkmNkHzWyVmVWZ2de62OZcM1tpZivM7A+5rEdERET6j5wNMGhmceBG4HSgGnjBzBa4+8qMbaYDXwdOcPdtZrZfruoRERGR/iWXLSnHAlXu/oa7NwN3AGd22OZzwI3uvg3A3TflsB4RERHpR3IZUvYH1mfMV4fLMh0IHGhm/zCzZ83sg9l2ZGafN7NFZraopqYmR+WKiIhIlOQypFiWZd5hvgCYDpwMXAD80sxGdLqR+y3uPtvdZ1dUVPR6oSIiIhI9uQwp1cDEjPkJwMYs29zj7i3uvgZYRRBaREREZJDLZUh5AZhuZlPMrAg4H1jQYZv5wCkAZjaG4PDPGzmsSURERPqJnIUUd08AFwMPAq8Af3L3FWZ2jZl9NNzsQWCLma0EHge+7O5bclWTiIiI9B85OwUZwN3vB+7vsOw7GdMOXB7+ExEREUnLaUgRERGR9hpwVgLbCL6EDwAmA5b1fJPBTSFFRESkDyRxHgGeA1o6rKsAPoIzVUGlHY3dIyIikmMpnD8CT9E5oADUAL8GVnW6UsfgppAiIiKSY4sJziBp9a5Egg/V1XFCQwPFqSCYpIC/AE0KKmk63CMiIpJDjvNsxvx76xu4oLY2fWDnI3X1XDdyBNvjcRqAlwjGlRGFFBER6cfeBn4V8ZaHFqB1YLpCdz5WV9eu58moVIrT6xu4s3woAI8AyyL+mLryNtDpsvH7QCFFRET6pcrKynyX0C319fXw2msAjE0kKfXOAeSARCI9bWVljDjwwD6rrzeNoHdfF4UUERHpl+bOnZvvErrl7bff5txzzw2mC+LUmTGkQ1B5o7Dt63jWrFn88Ic/7NMao0odZ0VERHJo7NixHBi2jCTM+NPQoSQz1r8Tj/NQWVl6fs6cOX1cYXSpJUVERCSHzIyPf/zjXHvttQA8W1rC6qJCDm9qZlcsxkvFRSQt6KUycuRITj311HyWGylqSREREcmxD3zgA3zgAx9Iz2+Jx/l7WSkvlhSnA4qZcdVVV1FSUpKvMiNHIUVERCTHzIyvf/3rXHTRRQwbNqzT+rKyMiorKznyyCPzUF106XCPiIhIH4jFYnzqU5/ivPPO45lnnuHtt9+msLCQGTNmcMMNN+S7vEhSSBEREelDxcXFnHzyyfkuo1/Q4R4RERGJJIUUERERiSSFFBEREYkkhRQRERGJJIUUERERiSSFFBEREYkkhRQRERGJJIUUERERiSRdzE1EJAd27tzJk08+yebNmykpKeHoo4+msrIy32WJ9CsKKSIivaipqYmbbrqJ++67j+bm5nbrZsyYweWXX860adPyVJ1I/6LDPSIivaS5uZmvfOUr3H333Z0CCsCyZcu4+OKLWbVqVR6qE+l/FFJERHrJH/7wB5YsWZKeT4xK0DijkebJzbg5AHV1dVxzzTWkUql8lSnSb+hwj4hIL2hpaWH+/Pnp+YYjG2g8shEsmI9viVN+bzmWMNavX8/zzz/Pu9/97jxVK9I/KKSIiHTTvHnzqKqqyrqutraWrVu3ApAqTdE4qy2gACRHJ2k6qImSFSUAXHfddUycOLHTflavXg3A3Llze7n69iorK3N+HyL7SiFFRKQXJJPJtunRyawH0xMVibbpRKLzBkBpaWmv1ybSXymkiIh00+5aHl5++WUuvvhiIDi0Q4pOQaWgpu0jd86cOXz5y1/ORZkiA4Y6zoqI9IJDDjmEUaNGARBriFGypAS8bX18c5ziVcXp+Tlz5vR1iSL9jkKKiEgvKCws5KyzzkrPly4tpXx+OaXPlzLk0SGULwg6zQJMnDiRY445Jl+livQbCikiIr3kE5/4BEceeWR6vmBrASXLSihaW4R5EFCGDBnClVdeSSymj1+RPdFfiYhILykqKuJHP/oR55xzDsXFxZ3Wz5w5kxtvvJEDDzwwD9WJ9D/qOCsi0ouKi4u55JJL+OxnP8uTTz5JTU0NpaWlHH300bocvkgPKaSIiORAeXk5Z5xxRr7LEOnXdLhHREREIkktKSIiInmydu1aXnrpJTZt2kRRURGNjY2UlJTku6zIUEgRERHpY2vWrOG///u/2w1ICXD22Wdz9tlnc+GFF1JQoK9oPQMiIiJ9aNWqVVx66aXU1dV1WldbW8tvfvMb1q5dy9VXX008Hs9DhdGhPikiIiJ9JJFIcOWVV6YDirkzo6mJ99Y3MCZj/KeFCxdy991356vMyFBLioiISB955pln2LhxIwDFqRRXbN/BpHCwyVQt3DF0KH8vCwaZ/Mtf/sLZZ589qC/8N3gfuYiISB9buHBhevrkhsZ0QIHgC/njtbWUpFIAbNiwgddff72vS4wUtaSIiIh0Yd68eVRVVfXa/jJDx7SWlk7ri4CJiQSri4oAuPbaaykvL++1+6+srNztaN5Ro5YUERGRPpLZEXZdYed2ghZgY8ZZPYO946xaUkRERLrQ260OjzzyCNdccw0Aj5eW8u7GRiqSqfT6vw0poy7sg7Lffvtx8803D+pTkQfvIxcREeljc+bMYfTo0WzZsoW6WIyrR43imMZGRqVSLC8qYm1hYXrbs846a1AHFNDhHhERkT5TVFTEd77zHYrCPictZjxdWsq9Q4a0CyhHHnkk5557br7KjAyFFBERkT505JFH8tOf/pSpU6d2WldYWMiZZ57Jj370o3SQGcwGdzuSiIhIHsyYMYPbbruNZcuWsWTJEhobG9lvv/045ZRTGDFiRL7LiwyFFBERkTwwM2bOnMnMmTPzXUpk6XCPiIiIRJJCioiIiESSQoqIiIhEkkKKiIiIRJJCioiIiESSQoqIiIhEUk5Dipl90MxWmVmVmX0ty/oLzazGzJaG/y7KZT0iIiLSf+TsOilmFgduBE4HqoEXzGyBu6/ssOkf3f3iXNUhIiIi/VMuW1KOBarc/Q13bwbuAM7M4f2JiIjIAJLLkLI/sD5jvjpc1tE5Zvaymf3ZzCZm25GZfd7MFpnZopqamlzUKiIiIhGTy5BiWZZ5h/m/ApPdfSbwCPDrbDty91vcfba7z66oqOjlMkVERCSKchlSqoHMlpEJwMbMDdx9i7s3hbO/AI7OYT0iIiLSj+RygMEXgOlmNgXYAJwPfCJzAzMb7+5vhbMfBV7Z004XL1682cze7O1i+4kxwOZ8FyF9Sq/54KTXfXAarK/7AV2tyFlIcfeEmV0MPAjEgVvdfYWZXQMscvcFwFwz+yiQALYCF3Zjv4P2eI+ZLXL32fmuQ/qOXvPBSa/74KTXvTNz79hNRKJKb+DBR6/54KTXfXDS696ZrjgrIiIikaSQ0r/cku8CpM/pNR+c9LoPTnrdO9DhHhEREYkktaSIiIhIJCmk7IaZuZn9NmO+IBwQ8d5u3LY2/H+ymX0iY/lsM5uXm4rT9/HRbAM6dtjmQjO7IZy+yszqzWy/jPW1GdPJcADIl8zsRTN7T+6qjxYz+6aZrQivirzUzI4L3wfXmtnqjMExv5lxm9bna0X4nF1uZrGM9cea2cJw8M1XzeyXZlaW+Zr0Uu33m9mIcHqumb1iZr/vzvujP9jT89zDfV1jZu/bzfovmNmn9mK/H8h4j9SGr/lSM/vN3tSZZf/DzOwXZvZ6+Dw8YWbHhO/R7b1xH+H9fNHM/iWcPjR8vpeY2TQze7K37qeL+259nZeb2V9b39O9sN/JZra8l/Z1u5mtyXit5/bGfru4r5M7fgab2afC52eFma00sysy6vp4L93vu8zszxnz/xd+Ll62p7+ffZHL66QMBHXA4WZW6u4NBIMlbujhPiYTXB/mDwDuvghY1JtFdhSe3r2ghzfbDPwX8NUs6xrcfRYEH7rAD4D37lOR/YCZHQ98BDjK3ZvMbAxQBHwPGAfMcPdGMysneO5aZT5f+xG89sOBK81sLHAncL67P2NmBpwDlPd2/e5+RsbsfwIfcvc14Xy33x9mVuDuiV4trnd0+Tz3dEfu/p09rL95bwp09wcJLsOAmT0BXBF+BrSzD8/xrQTXl6p0dzezSmD63tS6O+5+Y8bs2cCf3f274fxJ3d1P+H43d0/14O4zX+dfA18Evt+D2/eVL7v7n/e8WXtmFnf3ZA9ucjJQCzwd3v5DwKXA+919o5mVAJ/saR174u4bgY+H9zkOeI+7d3l9k93pyftdLSl79jfgw+H0BcD/ta4IWyCuyJhfbmaTO9z+h8BJYbq+LEzB92bc/tbw188bmek7/FW4PPx3abhscsYv7+Xhr+L3mdk/LPhVf2y4XWYryT+Z2XPhr55Hwi/JbG4FzjOzUXt4PoYB2/awzUAxHtjcelVkd98MbAc+B3zJ3RvD5bvc/apsO3D3TcDngYvDD+gvAr9292fC9e7uf3b3dzJv19XrZmbvzfi1tsTMys1svAUtM62/Nk8Kt11rZmPM7GZgKrAgfA9mvj8qzOwvZvZC+O+EcPlVZnaLmT0E9Mqv/lzq+DybWdzMfhw+ppfN7N9btzWzr5jZMgtaA34YLkv/4jSzH4a/Rl82s+vDZem/dTObZWbPhuvvNrOR4fInzOw6M3vezF5rfR26YmYXmdkd4efB38JlXwtv/7KZfSdj20+Hy5ea2U1mFjOzg4BZwJUedi509yp3/1uH+xlmZo9Z0Ar6spl9JFxebmZ/C5+H5RmP/8cZj/+6cNn3zOxSC65rdTHwhfB92a7FJlv9ZlYZ7v9m4EWCv6u99QzhGHBmNtTMHg0f1zIzOzNcPtmCVsNfWNCy8JCZlYbrjg4f7zMEf4utdZeY2W3hfpaY2Snh8gvNbL4FLThrzOxiCz6bl4Tvgd1+XprZBeE+l7c+l+HyWgtaH54Djg/r+ruZLTazB81sfLjd3IzX4g4Lvl++AFwWvhdOAr5OEH43Arh7o7v/Ikst3wn/HpaHf9uW7T7CZdk+ZzJbnh4C9mutocPfT1eP5QkLWqD/DlzS7Vfc3fWvi38EaXUm8GegBFhKkGLvDddfFb45WrdfTjAWEUBt+H96+47z4e2fBooJrjS4BSgkGB5gGTAEGAqsAI4kaJVJADMIAuZignBhBCNMzw/3eyFwQzg9krYO0hcB/y/LNlcBVwDfAa7OrD+cToaP/VVgB3B0vl+bPnr9h4aP+zXgJoLWo5nAkj29b7Is2waMBe4Czuzidt153f4KnJBRXwFBK843w2VxoDycXguMyTKdeT9/AE4MpycBr2S8JxYDpfl+Hfbyef488K1wWTFB6+UU4EPh31xZuG5U+P/tBL8SRwGrMp77ERnPxxXh9MvAe8Ppa4D/DqefyHidzgAe6VDbE8DsjPmLgDeBkRm3uYng7zkGPAC8BzgcmA8UhNvdQtA6ezZwZxfPTQGwPZwuzHhP7AesDqfPA/4n4zbDw+duRZbH/z3g0izTmffTVf2VQAo4Zl9eZ4L39p3ABzPue1g4PQaoCu97MsHn5Kxw3Z+Af83y2v0YWB5O/xdwWzh9MLCO4DP/wnC/5UAFweffF8LtfprxPNwOrCH4vFhK8Bn9rnA/FWGtjwFnhds7cG7G6/M0UJHxutwaTm8Eirt6L4bzW4HhXTx3twMfz3yvh9O/Bf5pN/eR7XNmcsbzlZ7u8Pezu8fyBHBTT19/He7ZA3d/OUyvFwD35+Au7vPgl3qTmW0i+JA4Ebjb3esAzOwugibVBcAad18WLl8BPOrubmbLCN44HU0A/him2SKCP6SuzAOWmtn/67A8s7n1eOA3Zna4h++8gcrda83saILn/hTgj8C1mduY2WcIfhWMJmj+XN9pR+GmPbz7rl63fwA/MbPfA3e5e7WZvQDcamaFBEF1aQ/u533AoeGPKoBhFhy+AljgwWHO/qT1gbwfmGltx+OHExwGeR/Bl1E9gLtv7XD7nUAj8Eszuw9o1//MzIYTfJD/PVz0a4IvzlZ3hf8vJvvfY0cPuXtry+T7CULUknB+KHAgMAI4BlgUvk6lBCPMr+jG/iF4Tq4zsxMJwsJECw5dvgz80ILWpL+6+z/MrD7c5hfZHv8edFX/JuB1d3+hB/vKVGpmSwmez8XAwxmP61ozmxPWvD/B5ycEn5OtfweLgclZXrvfhvVC8Jn7cwB3f9WCoVcODNc97u67gF1mtoPgCxyCH5IzM+psd7gnbNl5wt1rwvnfA3MIAmcS+Eu46UEEQfTh8PWNA63DxbwM/N7M5oe32xenmNlXgDKCML4ifCzZ7iPb50x37mN3jwWCz9Ae0eGe7lkAXE/GoZ5QgvbPYcle7LspYzpJkFh3927I3D6VMZ8iex+jnxP8ap4B/PvuanT37QS/rP9zN9s8Q/CrZVAMT+DuSXd/wt2vJGjm/idgUusXubvfFga4HQR/kJ2Y2VSC13YTwQdDdwbSzPq6ufsPCX6BlwLPmtnB7r6Q4MNvA/Bb61kHzxhwvLvPCv/tH34gQ9Anq9/o8DwbwSG51sc1xd0fCpd3Ga49OE5+LMEXyFkErQE90fr32Pq3vCeZz7EB38uoudLdbw+X35qx/CAP+oOsAGbZnjsLf4ogpB0Vvlc3AyXu/gowO9zPj83sG+7eEi6bT9BX6r7uPOg91N/xcfZU64+kAwgCe+thmn8h+Bw6Olz/Dm2fb119rnb12vfmZ2539tnobf1QDFiR8bzNcPf3h+s+DNxI8Jmx2Myy3d8eP1Ms6KdyE0GrygyCAX1bn6tO95Htc2Z3+8+8q908FtiL94FCSvfcClzT2oKRYS1wFICZHUXQnNzRLnreKXIhcJYFZ3wMAT4G7G0P+uG0dfb9dDe2/wnBl2LWP77wzRonODQ1oJnZQWaW2QlxFsGhgF8BN4R/+JhZnODDM9s+KoCbCQKHAzcAnzaz4zK2+VcLOqJlyvq6mdk0d1/m7tcRHMI42MwOADZ5cBz6V4TvyW56iCB8te5/Vg9uGxlZnucHgf8IW5cwswPDv6WHgM+aWVm4fFSH/QwlaDq/n6AzYrvnw913ANusrb/JJ4G/0zseBP4trBMzmxC2eDwCnBtOY2ajzWySu68i+DX/nYz+BQeZ2T912O9wgvdHwsxOp61Px/4Eh1J+S/B3f1QYvoe5+73AZQSHmfe1/l4RPvdzgSvC17X1cbVY0Idkt504wx9hO8IWJQhCTquFrfNmdiDBoc9V+1jyc8B7LegXFidojc/2XlkFVISt1JhZoZkdFobPie7+OPAVgha1oXT+TvkB8KPWzxAzK7bOZxe1BpLN4Xu8tf9I1vvI9jnTzcec9bF087ZZ6XBPN7h7NfCzLKv+AnwqbIp8gaDvQkcvAwkze4nguN2SLNt0vL8Xzex24Plw0S/dfYl17pTbHVcBd5rZBuBZsgepzPvebGZ3E3xAtWptboUgKX/ae9Ybvb8aCvzcglMeEwTHpj9P0GryXWC5me0CGgia/TeGt2t9vgrD27V+CeDu75jZ+cD1FpyRkiL4gLyL9q4i++t2afiBnARWEnS4PB/4spm1EPSj6klLylzgRjN7meDzYCFBx7z+oMvnGfglweGBF8Mv8BqC/gAPhEFskZk1ExzC/UbGPsuBe8IAarT/O2j1aeDmMOi8AXymNx6Mu98f/gh4Nswcu4BPuPsyM7saeCT8UmkheI3Whff9E6DKzBrCx3lFh13/FvirmS0i6Li6Olx+BMHhnhTQHO5zOHCXmRUT/Ii9fF/r7+HTsKf7WBJ+lp4P/J62x9XaZ25PPkNwaLSe8Kyr0E0Er+kygvfShR6c0bcvtb5lZl8HHid4L93v7vdk2a7ZgsOS88JDUgXAfxN8n/wuXGbAT919u5n9FfhzeDjpS+HzPpbg/dHaWnRrh/vYbma/IAi1awm+ryD4wZntPr6b5XNmjx2ed/NYuntoshNdcVZEREQiSYd7REREJJIUUkRERCSSFFJEREQkkhRSREREJJIUUkRERCSSFFJEpM9YOJ7Qvm4jIoODQoqIiIhEkkKKiHTJuj/y9igLRot92YLRYWeGy0dbMArtEjP7XzIuFR5eabd1ZN//Da/Kubtaas3s+xaMYvustY0M3dWI0VeZ2a/D+19rZmeb2Y8sGJX2AWu7Gm3WUVtFJP8UUkRkTyoJrrg8k+Dy2J8gGJDtCtqu1no1wejQM8NlvwmXXwk85e5HEoyBNQnAzA4hGCH1hHDclSTtL1OezRDgWXc/guDKuJ8Llz8FvDu8jzsILu/dahrB2CRnAr8jGCxuBsFVgj8cBpWfE4xpcjTBlTq/3/2nRkRySZfFF5E96c7I2ycSDEiHuz8WtqAMJxj48Oxw+X1m1jri72kEA5q9YG0j+27aQx3NtI3Kuxg4PZze3UjffwvHdllGcAnw1gEDW2vf06itIpJHCikisifdGQU22yAn3uH/TAb82t2/3oM6WrxtHI/MUYZ/DvzE3ReY2ckE4x61q93dU2aWefvW2ltHbT2+B3WISB/R4R4R6Q2Zo8ieDGx2950dln8IGBlu/yjw8XCQxdY+LbsdxXY3ejrSd6ZeH7VVRHqPQoqI9IargNnhaMo/pC0sXA3MMbMXgfcTjNyLu68EvgU8FN7mYboxyupu7vtOM3sS2NyTG7p7M8Gw9deFo+suBd6zl3WISC/TKMgiIiISSWpJERERkUhSSBEREZFIUkgRERGRSFJIERERkUhSSBEREZFIUkgRERGRSFJIERERkUhSSBEREZFI+v+i7VbvhTJT7AAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 648x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "model name\n",
       "DecisionTreeClassifier    0.685948\n",
       "MultinomialNB             0.915608\n",
       "RandomForestClassifier    0.653011\n",
       "SGDClassifier             0.951249\n",
       "Name: accuracy, dtype: float64"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Answer model selection\n",
    "entry = []\n",
    "for model in models:\n",
    "    model_name = model.__class__.__name__\n",
    "    accuracies = cross_val_score(model,X_train_answer,y_train_answer,scoring='accuracy',cv=cross_validation)\n",
    "    for fold_index, accuracy in enumerate(accuracies):\n",
    "        entry.append((model_name,fold_index,accuracy))\n",
    "    cv_tfidf=pd.DataFrame(entry, columns=['model name','fold_idx','accuracy'])\n",
    "\n",
    "plt.figure(figsize=(9,6))\n",
    "sns.boxplot(x='model name',y='accuracy',data=cv_tfidf)\n",
    "sns.stripplot(x='model name',y='accuracy',data=cv_tfidf,\n",
    "             size=9, jitter=True, edgecolor=\"gray\",linewidth=3)\n",
    "\n",
    "plt.show()\n",
    "cv_tfidf.groupby('model name').accuracy.mean() # The accuracy mean of those models"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### **`Discussion | Answer Category Classification | Model Selection`**\n",
    "- **Support Vector Machine** (SGDClassifier) performs the **best** with the highest accuracy rate of approximately **95%** on training data.\n",
    "- **Naive Bayes** performs **slightly less accurate**, with **91%**, comparing to **Support Vector Machine** by **3%** on training data. \n",
    "- **Random Forest** and **Decision Tree** do not perform as well as other models on training data. The accuracy for those models are roughly **64%** and **68%**, repectively\n",
    "#### => CONCLUSION: SUPPORT VECTOR MACHINE - 95% OF ACCURACY ON TRAINING"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# ___\n",
    "## Step 7: Model Evaluations | Support Vector Machine \n",
    "- Fit the best model and evaluate on the **testing set**\n",
    "- Confusion Matrix\n",
    "\n",
    "### A. Question | SVM{TFIDF} Evaluation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.7458\n",
      "Precision: 0.74719\n",
      "Recall: 0.7458\n",
      "F1 Score: 0.74471\n"
     ]
    }
   ],
   "source": [
    "# Question\n",
    "svm_tfidf_predictions_question = train_predict_evaluate_model(classifier=svm,\n",
    "                                                 train_features=X_train_question,\n",
    "                                                 train_labels=y_train_question,\n",
    "                                                 test_features=tfidf_test_question,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### **`Question | Confusion Matrix`**\n",
    "\n",
    "- Using `metrics.confusion_matrix` from scikitlearn to create the confusion matrix for predictions\n",
    "- Visualize confusion matrix with `heatmap` from `seaborn`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x432 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Create a confusion matrix for question\n",
    "cm_q = metrics.confusion_matrix(test_labels,svm_tfidf_predictions_question)\n",
    "# Visualize the confusion matrix with heatmap\n",
    "plt.figure(figsize=(8,6))\n",
    "sns.heatmap(cm_q,\n",
    "           annot = True,\n",
    "           fmt = \"d\",\n",
    "           center = 0,\n",
    "           xticklabels = category_id_df.category.values,\n",
    "           yticklabels = category_id_df.category.values)\n",
    "plt.ylabel('Actual')\n",
    "plt.xlabel('Predicted')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "               precision    recall  f1-score   support\n",
      "\n",
      "    LIFE_ARTS       0.63      0.66      0.65        47\n",
      "      CULTURE       0.80      0.69      0.74        64\n",
      "      SCIENCE       0.78      0.79      0.79        58\n",
      "STACKOVERFLOW       0.75      0.66      0.70       103\n",
      "   TECHNOLOGY       0.75      0.81      0.78       204\n",
      "\n",
      "     accuracy                           0.75       476\n",
      "    macro avg       0.74      0.72      0.73       476\n",
      " weighted avg       0.75      0.75      0.74       476\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# Evaluate each category\n",
    "print(metrics.classification_report(test_labels, svm_tfidf_predictions_question, target_names=category_id_df.category.values))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 45,   9,   5,   0,   5],\n",
       "       [  5,  32,   1,   0,   9],\n",
       "       [  2,   3,  45,   2,   6],\n",
       "       [  1,   0,   1,  68,  33],\n",
       "       [  3,   5,   6,  22, 168]], dtype=int64)"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cm_q"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### **`Examples of miss classification`**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 209,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Actual Label: CULTURE\n",
      "Predicted Label: LIFE_ARTS\n",
      "Document:-\n",
      "I see questions/information online about how to get rid of corpses but I'm not sure what the consequences to having them lying around are.  I know that I can use a morgue to store them and that they will eventually be taken away in a hearse. However, I've always just left them lying around. This doesn't seem to have any obvious consequences.  Do they upset prisoners or staff (e.g. increasing the danger level beyond what death normally does)? Make them sick? Slow down pathing? Is this something that's not implemented yet or is leaving dead bodies all over the place meant to be completely fine?  The wiki's pages for Death and Morgue don't shed any light on this. All I'm finding is the how of getting rid of corpses/glitches related to it but not why it matters in the first place. All I've seen that's related is that apparently prisoners can loot guard corpses for keys, but there's no obvious consequences for prisoner corpses which are what I tend to end up with.  So why not just leave them in the corridors? \n",
      "Actual Label: CULTURE\n",
      "Predicted Label: LIFE_ARTS\n",
      "Document:-\n",
      "To experiment I started a bot game, toggled invincibility and let the bots attack me. There were two clear hit effects: a major one that tilts view up a lot and sideways and a minor one that only tilts it up.  So I tried releasing single shots exactly at the tilts: during a major one a single bullet ended up here (slightly covered by text). Actually the crosshair never moves that high so there tends to be a good amount of additional upward deviation of single shots fired.  But when single-shooting during the minor tilts the marks seemed to end up exactly at the location of the crosshair.  Now, when switching to spraying while being shot this pattern emerged. This was mostly caused by the minor tilts and, I guess, \"standard recoil\". Occasionally a major tilt caused a single bullet to lie severly outside of the depicted cone. \n",
      "Actual Label: CULTURE\n",
      "Predicted Label: SCIENCE\n",
      "Document:-\n",
      "I understand that elites with the Avenger property get stronger as the others in the pack with that property die, but how exactly do they get stronger?  I'm looking for hard numbers. \n",
      "Actual Label: LIFE_ARTS\n",
      "Predicted Label: CULTURE\n",
      "Document:-\n",
      "There is a scene in 21 Jumpstreet where Dave Franco' character Eric says \"do you know what they'll do to me in prison it rhymes with grape\" and it made me curious as to why it was phrased that way.   In Knocked Up a similar joke was made \"it rhymes with smabortion\" which I believe I read was done due to censors. However the repeating setup for the joke has me curious if this time it was due to censors or if it was just considered a funny setup that bared repeating. So was it censoring or continuing a joke? \n",
      "Actual Label: TECHNOLOGY\n",
      "Predicted Label: STACKOVERFLOW\n",
      "Document:-\n",
      "what is the location of imei number for android 2.3.3 ?   my phone blacked out after I factory resetted my phone. And need to restore my imei back on my  htc droid incredible 2 \n"
     ]
    }
   ],
   "source": [
    "num = 0\n",
    "for document, label, predicted_label in zip(df_test.question_body,test_labels, svm_tfidf_predictions_question):\n",
    "    if predicted_label != label:\n",
    "        print('Actual Label:', label)\n",
    "        print('Predicted Label:', predicted_label)\n",
    "        print('Document:-')\n",
    "        print(re.sub('\\n',' ', document))\n",
    "        num += 1\n",
    "        if num == 5:\n",
    "            break"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### B. Answer | SVM{TFIDF} Evaluation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.7416\n",
      "Precision: 0.74135\n",
      "Recall: 0.7416\n",
      "F1 Score: 0.73954\n"
     ]
    }
   ],
   "source": [
    "# Answer\n",
    "svm_tfidf_predictions_answer = train_predict_evaluate_model(classifier=svm,\n",
    "                                                 train_features=X_train_answer,\n",
    "                                                 train_labels=y_train_answer,\n",
    "                                                 test_features=tfidf_test_answer,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### **`Answer | Confusion Matrix`**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x432 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Create a confusion matrix for question\n",
    "cm_a = metrics.confusion_matrix(test_labels,svm_tfidf_predictions_answer)\n",
    "# Visualize the confusion matrix with heatmap\n",
    "plt.figure(figsize=(8,6))\n",
    "sns.heatmap(cm_a,\n",
    "           annot = True,\n",
    "           fmt = \"d\",\n",
    "           center = 0,\n",
    "           xticklabels = category_id_df.category.values,\n",
    "           yticklabels = category_id_df.category.values)\n",
    "plt.ylabel('Actual')\n",
    "plt.xlabel('Predicted')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "               precision    recall  f1-score   support\n",
      "\n",
      "    LIFE_ARTS       0.68      0.81      0.74        47\n",
      "      CULTURE       0.88      0.78      0.83        64\n",
      "      SCIENCE       0.81      0.79      0.80        58\n",
      "STACKOVERFLOW       0.63      0.54      0.58       103\n",
      "   TECHNOLOGY       0.75      0.80      0.77       204\n",
      "\n",
      "     accuracy                           0.74       476\n",
      "    macro avg       0.75      0.75      0.74       476\n",
      " weighted avg       0.74      0.74      0.74       476\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# Evaluate each category\n",
    "print(metrics.classification_report(test_labels, svm_tfidf_predictions_answer, target_names=category_id_df.category.values))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 213,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 51,   9,   2,   0,   2],\n",
       "       [  6,  37,   0,   1,   3],\n",
       "       [  1,   4,  46,   1,   6],\n",
       "       [  1,   1,   3,  54,  44],\n",
       "       [  1,   4,   5,  31, 163]], dtype=int64)"
      ]
     },
     "execution_count": 213,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cm_a"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### **`Examples of Misclassification`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 214,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Actual Label: CULTURE\n",
      "Predicted Label: LIFE_ARTS\n",
      "Document:-\n",
      "There is no consequence for leaving corpses anywhere. All it does if leave a mark of failure in your path of making a fully functional prison. \n",
      "Actual Label: LIFE_ARTS\n",
      "Predicted Label: CULTURE\n",
      "Document:-\n",
      "Opinion: It was a funny(?) joke.  Although a similar setup was used in Knocked Up, the truth is similar setups have been used in plenty of comedy movies.  Whilst I understand your query regarding the censors, it seems difficult to imagine why censorship would be an issue, the main reason I can think of being:   There is a colossal amount of crass language, violence, partial nudity and drug taking in the film. Why would all this be considered acceptable but a line about rape wouldn't be?   It seems more likely to me that it was a funny line that works in the context of the scene as it seems so abhorrent that he can't bring himself to utter the actual word, instead resorting to childish rhyming (suggesting the mere thought of it makes him regress to his inner child in fear).   Whether it was a homage to his (extremely popular) line from Knocked Up is something I can find no reference to online, but solely in my own opinion, I can't see how this was a censorship issue. \n",
      "Actual Label: TECHNOLOGY\n",
      "Predicted Label: STACKOVERFLOW\n",
      "Document:-\n",
      "While I can't guarentee that this is what you are looking for (as you have not posted what your expected output should be), hopefully this will help.  Even if it's not exactly what you are looking for, you have the whole object for each term available to you so that you can manipulate the output exactly as you wish.  /** Grab the terms within the for the 'portfolio_cat' taxonomy for the current post */ $terms = wp_get_object_terms($post-&gt;ID, 'portfolio_cat'); // If you are in The Loop you can replace '$post-&gt;ID' with 'get_the_ID()'  /** Loop through each term and add them to an array */ $formatted_terms = array(); foreach($terms as $term) :      $formatted_terms[] = sprintf(         '&lt;a href=\"urlhere/%1$s\" title=\"View posts in $2%s\"&gt;%3$s&lt;/a&gt;',         $term-&gt;slug,            /** %1$s - The term slug (for the 'href' attribute)*/         esc_attr($term-&gt;name),  /** %2$s - The term name (formatted for the 'title' attribute) */         $term-&gt;name             /** %1$s - The term name (for display) */     );  endforeach;  /** Output the terms, formatted as you desire */ echo join(',', $formatted_terms);   The function that is used to achieve your desired output is wp_get_object_terms().  If you require, you can pass a 3rd parameter as an array containing arguments to order the terms and to only retrun certain fields, so please take a look at the Function Reference.  With the code above the $temrs are returned as an array of objects, and should you wish to use any more of the fields returned for each term here is how the object is formatted -  [0] =&gt; stdClass Object     (         [term_id] =&gt; 22         [name] =&gt; My Taxonomy Term         [slug] =&gt; my-taxonomy-term         [term_group] =&gt; 0         [term_taxonomy_id] =&gt; 29         [taxonomy] =&gt; portfolio_cat         [description] =&gt;          [parent] =&gt; 20         [count] =&gt; 5         [filter] =&gt; raw     )  \n",
      "Actual Label: TECHNOLOGY\n",
      "Predicted Label: STACKOVERFLOW\n",
      "Document:-\n",
      "You can use RewriteCond %{REMOTE_HOST} or RewriteCond %{REMOTE_ADDR} to set such redirects.  Try something like:   #A.B.C.D is the referrer IP@ RewriteCond %{REMOTE_HOST} ^A\\.B\\.C\\.D RewriteRule ^(.*)$ http://www.yoursite.com/landingpage.html [NC,L,U,QSA]  \n",
      "Actual Label: STACKOVERFLOW\n",
      "Predicted Label: TECHNOLOGY\n",
      "Document:-\n",
      "Use if all tables have identity field.  exec sp_MSforeachtable @command1 = 'DBCC CHECKIDENT(''?'', RESEED, 1)'   MSforeachtable is an undocumented, but extremely handy stored proc which executes a given command against all tables in your database.  To reseed ONLY tables with an identity column you can use the next script. It also makes use of sp_MSforeachtable but taking into account the correct tables.  EXEC sp_MSforeachtable ' IF (SELECT COUNT(1)      FROM INFORMATION_SCHEMA.TABLES      WHERE TABLE_TYPE = ''BASE TABLE''      AND ''[''+ TABLE_SCHEMA + ''].['' + TABLE_NAME + '']'' = ''?''      AND OBJECTPROPERTY(OBJECT_ID(TABLE_NAME), ''TableHasIdentity'') = 1) &gt; 0  BEGIN     DBCC CHECKIDENT (''?'', RESEED, 1) END'   Use Following query to get Last Identity value inserted in tables in a database.     SELECT TableName = OBJECT_NAME(OBJECT_ID) ,        ColumnName = name ,        OriginalSeed = seed_value ,        Step = increment_value ,        LastValue = last_value ,        IsNotForReplication = is_not_for_replication FROM sys.identity_columns   Refer this for more details \n"
     ]
    }
   ],
   "source": [
    "num = 0\n",
    "for document, label, predicted_label in zip(df_test.answer,test_labels, svm_tfidf_predictions_answer):\n",
    "    if predicted_label != label\n",
    "        print('Actual Label:', label)\n",
    "        print('Predicted Label:', predicted_label)\n",
    "        print('Document:-')\n",
    "        print(re.sub('\\n',' ', document))\n",
    "        num += 1\n",
    "        if num == 5:\n",
    "            break"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### C. Pair of Question/Answer Evaluation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 215,
   "metadata": {},
   "outputs": [],
   "source": [
    "test['predicted_question'] = svm_tfidf_predictions_question\n",
    "test['predicted_answer'] = svm_tfidf_predictions_answer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 216,
   "metadata": {},
   "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>qa_id</th>\n",
       "      <th>question_title</th>\n",
       "      <th>question_body</th>\n",
       "      <th>question_user_name</th>\n",
       "      <th>question_user_page</th>\n",
       "      <th>answer</th>\n",
       "      <th>answer_user_name</th>\n",
       "      <th>answer_user_page</th>\n",
       "      <th>url</th>\n",
       "      <th>category</th>\n",
       "      <th>host</th>\n",
       "      <th>predicted_question</th>\n",
       "      <th>predicted_answer</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>39</td>\n",
       "      <td>Will leaving corpses lying around upset my pri...</td>\n",
       "      <td>I see questions/information online about how t...</td>\n",
       "      <td>Dylan</td>\n",
       "      <td>https://gaming.stackexchange.com/users/64471</td>\n",
       "      <td>There is no consequence for leaving corpses an...</td>\n",
       "      <td>Nelson868</td>\n",
       "      <td>https://gaming.stackexchange.com/users/97324</td>\n",
       "      <td>http://gaming.stackexchange.com/questions/1979...</td>\n",
       "      <td>CULTURE</td>\n",
       "      <td>gaming.stackexchange.com</td>\n",
       "      <td>LIFE_ARTS</td>\n",
       "      <td>LIFE_ARTS</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>46</td>\n",
       "      <td>Url link to feature image in the portfolio</td>\n",
       "      <td>I am new to Wordpress. i have issue with Featu...</td>\n",
       "      <td>Anu</td>\n",
       "      <td>https://wordpress.stackexchange.com/users/72927</td>\n",
       "      <td>I think it is possible with custom fields.\\n\\n...</td>\n",
       "      <td>Irina</td>\n",
       "      <td>https://wordpress.stackexchange.com/users/27233</td>\n",
       "      <td>http://wordpress.stackexchange.com/questions/1...</td>\n",
       "      <td>TECHNOLOGY</td>\n",
       "      <td>wordpress.stackexchange.com</td>\n",
       "      <td>TECHNOLOGY</td>\n",
       "      <td>TECHNOLOGY</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>70</td>\n",
       "      <td>Is accuracy, recoil or bullet spread affected ...</td>\n",
       "      <td>To experiment I started a bot game, toggled in...</td>\n",
       "      <td>Konsta</td>\n",
       "      <td>https://gaming.stackexchange.com/users/37545</td>\n",
       "      <td>You do not have armour in the screenshots. Thi...</td>\n",
       "      <td>Damon Smithies</td>\n",
       "      <td>https://gaming.stackexchange.com/users/70641</td>\n",
       "      <td>http://gaming.stackexchange.com/questions/2154...</td>\n",
       "      <td>CULTURE</td>\n",
       "      <td>gaming.stackexchange.com</td>\n",
       "      <td>LIFE_ARTS</td>\n",
       "      <td>CULTURE</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>132</td>\n",
       "      <td>Suddenly got an I/O error from my external HDD</td>\n",
       "      <td>I have used my Raspberry Pi as a torrent-serve...</td>\n",
       "      <td>robbannn</td>\n",
       "      <td>https://raspberrypi.stackexchange.com/users/17341</td>\n",
       "      <td>Your Western Digital hard drive is disappearin...</td>\n",
       "      <td>HeatfanJohn</td>\n",
       "      <td>https://raspberrypi.stackexchange.com/users/1311</td>\n",
       "      <td>http://raspberrypi.stackexchange.com/questions...</td>\n",
       "      <td>TECHNOLOGY</td>\n",
       "      <td>raspberrypi.stackexchange.com</td>\n",
       "      <td>TECHNOLOGY</td>\n",
       "      <td>TECHNOLOGY</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>200</td>\n",
       "      <td>Passenger Name - Flight Booking Passenger only...</td>\n",
       "      <td>I have bought Delhi-London return flights for ...</td>\n",
       "      <td>Amit</td>\n",
       "      <td>https://travel.stackexchange.com/users/29089</td>\n",
       "      <td>I called two persons who work for Saudia (tick...</td>\n",
       "      <td>Nean Der Thal</td>\n",
       "      <td>https://travel.stackexchange.com/users/10051</td>\n",
       "      <td>http://travel.stackexchange.com/questions/4704...</td>\n",
       "      <td>CULTURE</td>\n",
       "      <td>travel.stackexchange.com</td>\n",
       "      <td>CULTURE</td>\n",
       "      <td>CULTURE</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   qa_id                                     question_title  \\\n",
       "0     39  Will leaving corpses lying around upset my pri...   \n",
       "1     46         Url link to feature image in the portfolio   \n",
       "2     70  Is accuracy, recoil or bullet spread affected ...   \n",
       "3    132     Suddenly got an I/O error from my external HDD   \n",
       "4    200  Passenger Name - Flight Booking Passenger only...   \n",
       "\n",
       "                                       question_body question_user_name  \\\n",
       "0  I see questions/information online about how t...              Dylan   \n",
       "1  I am new to Wordpress. i have issue with Featu...                Anu   \n",
       "2  To experiment I started a bot game, toggled in...             Konsta   \n",
       "3  I have used my Raspberry Pi as a torrent-serve...           robbannn   \n",
       "4  I have bought Delhi-London return flights for ...               Amit   \n",
       "\n",
       "                                  question_user_page  \\\n",
       "0       https://gaming.stackexchange.com/users/64471   \n",
       "1    https://wordpress.stackexchange.com/users/72927   \n",
       "2       https://gaming.stackexchange.com/users/37545   \n",
       "3  https://raspberrypi.stackexchange.com/users/17341   \n",
       "4       https://travel.stackexchange.com/users/29089   \n",
       "\n",
       "                                              answer answer_user_name  \\\n",
       "0  There is no consequence for leaving corpses an...        Nelson868   \n",
       "1  I think it is possible with custom fields.\\n\\n...            Irina   \n",
       "2  You do not have armour in the screenshots. Thi...   Damon Smithies   \n",
       "3  Your Western Digital hard drive is disappearin...      HeatfanJohn   \n",
       "4  I called two persons who work for Saudia (tick...    Nean Der Thal   \n",
       "\n",
       "                                   answer_user_page  \\\n",
       "0      https://gaming.stackexchange.com/users/97324   \n",
       "1   https://wordpress.stackexchange.com/users/27233   \n",
       "2      https://gaming.stackexchange.com/users/70641   \n",
       "3  https://raspberrypi.stackexchange.com/users/1311   \n",
       "4      https://travel.stackexchange.com/users/10051   \n",
       "\n",
       "                                                 url    category  \\\n",
       "0  http://gaming.stackexchange.com/questions/1979...     CULTURE   \n",
       "1  http://wordpress.stackexchange.com/questions/1...  TECHNOLOGY   \n",
       "2  http://gaming.stackexchange.com/questions/2154...     CULTURE   \n",
       "3  http://raspberrypi.stackexchange.com/questions...  TECHNOLOGY   \n",
       "4  http://travel.stackexchange.com/questions/4704...     CULTURE   \n",
       "\n",
       "                            host predicted_question predicted_answer  \n",
       "0       gaming.stackexchange.com          LIFE_ARTS        LIFE_ARTS  \n",
       "1    wordpress.stackexchange.com         TECHNOLOGY       TECHNOLOGY  \n",
       "2       gaming.stackexchange.com          LIFE_ARTS          CULTURE  \n",
       "3  raspberrypi.stackexchange.com         TECHNOLOGY       TECHNOLOGY  \n",
       "4       travel.stackexchange.com            CULTURE          CULTURE  "
      ]
     },
     "execution_count": 216,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 220,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Check to see if the pair of question and answer is predicted as the same category\n",
    "Count = []\n",
    "for i in range(0, len(test)):\n",
    "    if test['predicted_question'][i] == test['predicted_answer'][i] and test['predicted_question'][i] == test['category'][i]:\n",
    "        a = \"correct\"\n",
    "    else:\n",
    "        a = \"incorrect\"\n",
    "    Count.append(a)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 221,
   "metadata": {},
   "outputs": [],
   "source": [
    "test['qa_pair_prediction'] = Count"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 222,
   "metadata": {},
   "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>qa_id</th>\n",
       "      <th>question_title</th>\n",
       "      <th>question_body</th>\n",
       "      <th>question_user_name</th>\n",
       "      <th>question_user_page</th>\n",
       "      <th>answer</th>\n",
       "      <th>answer_user_name</th>\n",
       "      <th>answer_user_page</th>\n",
       "      <th>url</th>\n",
       "      <th>category</th>\n",
       "      <th>host</th>\n",
       "      <th>predicted_question</th>\n",
       "      <th>predicted_answer</th>\n",
       "      <th>qa_pair_prediction</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>39</td>\n",
       "      <td>Will leaving corpses lying around upset my pri...</td>\n",
       "      <td>I see questions/information online about how t...</td>\n",
       "      <td>Dylan</td>\n",
       "      <td>https://gaming.stackexchange.com/users/64471</td>\n",
       "      <td>There is no consequence for leaving corpses an...</td>\n",
       "      <td>Nelson868</td>\n",
       "      <td>https://gaming.stackexchange.com/users/97324</td>\n",
       "      <td>http://gaming.stackexchange.com/questions/1979...</td>\n",
       "      <td>CULTURE</td>\n",
       "      <td>gaming.stackexchange.com</td>\n",
       "      <td>LIFE_ARTS</td>\n",
       "      <td>LIFE_ARTS</td>\n",
       "      <td>incorrect</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>46</td>\n",
       "      <td>Url link to feature image in the portfolio</td>\n",
       "      <td>I am new to Wordpress. i have issue with Featu...</td>\n",
       "      <td>Anu</td>\n",
       "      <td>https://wordpress.stackexchange.com/users/72927</td>\n",
       "      <td>I think it is possible with custom fields.\\n\\n...</td>\n",
       "      <td>Irina</td>\n",
       "      <td>https://wordpress.stackexchange.com/users/27233</td>\n",
       "      <td>http://wordpress.stackexchange.com/questions/1...</td>\n",
       "      <td>TECHNOLOGY</td>\n",
       "      <td>wordpress.stackexchange.com</td>\n",
       "      <td>TECHNOLOGY</td>\n",
       "      <td>TECHNOLOGY</td>\n",
       "      <td>correct</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>70</td>\n",
       "      <td>Is accuracy, recoil or bullet spread affected ...</td>\n",
       "      <td>To experiment I started a bot game, toggled in...</td>\n",
       "      <td>Konsta</td>\n",
       "      <td>https://gaming.stackexchange.com/users/37545</td>\n",
       "      <td>You do not have armour in the screenshots. Thi...</td>\n",
       "      <td>Damon Smithies</td>\n",
       "      <td>https://gaming.stackexchange.com/users/70641</td>\n",
       "      <td>http://gaming.stackexchange.com/questions/2154...</td>\n",
       "      <td>CULTURE</td>\n",
       "      <td>gaming.stackexchange.com</td>\n",
       "      <td>LIFE_ARTS</td>\n",
       "      <td>CULTURE</td>\n",
       "      <td>incorrect</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>132</td>\n",
       "      <td>Suddenly got an I/O error from my external HDD</td>\n",
       "      <td>I have used my Raspberry Pi as a torrent-serve...</td>\n",
       "      <td>robbannn</td>\n",
       "      <td>https://raspberrypi.stackexchange.com/users/17341</td>\n",
       "      <td>Your Western Digital hard drive is disappearin...</td>\n",
       "      <td>HeatfanJohn</td>\n",
       "      <td>https://raspberrypi.stackexchange.com/users/1311</td>\n",
       "      <td>http://raspberrypi.stackexchange.com/questions...</td>\n",
       "      <td>TECHNOLOGY</td>\n",
       "      <td>raspberrypi.stackexchange.com</td>\n",
       "      <td>TECHNOLOGY</td>\n",
       "      <td>TECHNOLOGY</td>\n",
       "      <td>correct</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>200</td>\n",
       "      <td>Passenger Name - Flight Booking Passenger only...</td>\n",
       "      <td>I have bought Delhi-London return flights for ...</td>\n",
       "      <td>Amit</td>\n",
       "      <td>https://travel.stackexchange.com/users/29089</td>\n",
       "      <td>I called two persons who work for Saudia (tick...</td>\n",
       "      <td>Nean Der Thal</td>\n",
       "      <td>https://travel.stackexchange.com/users/10051</td>\n",
       "      <td>http://travel.stackexchange.com/questions/4704...</td>\n",
       "      <td>CULTURE</td>\n",
       "      <td>travel.stackexchange.com</td>\n",
       "      <td>CULTURE</td>\n",
       "      <td>CULTURE</td>\n",
       "      <td>correct</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   qa_id                                     question_title  \\\n",
       "0     39  Will leaving corpses lying around upset my pri...   \n",
       "1     46         Url link to feature image in the portfolio   \n",
       "2     70  Is accuracy, recoil or bullet spread affected ...   \n",
       "3    132     Suddenly got an I/O error from my external HDD   \n",
       "4    200  Passenger Name - Flight Booking Passenger only...   \n",
       "\n",
       "                                       question_body question_user_name  \\\n",
       "0  I see questions/information online about how t...              Dylan   \n",
       "1  I am new to Wordpress. i have issue with Featu...                Anu   \n",
       "2  To experiment I started a bot game, toggled in...             Konsta   \n",
       "3  I have used my Raspberry Pi as a torrent-serve...           robbannn   \n",
       "4  I have bought Delhi-London return flights for ...               Amit   \n",
       "\n",
       "                                  question_user_page  \\\n",
       "0       https://gaming.stackexchange.com/users/64471   \n",
       "1    https://wordpress.stackexchange.com/users/72927   \n",
       "2       https://gaming.stackexchange.com/users/37545   \n",
       "3  https://raspberrypi.stackexchange.com/users/17341   \n",
       "4       https://travel.stackexchange.com/users/29089   \n",
       "\n",
       "                                              answer answer_user_name  \\\n",
       "0  There is no consequence for leaving corpses an...        Nelson868   \n",
       "1  I think it is possible with custom fields.\\n\\n...            Irina   \n",
       "2  You do not have armour in the screenshots. Thi...   Damon Smithies   \n",
       "3  Your Western Digital hard drive is disappearin...      HeatfanJohn   \n",
       "4  I called two persons who work for Saudia (tick...    Nean Der Thal   \n",
       "\n",
       "                                   answer_user_page  \\\n",
       "0      https://gaming.stackexchange.com/users/97324   \n",
       "1   https://wordpress.stackexchange.com/users/27233   \n",
       "2      https://gaming.stackexchange.com/users/70641   \n",
       "3  https://raspberrypi.stackexchange.com/users/1311   \n",
       "4      https://travel.stackexchange.com/users/10051   \n",
       "\n",
       "                                                 url    category  \\\n",
       "0  http://gaming.stackexchange.com/questions/1979...     CULTURE   \n",
       "1  http://wordpress.stackexchange.com/questions/1...  TECHNOLOGY   \n",
       "2  http://gaming.stackexchange.com/questions/2154...     CULTURE   \n",
       "3  http://raspberrypi.stackexchange.com/questions...  TECHNOLOGY   \n",
       "4  http://travel.stackexchange.com/questions/4704...     CULTURE   \n",
       "\n",
       "                            host predicted_question predicted_answer  \\\n",
       "0       gaming.stackexchange.com          LIFE_ARTS        LIFE_ARTS   \n",
       "1    wordpress.stackexchange.com         TECHNOLOGY       TECHNOLOGY   \n",
       "2       gaming.stackexchange.com          LIFE_ARTS          CULTURE   \n",
       "3  raspberrypi.stackexchange.com         TECHNOLOGY       TECHNOLOGY   \n",
       "4       travel.stackexchange.com            CULTURE          CULTURE   \n",
       "\n",
       "  qa_pair_prediction  \n",
       "0          incorrect  \n",
       "1            correct  \n",
       "2          incorrect  \n",
       "3            correct  \n",
       "4            correct  "
      ]
     },
     "execution_count": 222,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 287,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x1c12c428148>"
      ]
     },
     "execution_count": 287,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x432 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(6,6))\n",
    "test.qa_pair_prediction.value_counts().plot(kind='pie',autopct = '%1.1f%%')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# ____\n",
    "## Step 8: Conclusion\n",
    "\n",
    "### `Feature`\n",
    "- TFIDF features | Stopwords Removal | Stemming | Unigram & Bigram Combination | logarithmic for frequency | minimum numbers of document a word must be present to kept is 1 \n",
    "\n",
    "### `Model`\n",
    "- Supported Vector Machine\n",
    "\n",
    "### `Evaluation`\n",
    "- **Question Accuracy:** 98% on training | 74% on testing\n",
    "- **Answer Accuracy:** 95% on training | 73% on testing\n",
    "- **Pair of Question & Answer Accuracy:** 60.1% correct on testing - combination of the answer and question predicted labels - we don't actually fit the pair of Google qa in the model for this part\n",
    "\n",
    "### `Reasoning`\n",
    "- In multiclassification problem, we still need to have a balanced data for a better model performance. However, with 5 target in the dataset, which means if we randomly pick 1 to predict the accuracy would be 100%/5=20%, which is the baseline for predictions - Therefore, the accuracy rate of 74% and 73% on testing exceed the the baseline more than 3 times. The model does perform good enough under this circumstances."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# ___\n",
    "## Step 9: Appendix\n",
    "\n",
    "### A. Feature Engineering\n",
    "\n",
    "### `BOW | Unigram`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 239,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.feature_extraction.text import CountVectorizer #tokenizes, count words\n",
    "bow_vectorizer = CountVectorizer(lowercase=True, # Lowercase\n",
    "                                min_df=1,\n",
    "                                ngram_range=(1,1)) # unigram\n",
    "# QUESTION features engineering\n",
    "bow_train_question = bow_vectorizer.fit_transform(train_question)\n",
    "bow_test_question = bow_vectorizer.transform(test_question)\n",
    "\n",
    "# ANSWER features engineering\n",
    "bow_train_answer = bow_vectorizer.fit_transform(train_answer)\n",
    "bow_test_answer = bow_vectorizer.transform(test_answer)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### `Word2Vec`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 297,
   "metadata": {},
   "outputs": [],
   "source": [
    "# tokenize documents for word2vec\n",
    "tokenized_train_question = [nltk.word_tokenize(text)\n",
    "                   for text in train_question]\n",
    "tokenized_test_question = [nltk.word_tokenize(text)\n",
    "                   for text in test_question]\n",
    "tokenized_train_answer = [nltk.word_tokenize(text)\n",
    "                   for text in train_answer] \n",
    "tokenized_test_answer = [nltk.word_tokenize(text)\n",
    "                   for text in test_answer]  \n",
    "\n",
    "# build word2vec model for question                 \n",
    "wv_model_q = gensim.models.Word2Vec(tokenized_train_question,\n",
    "                               size=200,                          #set the size or dimension for the word vectors \n",
    "                               window=60,                        #specify the length of the window of words taken as context\n",
    "                               min_count=10)                   #ignores all words with total frequency lower than                     \n",
    "\n",
    "# build word2vec model for answer                 \n",
    "wv_model_a = gensim.models.Word2Vec(tokenized_train_answer,\n",
    "                               size=200,                          #set the size or dimension for the word vectors \n",
    "                               window=60,                        #specify the length of the window of words taken as context\n",
    "                               min_count=10)                   #ignores all words with total frequency lower than                     "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 298,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\Jennie\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py:9: DeprecationWarning: Call to deprecated `__getitem__` (Method will be removed in 4.0.0, use self.wv.__getitem__() instead).\n",
      "  if __name__ == '__main__':\n"
     ]
    }
   ],
   "source": [
    "def average_word_vectors(words, model, vocabulary, num_features):\n",
    "    \n",
    "    feature_vector = np.zeros((num_features,),dtype=\"float64\")\n",
    "    nwords = 0.\n",
    "    \n",
    "    for word in words:\n",
    "        if word in vocabulary: \n",
    "            nwords = nwords + 1.\n",
    "            feature_vector = np.add(feature_vector, model[word])\n",
    "    \n",
    "    if nwords:\n",
    "        feature_vector = np.divide(feature_vector, nwords)\n",
    "        \n",
    "    return feature_vector \n",
    "   \n",
    "\n",
    "def averaged_word_vectorizer(corpus, model, num_features):\n",
    "    vocabulary = set(model.wv.index2word)\n",
    "    features = [average_word_vectors(tokenized_sentence, model, vocabulary, num_features)\n",
    "                    for tokenized_sentence in corpus]\n",
    "    return np.array(features)\n",
    "\n",
    "# averaged word vector features from word2vec\n",
    "avg_wv_trainq_features = averaged_word_vectorizer(corpus=tokenized_train_question,\n",
    "                                                 model=wv_model_q ,\n",
    "                                                 num_features=200) \n",
    "avg_wv_testq_features = averaged_word_vectorizer(corpus=tokenized_test_question,\n",
    "                                                 model=wv_model_q ,\n",
    "                                                 num_features=200)\n",
    "avg_wv_traina_features = averaged_word_vectorizer(corpus=tokenized_train_answer,\n",
    "                                                 model=wv_model_a,\n",
    "                                                 num_features=200)\n",
    "avg_wv_testa_features = averaged_word_vectorizer(corpus=tokenized_test_answer,\n",
    "                                                model=wv_model_a,\n",
    "                                                num_features=200) "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Over Sampling**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 310,
   "metadata": {},
   "outputs": [],
   "source": [
    "oversample = SMOTE()\n",
    "X, y = oversample.fit_resample(bow_train_question, train_labels)\n",
    "A, b = oversample.fit_resample(bow_train_answer,train_labels)\n",
    "w2v_q, q_labels = oversample.fit_resample(avg_wv_trainq_features, train_labels)\n",
    "w2v_a, a_labels = oversample.fit_resample(avg_wv_traina_features, train_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## B. Model Fitting & Evaluation\n",
    "\n",
    "# `BOW Feature`\n",
    "\n",
    "- Naive Bayes tend to perform the best with BOW features yet the accuracy rate is not as high as TFIDF features\n",
    "\n",
    "### Question"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 281,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 648x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "model name\n",
       "DecisionTreeClassifier    0.806712\n",
       "MultinomialNB             0.851784\n",
       "RandomForestClassifier    0.402040\n",
       "SGDClassifier             0.872676\n",
       "Name: accuracy, dtype: float64"
      ]
     },
     "execution_count": 281,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "models = [mnb, svm, dct, rf]\n",
    "from sklearn.model_selection import cross_val_score\n",
    "# Fitting the model into a plot\n",
    "CV = 5 # 5-fold cross validation\n",
    "cv_df_bow = pd.DataFrame(index=range(CV*len(models)))\n",
    "entries=[]\n",
    "for model in models:\n",
    "    model_name = model.__class__.__name__\n",
    "    accuracies = cross_val_score(model,bow_train_question,train_labels,scoring='accuracy',cv=CV)\n",
    "    for fold_idx, accuracy in enumerate(accuracies):\n",
    "        entries.append((model_name,fold_idx,accuracy))\n",
    "    cv_df_bow=pd.DataFrame(entries, columns=['model name','fold_idx','accuracy'])\n",
    "\n",
    "# Build a plots to visualize the model selections\n",
    "plt.figure(figsize=(9,6))\n",
    "sns.boxplot(x='model name',y='accuracy',data=cv_df_bow)\n",
    "sns.stripplot(x='model name',y='accuracy',data=cv_df_bow,\n",
    "             size=9, jitter=True, edgecolor=\"gray\",linewidth=3)\n",
    "\n",
    "plt.show()\n",
    "cv_df_bow.groupby('model name').accuracy.mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 283,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 648x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "model name\n",
       "DecisionTreeClassifier    0.788775\n",
       "MultinomialNB             0.838181\n",
       "RandomForestClassifier    0.582384\n",
       "SGDClassifier             0.875461\n",
       "Name: accuracy, dtype: float64"
      ]
     },
     "execution_count": 283,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Over Sampling Question BOW\n",
    "models = [mnb, svm, dct, rf]\n",
    "from sklearn.model_selection import cross_val_score\n",
    "# Fitting the model into a plot\n",
    "CV = 5 # 5-fold cross validation\n",
    "cv_df_bow = pd.DataFrame(index=range(CV*len(models)))\n",
    "entries=[]\n",
    "for model in models:\n",
    "    model_name = model.__class__.__name__\n",
    "    accuracies = cross_val_score(model,X,y,scoring='accuracy',cv=CV)\n",
    "    for fold_idx, accuracy in enumerate(accuracies):\n",
    "        entries.append((model_name,fold_idx,accuracy))\n",
    "    cv_df_bow=pd.DataFrame(entries, columns=['model name','fold_idx','accuracy'])\n",
    "\n",
    "# Build a plots to visualize the model selections\n",
    "plt.figure(figsize=(9,6))\n",
    "sns.boxplot(x='model name',y='accuracy',data=cv_df_bow)\n",
    "sns.stripplot(x='model name',y='accuracy',data=cv_df_bow,\n",
    "             size=9, jitter=True, edgecolor=\"gray\",linewidth=3)\n",
    "\n",
    "plt.show()\n",
    "cv_df_bow.groupby('model name').accuracy.mean()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Answer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 280,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAi8AAAFzCAYAAAD2cOlVAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjEsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy8QZhcZAAAgAElEQVR4nOzdeXwdddn//9d1TtY23ZuyNC1taaGyL2G5WVxZBUEWFURpvb/Ym1uxKsItiLdg9fZGbm/UCooVq8IPbkRQqAhCEQsoBZrSCrS1kC5AKHRLW5JmPSfX74+ZJCfJSXvaZpJM834+Hnlk5jOfmbnmzDlzrvOZz8yYuyMiIiISF4m+DkBERERkVyh5ERERkVhR8iIiIiKxouRFREREYkXJi4iIiMSKkhcRERGJlby+DqCnjB492idMmNDXYYiIiEgPWbx48SZ3L+1cvtckLxMmTKCioqKvwxAREZEeYmZvZCvXaSMRERGJFSUvIiIiEiuRJi9mdpaZrTSzSjO7Lsv08Wb2VzNbYmYvm9lHM6ZdH8630szOjDJOERERiY/I+ryYWRK4HTgdqAIWmdk8d1+eUe2bwP3u/jMzOwR4FJgQDl8CHArsDzxpZge5ezqqeEVERCQeomx5OR6odPfV7t4E3Aec36mOA0PD4WHAunD4fOA+d2909zVAZbg8ERERGeCiTF7GAm9ljFeFZZluAj5jZlUErS5f2oV5RUREZACKMnmxLGXeafxS4NfuXgZ8FLjbzBI5zouZzTCzCjOr2Lhx4x4HLCIiIv1flMlLFTAuY7yM9tNCrf4fcD+Auy8EioDROc6Lu89x93J3Ly8t7XIPGxEREdkLRZm8LAKmmNlEMysg6IA7r1OdN4GPAJjZ+wiSl41hvUvMrNDMJgJTgBcjjFVERERiIrKrjdw9ZWZXAY8DSWCuuy8zs1lAhbvPA74G/MLMvkpwWmi6uzuwzMzuB5YDKeCLutJIREREACzIFeKvvLzc9XiA7LZt28bWrVsZPHgwo0aNwixblyIREZH+xcwWu3t55/K95tlG0tXChQu5//77Wbx4cVvZgQceyIUXXsjZZ59NXp52v4iIxI9aXvZC7s7Pf/5z7r333m7rnHTSScyaNYuCgoJejExERCR3ankZAKqqqnjsscd4/vnnef3119vKHUgVDSevsQYLuw4999xz3HbbbVx99dV9FK2IiMjuUfKyF2hsbOTWW2/lscce6zLNgdp9j2HLxNOwdCPD3/wbQ94NTiM98sgjTJ8+nZEjR/ZyxCIiIrtPT5Xux1KpFKtWreKVV17hnXfeyVqnpaWFG2+8MWviAsHd/oa8+xIj1v4Fzytiy8SP0FiyX9vy58+fH1X4IiIikVDLS4Rmz55NZWXlLs+XTqfZuHEjmzZtIpVKtZUPGjSIMWPGMHz48LayVatWUVNT0zZeP2wCdaWHgqcoefcfFG5/F4Ah7yxme+mhNJXsR92ogymsDZKh3/72t/z973/faUyTJ09m5syZu7wtIiIiPU3JSz/T2tpSX1/fZVpdXR1r165lzJgx7LfffpgZ27dvb5tes+/RbJl4OoSXQm8fcwSlKx6geOsaAEreXUL15P1IpNqXrcumRUQkbgZs8rK7rSJRe/PNNzskLum8YlKFwyio24B5CwAbNmygqKiI4cOH09LS0lZ3W9lJbYkLAJbgvbJ/aUteira9iaUaKNnwSluVwYMH5xRXZWVl5C0vat0REZFcDNjkpbKykiWvLKdlUD/qrJpuJln3XtvolgkfpmbfYyCRJNG0nVGrHqN4yyoA3qhax5rN9STDui2JPFryuyYiqcJhbcOJVD37vnI3yeY6ANwSrNpUD5sbotumHCXqqvs6BBERiYkBm7wAtAwaScMh5/Z1GG3y31xEsm4zAHUjD6Jm/+PaprUUDGbTlHMZW/FTEi3NWEuKpgknUfTPR7F0M4mWFIXvvUXjsPEdllm0dXXbcCLdRKK+PUloPPADpEcdGPFW5aZo+SN9HYKIiMTEgE1eqqqqSNRt61dfmla/tW24fkTXpMLzimgcOo7iMCEpWPU0nsjD0s0AjFz9OBunXkSqOGhNKqhZx/A3nu26HMCLhpK/fgX561dEsCW7LlG3maqq1M4riojIgDdgk5f+qb2/SrK5NmuNZFNGuRmePxhvrseA/Ppq9lvyC5qG7I+1pCjYvqGtqgMk8vC8Ijy/GBLJzosWERGJhQGbvJSVlbG+Ma9fnTbK2/gahaufAaDk3aXU7HsMnlfUNr1oy2oK6oKExC1Jw/vOgfwikptXUVi5AMMxoLBmXYfleiKfhoPPpGXovr22LbuqaPkjlJX13/hERKT/GLDJS3+UGjmJgjdewNKN5DXVsO8/fkPN/uWkikdStHUtJe++1F531CTIDxKb9KgDaSgYTP7bS8nbVtVWx81Ij5hA09hj8EEjen17REREoqDkpZ+whvfIX78cD1tPAPIbtzJyzZNd6rbkF9NcdmzHsiH70jj1LJoaa7GGrYAFV1LlF0cfvIiISC9S8tIPJDetonD10233cdmRlqLhNBx0Gl5YknW6F5Z0O01ERGRvoOSljyW2raNwVdBfZUfSxSNpHldOengZmB5JJSIiA5eSlz5WUFXRlrg0F49ky4QP0zDsAPLrNjP8rWfbbkqXaHyP9JB9lLiIiMiAp2/CPmR11SRrW68eSrDhkE/RMOJASOTRXLIPG6deSFPx6KBuS4q8zav6MlwREZF+YUC3vCTqqiO9SZ01vIe1NHdfIeO5RA3DJpAuHNppAQm2jzmMgjcWAFDwxvMUvLUogkh3zhP5eNHQnVfcTcHjAXSptIiI7NyATV4mT54c+TqqqlJZnw7dKpVK0diYznl5+ckEhYUFPRHaLisuLo74Piz79so+ERGR+BuwyUt/eHrx6tWrmT59OgBF29aSbHyvY+uLtzB4w6tto1dddRUXXnhhL0cpIiLSv6jPSx+aNGkShx56KADmLYxZ/luKtqyClhT5tesp/efvKajfBEBhYSGnn356X4YrIiLSLwzYlpf+4oorruBrX/saLS0t5NdXM2bFA1nrfeYzn2HIkCG9HJ2IiEj/o5aXPnbsscdyww03kJ+f322diy66iMsvv7wXoxIREem/1PLSD5x++ukccsgh/OEPf+Cpp56iurqaQYMGcfzxx3PBBRdw5JFH9nWIIiIi/Ya57/jOrnFRXl7uFRUVfR2GiIiI9BAzW+zu5Z3LddpIREREYkXJi4iIiMRKpMmLmZ1lZivNrNLMrssy/YdmtjT8e83MtmZMS2dMmxdlnCIiIhIfkXXYNbMkcDtwOlAFLDKzee6+vLWOu381o/6XgKMzFlHv7kdFFZ+IiIjEU5QtL8cDle6+2t2bgPuA83dQ/1Lg/yKMR0RERPYCUV4qPRZ4K2O8CjghW0UzOwCYCDyVUVxkZhVACrjZ3R/KMt8MYAbA+PHjeyjseKmurmb58uU0Njayzz77cMghh5BIqCuTiIjsvaJMXixLWXfXZV8CPODumU8pHO/u68xsEvCUmb3i7qs6LMx9DjAHgkuleyLouHjnnXeYM2cOCxYsIJ1uf9nGjh3LZZddxjnnnINZtl0gIiISb1H+RK8CxmWMlwHruql7CZ1OGbn7uvD/amABHfvDDGhr167lyiuv5C9/+UuHxAXg7bff5pZbbuG2225jb7mHj4iISKYok5dFwBQzm2hmBQQJSperhszsYGAEsDCjbISZFYbDo4GTgeWd5x2I0uk03/zmN9myZUtbWePgfakbOYWWZGFb2e9+9zueeuqpbIsQERGJtchOG7l7ysyuAh4HksBcd19mZrOACndvTWQuBe7zjs0E7wN+bmYtBAnWzZlXKQ1kzz//PG+++SYALYk8Nk69iMbhEwCwdCOjXv8Tg6pfB+D+++/nIx/5SF+FKiIiEolIn23k7o8Cj3Yq+1an8ZuyzPcccHiUscXVX/7yl7bh2n2PbUtcADxZSPWBZ1G8ZTXmaVasWMG6devYf//9+yBSERGRaOiylJjZvHlz23DDsK5XWLXkD6Jp8Jis9UVERPYGSl5iZtCgQW3DeQ3VXSu0pMlr2Jq1voiIyN5AyUvMlJe3P1xz6LpFJJrrO0wf8k4FyVRQNnr0aA444IBejU9ERCRqSl5i5swzz2xrTclrfI/9lv6SYW8+S8m7Syld8QAj3ljQVvf8888nLy/Sbk0iIiK9TslLzJSUlHDttde2jSebtzOs6jlGrn6c4i3t9/CbOnUqn/rUp/oiRBERkUjpZ3kMfeQjHyGZTPLDH/6ww/1eWr3//e/nuuuuo6ioqA+iExERiZaSl5j64Ac/yEknncQzzzzD0qVLaWpqYp999uGMM85g3LhxO1+AiIhITNnecgv58vJyr6io6OswREREpIeY2WJ3L+9crj4vIiIiEitKXkRERCRWlLyIiIhIrCh5ERERkVjR1UYie4nm5maeffZZXnrpJRobGxk9ejSnnXYaBx54YF+HJiLSo5S8iOwFFi5cyC233NLlQZz33HMPJ5xwAjfccAPDhw/vo+hERHqWThuJxNzf//53rr/++m6fIP7CCy8wc+ZMampqejkyEZFoKHkRibHGxka+//3v09LSAsDw/GY+e8B6/mPqW3ywtP3p4mvXruWuu+7qqzBFRHqUkheRGFuwYAFbtwZJysiCZn51/Gt8/sB3OXf/amYd/gbXHvxWW90//elPNDQ09FWoIiI9Rn1eRHrY7NmzqaysjHQdVVVVAKTT6bayC8ZuYlRhqkO9c/av5q61+7C+sYDa2lq+8IUvUFJSkvN6Jk+ezMyZM3smaBGRHqKWF5EYqq+vp76+vkPyMm5QY5d6CYOyjPLW00siInGmlheRHtYbLRWt65gwYQIPPfQQAC9tKeHD+2zrUK8ulWD5e4Paxr/xjW9w0EEHRR6fiEiUlLyI9FPuzvLly9v6tZSUlHDiiSdy3HHHtdX58Ic/3Ja8/OmdUZwwqoZTRr+HGTSkjR+sLKM+nQRg/PjxTJkypU+2RUSkJyl5EemH1q1bx3e+8x2WLVvWofzBBx9k//33Z/DgwQwePJgjjzySqVOn8s9//pO0Gze8MpEDS+rZr6iJf2wdTE2q/SP+yU9+EjPr7U0REelx6vMi0s+sX7+eq666qkvi0mrdunW8/vrrrF+/HoBvf/vbjBkzpm36qtpi/rZpWIfE5dxzz+VjH/tYtIGLiPQSJS8i/cxtt93Gpk2bwjHvtt4777zDT37yE/bdd1/uuOMOPvrRj1JQUNChzv77789Xv/pVrr32WrW6iMhew9y7PzjGSXl5uVdUVPR1GCJ7ZMOGDXzqU5/qcBVRq5K8NPsVNbJmexEpb//dcc0113DeeecB8N5777Fs2TLq6+spLS3l0EMPJZHQbxQRiSczW+zu5Z3L1edFBpTeuAfLnqiurs5IXBwIWkv+deK7XDp+A4VJZ0tTHj9YOZZnNwbPKpo9ezbz58+PbcuK7iUjIrtKyYsMKJWVlbz26kuML+nastEfNDYmgGQ4FiQj5SNqmD5xfVudEQUpbnjfW3xiSwk1qTyamprYsmoxg/Li14r6Zm1y55VERDpRe7IMKFVVVfTnM6V51jW4k0u3dSkblNfCsSNq28abY3rvOff2uwWLiOQq0uTFzM4ys5VmVmlm12WZ/kMzWxr+vWZmWzOmTTOz18O/aVHGKdJfDM53Ep066W5uzM9ad3NTe3k8TxiJiOyeyE4bmVkSuB04HagCFpnZPHdf3lrH3b+aUf9LwNHh8EjgRqCc4MT/4nDeLVHFKwNDWVkZDal3+GZ57c4r95HfVhbxx7XFbeN/WjeST43fyLD89lNdr2wdxCvbBgNgONcdU8vIon7cpNSN71aUUFRW1tdhiEjMRNnycjxQ6e6r3b0JuA84fwf1LwX+Lxw+E5jv7tVhwjIfOCvCWEX6jYsmNXDoiOa28S3N+Xx+0RQerBrF4uoS7ly9L9f8Y1Lb9GNKm2OZuIiI7K4oO+yOBd7KGK8CTshW0cwOACYCT+1g3rERxCjS7+Ql4Nqja/nZskG8sL4AMN5tKOTHr2VroXAmDU1lKRcR2XtF2fKS7TR8dz8PLwEecPfWdvGc5jWzGWZWYWYVGzdu3M0wRfqfvAR86fA6LptSz45uVAfG71YN4tl1BTuoIyKyd4kyeakCxmWMlwHruql7Ce2njHKe193nuHu5u5eXlpbuYbgi/c+amiStuXwCZ0xhE0cNr+W8/TdRVtzQVu/u14pp7J9Xf4uI9LgoTxstAqaY2UTgbYIE5dOdK5nZwcAIYGFG8ePA98xsRDh+BnB9hLGK9DvvNVl42ijwv0et5tiR7R2Nt6cSTH/hYNY3FlCXSvD8uwV8YGxTX4QqItKrIkte3D1lZlcRJCJJYK67LzOzWUCFu88Lq14K3OcZzylw92oz+w5BAgQwy92ro4pVBpY3a5N8t6Kkr8PYqdpmI+1Bq8uUkroOiQvA4LwWPjZ2M3eu3g+A+1cV8+w78Tp99GZtkoP6OggRiZ1I77Dr7o8Cj3Yq+1an8Zu6mXcuMDey4GRAmjx5cl+HkLOGrVth+1oARhZm75Q7qqC9PDF4FEUHHNAbofWYg4jXPhGR/kGPB5ABJU7P0Hnttde44oorAHh562BqmpMMye/YseXZTUPbhs855xxmzJjRqzGKiPQFPR5ApJ+aMmUKEydOBKA+neTby8azqTH4vdGYNu55o5TnNg1rq3/mmWf2SZwiIr1NLS8i/ZSZ8dnPfpZZs2YB8GL1UD7x3CFMGNzA+oZ8alPtH99TTz2VA2J2ykhEZHep5UWkHzvttNOYNq390V5pN1bVFndIXA455BCuv14X44nIwKHkRaSfu/zyyznjjDPIy+vaUHrsscdyyy23UFLS/6+eEhHpKUpeRPqxpqYmvv71r/PEE0+QSnW94mjx4sX853/+J42NjX0QnYhI31DyItKPzZkzh4qKirbx4fnNHD/yPUYWtD+4ccmSJdx+++19EZ6ISJ9Q8iLST9XW1jJv3ry28csOWM+DJ6/gB0et4YGTljN9wrtt0x599FG2bdvWF2GKiPQ6JS8i/dTf/vY3GhqC5xdNHFzPjEnvkp8IbkSdl4B/nbSeg4bUAcHppWeeeabPYhUR6U26VFqkh82ePZvKyso9Xs769evbho8bWYNledb68SNreK1mEAB33XUX8+fP3+P1Zpo8eXKsbuwnIgODWl5E+qlEov3jua6+MGuddfXtzzJKJpORxyQi0h+o5UWkh/VUS8Xq1auZPn06AAs3D2VlTTEHD6lvm76qtohnNrbfYffb3/42Bx98cI+sW0SkP1PyItJPTZo0iSOPPJJ//OMfpN2Y+dKBfGz/at43tI6VNcU8/PYoUh60zhx66KFKXERkwFDyItKPXX311XzhC19g+/bt1KeT3P9WaZc6xcXFXH311X0QnYhI31CfF5F+bOLEicyePZtx48ZlnT527Fh+9KMfMWXKlF6OTESk76jlRaSfmzJlCnfffTcvvvgizzzzDO+99x5Dhgzh1FNP5YQTTlBHXREZcJS8iMRAIpHgxBNP5MQTT+zrUERE+pxOG4mIiEisKHkRERGRWFHyIiIiIrGi5EVERERiRcmLiIiIxIqSFxEREYkVJS8iIiISK0peREREJFaUvIiIiEisKHkRERGRWFHyIiIiIrGi5EVERERiJdLkxczOMrOVZlZpZtd1U+eTZrbczJaZ2b0Z5WkzWxr+zYsyThEREYmPyJ4qbWZJ4HbgdKAKWGRm89x9eUadKcD1wMnuvsXMxmQsot7dj4oqPhEREYmnKFtejgcq3X21uzcB9wHnd6rzeeB2d98C4O4bIoxHRERE9gJRJi9jgbcyxqvCskwHAQeZ2d/N7HkzOytjWpGZVYTlH8+2AjObEdap2LhxY89GLyIiIv1SZKeNAMtS5lnWPwX4IFAGPGtmh7n7VmC8u68zs0nAU2b2iruv6rAw9znAHIDy8vLOyxYREZG9UJQtL1XAuIzxMmBdljoPu3uzu68BVhIkM7j7uvD/amABcHSEsYqIiEhMRJm8LAKmmNlEMysALgE6XzX0EPAhADMbTXAaabWZjTCzwozyk4HliIiIyIAX2Wkjd0+Z2VXA40ASmOvuy8xsFlDh7vPCaWeY2XIgDVzr7pvN7CTg52bWQpBg3Zx5lZKIiIgMXOa+d3QVKS8v94qKir4OQ0RERHqImS129/LO5brDroiIiMSKkhcRERGJFSUvIiIiEitKXkRERCRWlLyIiIhIrCh5ERERkVhR8iIiIiKxouRFREREYkXJi4iIiMSKkhcRERGJFSUvIiIiEitKXkRERCRWlLyIiIhIrCh5ERERkVhR8iIiIiKxouRFREREYkXJi4iIiMRKTsmLmT1oZueYmZIdERER6VO5JiM/Az4NvG5mN5vZ1AhjEhEREelWTsmLuz/p7pcBxwBrgflm9pyZfc7M8qMMUERERCRTzqeBzGwUMB24AlgC/JggmZkfSWQiIiIiWeTlUsnMfg9MBe4GPubu74STfmtmFVEFJyIiItJZTskLcJu7P5VtgruX92A8IiIiIjuU62mj95nZ8NYRMxthZl+IKCYRERGRbuWavHze3be2jrj7FuDz0YQkIiIi0r1ck5eEmVnriJklgYJoQhIRERHpXq59Xh4H7jezOwAHrgT+HFlUIiIiIt3INXn5OvBvwL8DBjwB3BlVUCIiIiLdyfUmdS3u/jN3v9jdL3L3n7t7emfzmdlZZrbSzCrN7Lpu6nzSzJab2TIzuzejfJqZvR7+Tct9k0RERGRvlut9XqYA/w0cAhS1lrv7pB3MkwRuB04HqoBFZjbP3Zd3Wu71wMnuvsXMxoTlI4EbgXKC01SLw3m37OL2iYiIyF4m1w67vyJ4vlEK+BBwF8EN63bkeKDS3Ve7exNwH3B+pzqfB25vTUrcfUNYfiYw392rw2nzgbNyjFVERET2YrkmL8Xu/hfA3P0Nd78J+PBO5hkLvJUxXhWWZToIOMjM/m5mz5vZWbswr4iIiAxAuXbYbTCzBMFTpa8C3gbG7GQey1LmWdY/BfggUAY8a2aH5TgvZjYDmAEwfvz4nYQjIiIie4NcW16+AgwCZgLHAp8BdtaJtgoYlzFeBqzLUudhd2929zXASoJkJpd5cfc57l7u7uWlpaU5boqIiIjE2U6Tl7Dj7Sfdvdbdq9z9c+EVR8/vZNZFwBQzm2hmBcAlwLxOdR4i6EODmY0mOI20muC+MmeEjyEYAZwRlomIiMgAt9PTRu6eNrNjzczcvcupmx3MlwpPMT0OJIG57r7MzGYBFe4+j/YkZTmQBq51980AZvYdggQIYJa7V+/apomIiMjeyHLJR8zsfwlO5/wO2N5a7u6/jy60XVNeXu4VFRV9HYaIiIj0EDNb7O7lnctz7bA7EthMxyuMHOg3yYuIiIgMDDklL+7+uagDEREREclFrnfY/RVZLlV293/t8YhEREREdiDX00aPZAwXAReQ5dJlERERkajletrowcxxM/s/4MlIIhIRERHZgVxvUtfZFEC3tBUREZFel2uflxo69nl5F/h6JBGJiIiI7ECup42GRB2IiIiISC5yOm1kZheY2bCM8eFm9vHowhIRERHJLtc+Lze6+7bWEXffCtwYTUgiIiIi3cs1eclWL9fLrEVERER6TK7JS4WZ3WpmB5rZJDP7IbA4ysBEREREssk1efkS0AT8FrgfqAe+GFVQIiIiIt3J9Wqj7cB1EcciIiIislO5Xm0038yGZ4yPMLPHowtLREREJLtcTxuNDq8wAsDdtwBjoglJREREpHu5Ji8tZtb2OAAzm0CWp0yLiIiIRC3Xy51vAP5mZk+H4+8HZkQTkoiIiEj3cu2w+2czKydIWJYCDxNccSQiIiLSq3J9MOMVwJeBMoLk5URgIfDh6EITERER6SrXPi9fBo4D3nD3DwFHAxsji0pERESkG7kmLw3u3gBgZoXu/k/g4OjCEhEREcku1w67VeF9Xh4C5pvZFmBddGGJiIiIZJdrh90LwsGbzOyvwDDgz5FFJSIiItKNXX4ytLs/vfNaIiIiItHItc+LiIiISL+g5EVERERiRcmLiIiIxIqSFxEREYmVSJMXMzvLzFaaWaWZXZdl+nQz22hmS8O/KzKmpTPK50UZp4iIiMTHLl9tlCszSwK3A6cDVcAiM5vn7ss7Vf2tu1+VZRH17n5UVPGJiIhIPEXZ8nI8UOnuq929CbgPOD/C9YmIiMgAEGXyMhZ4K2O8Kizr7CIze9nMHjCzcRnlRWZWYWbPm9nHs63AzGaEdSo2btSjlkRERAaCKJMXy1Lmncb/CExw9yOAJ4HfZEwb7+7lwKeBH5nZgV0W5j7H3cvdvby0tLSn4hYREZF+LMrkpQrIbEkpo9PzkNx9s7s3hqO/AI7NmLYu/L8aWEDwJGsREREZ4KJMXhYBU8xsopkVAJcAHa4aMrP9MkbPA1aE5SPMrDAcHg2cDHTu6CsiIiIDUGRXG7l7ysyuAh4HksBcd19mZrOACnefB8w0s/OAFFANTA9nfx/wczNrIUiwbs5ylZKIiIgMQObeuRtKPJWXl3tFRUVfhyEiIiI9xMwWh/1fO9AddkVERCRWlLyIiIhIrCh5ERERkVhR8iIiIiKxouRFREREYkXJi4iIiMSKkhcRERGJFSUvIiIiEitKXkRERCRWlLyIiIhIrCh5ERERkVhR8iIiIiKxouRFREREYkXJi4iIiMSKkhcRERGJFSUvIiIiEitKXkRERCRWlLyIiIhIrCh5ERERkVhR8iIiIiKxouRFREREYkXJi4iIiMSKkhcRERGJFSUvIiIiEitKXkRERCRWlLyIiIhIrCh5ERERkVhR8iIiIiKxEmnyYmZnmdlKM6s0s+uyTJ9uZhvNbGn4d0XGtGlm9nr4Ny3KOEVERCQ+8qJasJklgduB04EqYJGZzXP35Z2q/tbdr+o070jgRqAccGBxOO+WqOIVERGReIiy5eV4oNLdV7t7E3AfcH6O854JzHf36jBhmQ+cFVGcIiIiEiNRJi9jgbcyxqvCss4uMrOXzewBMxu3i/OKiIjIABNl8mJZyrzT+B+BCe5+BPAk8JtdmBczm+YZoHUAAB8tSURBVGFmFWZWsXHjxj0KVkREROIhyuSlChiXMV4GrMus4O6b3b0xHP0FcGyu84bzz3H3cncvLy0t7bHARUREpP+KMnlZBEwxs4lmVgBcAszLrGBm+2WMngesCIcfB84wsxFmNgI4IywTERGRAS6yq43cPWVmVxEkHUlgrrsvM7NZQIW7zwNmmtl5QAqoBqaH81ab2XcIEiCAWe5eHVWsIiK9KZVKsXDhQl5//XUAJkyYwCmnnEJBQUEfRyYSD+bepStJLJWXl3tFRUVfhyEiskOPPfYYc+bMYfPmzR3Khw8fzrRp07jwwgsxy9btT2TgMbPF7l7euTyylhcREenovvvu46c//WnWaVu3buXHP/4xW7Zs4Yorgvt1trS08Nxzz/Hwww+zbNkympub2W+//Tj77LM555xzGDp0aG+GL9JvqOVFRKQXrFmzhunTp9N6zG0pbqFpUhMYFKwuIFHX3gXxiiuu4LzzzuPmm2/mueeey7q8ESNG8P3vf5+pU6f2SvwifaG7lhclLyIiveDWW2/loYceAiBVmqLm7BrIh8TWBIOeG0T+O/m7vMxhw4Zx5513ss8++/R0uCL9gk4biYhEZPbs2VRWVu6wzrJly9qG68vrIR+S1UlK/lRComnHF342Tm2k4dAGvNjJX5tP8YvFJJoSbNu2jX//93+nrKysR7YDYPLkycycObPHlicSBT1VWkSkF6TT6bbh1OgUtMDgpwa3JS6O01zWTMPUBlryW9rqNk1sou7kOlqGt+CFTtPBTWz/wPa26dXV1bS0tNcXGQjU8iIie7VcWkV6QzKZbEsy8jbkgUNyWxIAz3NqPlpDujRIcPI25ZHYFCQ1jQc1dllWalyKluIWEvUJWlpaaGpqoqioqEfirKys7JWWF7XwyJ5Q8iIie7XKykqWLFsCw/s4kDyw5uAS6OJFxaRHtLfENE5tbEtcuujuqumM8hXvrggeoNIcFiSBwh3M29e29nUAEndKXkRkr1ZVVRXcBjPKL8w0WZ6+1okHp4YMI686j+SWZNuk1D6pjosbkSZvU3B4Lni9gNTYjtPz3s5ruzrJcdgK1ilT8ZqMgHYliTGC5CdKqXC/iOwmJS8islcbPnw49fX1ka6jsbFx5/1ODNy97VJp8/aMIrk5SfOE5rbxxoMbKXy9EIDCVYV4kdNweANe6BSsLaD4heKMxWbPTDLLDcv5xneJRILCgsKc6u62gmC/iOwuJS8islebO3duX4fQwTPPPMMdd9zRoeWhcHkhTVOaaBkaJEDpMWnSw9JtfWKKlhVRtGzHfVpaClpontSM5wdXJCVr2ptPvvvd73LqqadGsDUifUP3eRER6WUtLS289NJLfOtb36K2thYATzpNk5poGdJCflV+0Kl3BwoLC2lsDDrzpoemqTm3Bi8Oj+ctMHjBYArWBM9KOvroo/nxj38c3QaJRKS7+7zoUmkRkV6WSCQoLy/ne9/7XtvDGC1tFL5eSPFLxR0SF8/SmWbq1KmMGzeubbzh6Ib2xAUgAXUn1LWNLlmyhO3btyOyt1DyIiLSR4466ih+9KMfMWnSpG7rZOvT8uabb3box5ManepSxwc7LYPa++G0tvCI7A3U50VEpA8ddthh/OpXv+Lll19m8eLFLF++nNWrV7Np06a2OqWlpVRXV5NOp8nPz+eMM85g5cqVvP322wDkbcyjaXhTh+UmahIdnpc0ZMiQ3tkgkV6g5EVEpI+ZGUceeSRHHnkkEFyVtHnzZurq6hgxYgSNjY1ccsklpNNpEokE06ZN47HHHmPFihUAFL1URPP+zfjg8NRRCoqfb78i6bjjjmPQoEG9vl0iUdFpIxGRfsbMGD16NOPHj2fIkCGMHj2as88+GzPj7LPPZtSoUZx77rkUFwcJSrI2ybDfDWPQgkEUP1fMsN8No+DNgrblXXzxxX21KSKRUPIiIhID06ZN44gjjmDatGlAcJ+Ub3zjGySTwSXRljYKVxVStKKow+miT3ziE5x44ol9ErNIVHSptIhIjFVUVPCTn/yENWvWdCgfOXIkl112GRdffHHON6gT6W+6u1RayYuISMy5O6+++iorVqwgnU5TVlbGiSeeSH5+fl+HJrJHukte1GFXRCTmzIzDDz+cww8/vK9DEekV6vMiIiIisaLkRURERGJFyYuIiIjEipIXERERiRUlLyIiIhIrSl5EREQkVpS8iIiISKwoeREREZFYUfIiIiIisRJp8mJmZ5nZSjOrNLPrdlDvYjNzMysPxyeYWb2ZLQ3/7ogyThEREYmPyB4PYGZJ4HbgdKAKWGRm89x9ead6Q4CZwAudFrHK3Y+KKj4RERGJpyhbXo4HKt19tbs3AfcB52ep9x3gFqAhwlhERERkLxFl8jIWeCtjvCosa2NmRwPj3P2RLPNPNLMlZva0mZ2abQVmNsPMKsysYuPGjT0WuIiIiPRfUSYvlqXM2yaaJYAfAl/LUu8dYLy7Hw1cDdxrZkO7LMx9jruXu3t5aWlpD4UtIiIi/VmUyUsVMC5jvAxYlzE+BDgMWGBma4ETgXlmVu7uje6+GcDdFwOrgIMijFVERERiIsrkZREwxcwmmlkBcAkwr3Wiu29z99HuPsHdJwDPA+e5e4WZlYYdfjGzScAUYHWEsYqIiEhMRHa1kbunzOwq4HEgCcx192VmNguocPd5O5j9/cAsM0sBaeBKd6+OKlYRERGJD3P3ndeKgfLycq+oqOjrMERERKSHmNlidy/vXK477IqIiEisKHkRERGRWFHyIiIiIrGi5EVERERiRcmLiIiIxIqSFxEREYkVJS8iIiISK0peREREJFaUvIiIiEisKHkRERGRWFHyIiIiIrGi5EVERERiRcmLiIiIxIqSFxEREYkVJS8iIiISK0peREREJFaUvIiIiEisKHkRERGRWFHyIiIiIrGi5EVERERiRcmLiIiIxIqSFxEREYkVJS8iIiISK0peRERE+rFUKkVNTQ2pVKqvQ+k38vo6ABERkb1dXV0dCxcuZNOmTRQXF3PMMcdQVlZGc3MzL774Im+//TZ5eXkceuihHHzwwWzfvp25c+fy17/+lc2bN+PuJJNJTjrpJC666CKOOeaYDstPp9MsWbKEtWvXYmYceOCBHHnkkZhZH21xtJS8iIiIRKSpqYk777yThx9+mPr6+g7Txo8fz7Zt29i2bVuH8n333ZeNGzeSTqc7lKfTaZ599lmeffZZLr30Uq688krMjPnz53PnnXfyzjvvdFn+lVdeySmnnBLNxvUhc/e+jqFHlJeXe0VFRV+HISIiAkBzczPXXXcdixYtimT5X/7yl2lqauJnP/vZDut9/etf55xzzokkhqiZ2WJ3L+9crpYXERGRCNx7770dEpd9Uine19TMhmSSFQX5eHhKJ+HOSQ0N1Jjxj8JCCMtLU2k+WF/P8JYWXiko4IWiwrZ5AObOnUtNTU3b+KCWFo5qbKQFY0lhAY2JoFvrD37wA4466ijGjh3bG5vdKyLtsGtmZ5nZSjOrNLPrdlDvYjNzMyvPKLs+nG+lmZ0ZZZwiIiI9KZVK8Yc//KFt/MztdXy7eguX1tby5W3buGbrVvLCMx8tZhzf0MiU5lRb4rJPKsU3t2zhtPp6yhsb+VxNDZfW1HZYR2bicmBTM9/bXM20mlo+V1PDf22uZv+wg286neahhx6KepN7VWQtL2aWBG4HTgeqgEVmNs/dl3eqNwSYCbyQUXYIcAlwKLA/8KSZHeTuHU8AioiIZDF79mwqKysjXUdVVVWXfiyt0uk0DQ0NAAxNpzl/+3Yyu85Obk5xcn0DTw8qBmBxYSHVyfb2hI/U11PUqVvHKQ0N/LFkMDWJru0On6itpTij/hB3Lqzdzm3DhwFw//3388gjj3S7LcXFxZSVle14g/fQ5MmTmTlzZo8sK8rTRscDle6+GsDM7gPOB5Z3qvcd4Bbgmoyy84H73L0RWGNmleHyFkYYr4iI7CUWLFjApk2b+joMAA5IpUhmKZ/U3MzTBMnL9oTRlHFKaFS662/1JDAyne6SvJg7E7NcRj2publt2N3Zvn17tzFu37498terqqoqFsnLWOCtjPEq4ITMCmZ2NDDO3R8xs2s6zft8p3n3npN1IiISqeHDh3fbKtJTGhsbaWlpyTrN3Wm9IKYqL48WuvbTeDO//St4kDt56fZlLSso4LCm5g71tyUSvJXX9WvbzahKJinrlPC82aluIkuLTea0wsLCbqf3hOHDh/fYsqJMXrJdXN7WpmVmCeCHwPRdnTdjGTOAGRBcEiYiIgJBZ9a+1NjYyMUXX8y2bdvYkkzyZHExZ2QkU+uSSf5WVNQ2fkRjEwmc54uDsqeLi5nS3MzRjU0YUGPGr4YOoSWjdaaoqKjt1NSDJSV8Yds28sNpDQYPlwxuq/vxj3+cq6++OroN7mVRJi9VwLiM8TJgXcb4EOAwYEF4E519gXlmdl4O8wLg7nOAORBcKt2TwYuIiOyuwsJCzj33XO655x4AHhxSwtLCQg5pamJDXpLFhYWkwkQk6U5lfh51lsDccTPSZvx82DDGpFIMb2lhdX5+W/1Wl112Gb/85S8BWF5YwLdGjaS8oZEWgxcLC3kvGZysMjMuuOCCXtz66EWZvCwCppjZROBtgg64n26d6O7bgNGt42a2ALjG3SvMrB6418xuJeiwOwV4McJYRUREetTll1/OkiVLWL486Oq5qiCfVQX5Xeqlzfjz4MFdygE25OWxoVNZIpHgmmuu4dxzz6WhoaEtQapOJnli8KAuy7jyyiuZOHHinm1MPxPZpdLungKuAh4HVgD3u/syM5sVtq7saN5lwP0EnXv/DHxRVxqJiEicFBcXc+utt3L22WeTTHbtsjtq1Kis/UzGjBnDCSec0GVaQUEBp5xyCvfeey/nnnsuADNmzOBLX/oSw4YN67KckSNHct1113HppZf20Bb1H7rDroiISMQ2bdrE008/zaZNmygqKuK4447jfe97H3V1dSxYsKDDs42OO+44EokEdXV1vPbaazQ1NTFmzBgmTJjQ7fIbGxt59tlnWbNmDWbGlClTOPnkk8nL0sE3Trq7w66SFxEREemXukteIr3DroiIiEhPU/IiIiIisaLkRURERGJFyYuIiIjEipIXERERiRUlLyIiIhIrSl5EREQkVpS8iIiISKwoeREREZFYUfIiIiIisaLkRURERGJFyYuIiIjEyl7zYEYz2wi80ddx9JHRwKa+DkJ6nfb7wKN9PjAN5P1+gLuXdi7ca5KXgczMKrI9dVP2btrvA4/2+cCk/d6VThuJiIhIrCh5ERERkVhR8rJ3mNPXAUif0H4feLTPBybt907U50VERERiRS0vIiIiEitKXnaTmbmZ3Z0xnmdmG83skRzmrQ3/TzCzT2eUl5vZ7GgiblvHeWZ23U7qTDez28Lhm8yszszGZEyvzRhOm9lSM/uHmb1kZidFF33/YmY3mNkyM3s5fA1OCN8H3zOz18OypWZ2Q8Y8ra/XsvA1u9rMEhnTjzezZ8xspZn908zuNLNBmfukh2J/1MyGh8MzzWyFmd2Ty/sjDnb2Ou/ismaZ2Wk7mH6lmV2+G8s9M+M9Uhvu86VmdtfuxJll+UPN7Bdmtip8HRaY2XHhe3RrT6wjXM8XzeyycPiQ8PVeYmYHmtmzPbWebtbdup9fNbM/tr6ne2C5E8zs1R5a1q/NbE3Gvp7ZE8vtZl0f7HwMNrPLw9dnmZktN7NrMuK6uIfWu7+ZPZAx/n/hcfGrO/v87K68nl7gALIdOMzMit29HjgdeHsXlzEB+DRwL4C7VwAVPRlkZ+4+D5i3i7NtAr4GfD3LtHp3PwqCgzHw38AH9ijIGDCzfwHOBY5x90YzGw0UAN8F9gUOd/cGMxtC8Nq1yny9xhDs+2HAjWa2D/A74BJ3X2hmBlwEDOnp+N39oxmjXwDOdvc14XjO7w8zy3P3VI8G1zO6fZ13dUHu/q2dTL9jdwJ098eBx8MYFwDXhMeADvbgNZ4LrAAmu7ub2WRgyu7EuiPufnvG6IXAA+7+nXD81FyXE77fzd1bdmH1mfv5N8AXgf/ahfl7y7Xu/sDOq3VkZkl3T+/CLB8EaoHnwvnPBr4CnOHu68ysCPjsrsaxM+6+Drg4XOe+wEnufsDuLCvX97taXvbMY8A54fClwP+1TghbLK7JGH/VzCZ0mv9m4NQwG/9qmDU/kjH/3PDX0urMbD38Fflq+PeVsGxCxi/1V8Nf0aeZ2d8taAU4PqyX2aryMTN7IfyV9GT45ZnNXOBTZjZyJ6/HUGDLTursLfYDNrl7I4C7bwK2Ap8HvuTuDWF5jbvflG0B7r4BmAFcFR64vwj8xt0XhtPd3R9w9/WZ83W338zsAxm/7paY2RAz28+ClpzWX6enhnXXmtloM7sDmATMC9+Dme+PUjN70MwWhX8nh+U3mdkcM3sC6JFWgih1fp3NLGlm/xNu08tm9m+tdc3sP8zsFQtaD24Oy9p+oZrZzeGv15fN7AdhWdtn3cyOMrPnw+l/MLMRYfkCM/u+mb1oZq+17ofumNkVZnZfeDx4LCy7Lpz/ZTP7VkbdaWH5UjP7qZklzOxg4CjgRg87Nrp7pbs/1mk9Q83sKQtaTV82s3PD8iFm9lj4Oryasf3/k7H93w/LvmtmXzGz84CrgCvD92WHFp5s8ZvZ5HD5dwAvEXyudtdCYGy43BIz+0u4Xa+Y2flh+QQLWhl/YUFLxBNmVhxOOzbc3oUEn8XWuIvM7FfhcpaY2YfC8ulm9pAFLT5rzOwqC47NS8L3wA6Pl2Z2abjMV1tfy7C81oLWiheAfwnjetrMFpvZ42a2X1hvZsa+uM+C75crga+G74VTgesJkuJ1AO7e4O6/yBLLt8LPw6vhZ9uyrSMsy3acyWypegIY0xpDp89Pd9uywIIW66eBL+e0t91df7vxR5DdHgE8ABQBSwmy3kfC6TeFb5rW+q8CE1rnDf+31e88Hs7/HFBIcHfFzUA+cCzwCjAYKAGWAUcTtOKkgMMJktLFBEmHAecDD4XLnQ7cFg6PoL3T9hXA/2apcxNwDfAt4NuZ8YfD6XDb/wlsA47t633TS/u/JNzu14CfErQ2HQEs2dn7JkvZFmAf4PfA+d3Ml8t++yNwckZ8eQStPjeEZUlgSDi8FhidZThzPfcCp4TD44EVGe+JxUBxX++H3XydZwDfDMsKCVo7JwJnh5+5QeG0keH/XxP8qhwJrMx47YdnvB7XhMMvAx8Ih2cBPwqHF2Tsp48CT3aKbQFQnjF+BcEdw0dkzPNTgs9zAvgzcBJwGPAQkBfWm0PQmnsh8LtuXps8YGs4nJ/xnhgDvB4Ofwr4WcY8w8LXblmW7f8u8JUsw5nr6S7+yUALcNye7GeC9/bvgLMy1j00HB4NVIbrnkBwnDwqnHY/8Jks++5/gFfD4a8BvwqHpwJvEhzzp4fLHQKUEhz/rgzr/TDjdfg1sIbgeLGU4Bi9f7ic0jDWp4CPh/Ud+GTG/nkOKM3YL3PD4XVAYXfvxXC8GhjWzWv3a+DizPd6OHw38LEdrCPbcWZCxuvVNtzp87OjbVkA/HRX9r1OG+0Bd385zHYvBR6NYBV/8uCXfaOZbSA4eJwC/MHdtwOY2e8JmmbnAWvc/ZWwfBnwF3d3M3uF4A3VWRnw2zD7LSD4gHVnNrDUzP63U3lms+2/AHeZ2WEeviP3Vu5ea2bHErz2HwJ+C3wvs46ZfY7gV8QogmbUt7pZnO3i6rvbb38HbjWze4Dfu3uVmS0C5ppZPkECu3QX1nMacEj4IwxgqAWnwQDmeXC6NE5aN+QM4AhrP98/jOB0ymkEX1J1AO5e3Wn+94AG4E4z+xPQoX+bmQ0jOMA/HRb9huALtdXvw/+Lyf557OwJd29tyTyDILlaEo6XAAcBw4HjgIpwPxUDbxEkGbkw4PtmdgpBEjHOglOgLwM3W9D69Ed3/7uZ1YV1fpFt+3eiu/g3AKvcfdEuLCtTsZktJXg9FwPzM7bre2b2/jDmsQTHTwiOk62fg8XAhCz77u4wXgiOuT8BcPd/mtkbYewAf3X3GqDGzLYRfLFD8APziIw4O5w2CluCFrj7xnD8HuD9BIloGngwrHowQYI6P9y/SeCdcNrLwD1m9lA43574kJn9BzCIIElfFm5LtnVkO87kso4dbQsEx9Cc6bTRnpsH/ICMU0ahFB1f36LdWHZjxnCaIMPd0bsks35LxngL2fs3/YTgV/bhwL/tKEZ330rwS/wLO6izkOBXTpfnUOyN3D3t7gvc/UaC5vKPAeNbv+Dd/VdhYreN4IPahZlNIti3GwgOGMfmsOqs+83dbyb4xV4MPG9mU939GYKD4tvA3bZrHUsTwL+4+1Hh39jwQA1Bn6/Y6PQ6G8GpvdbtmujuT4Tl3SbdHpyHP57gi+XjBK0Hu6L189j6Wd6ZzNfYgO9mxDzZ3X8dls/NKD/Yg/4my4CjbOedlC8nSN6OCd+rm4Aid18BlIfL+R8z+4a7N4dlDxH0xfpTLhu9k/g7b+euav3xdABBIt96uucyguPQseH09bQf37o7rna373vymJvLMhu8vZ+LAcsyXrfD3f2McNo5wO0Ex4zFZpZtfTs9pljQD+anBK0whwO/oP216rKObMeZHS0/c1U72BbYxfeBkpc9NxeY1drikWEtcAyAmR1D0CzdWQ273hnzGeDjFlyBMhi4ANjdHv3DaO9kPC2H+rcSfFlm/VCGb+IkwSmuvZqZHWxmmZ0fjyI4pfBL4LbwgICZJQkOqtmWUQrcQZCIOHAbMM3MTsio8xkLOsBlyrrfzOxAd3/F3b9PcCpkqpkdAGzw4Dz3Lwnfkzl6giApa13+Ubswb7+R5XV+HPj3sDUKMzso/Cw9AfyrmQ0Ky0d2Wk4JQRP8owSdIDu8Hu6+Ddhi7f1ZPgs8Tc94HPh/YZyYWVnYQvIk8MlwGDMbZWbj3X0lwa//b2X0XzjYzD7WabnDCN4fKTM7nfY+I2MJTsncTfC5PyZMyoe6+yPAVwlOV+9p/D0ifO1nAteE+7V1u5ot6KOyw86j4Y+zbWELFATJT6tnWsfN7CCCU6gr9zDkF4APWNDvLEnQep/tvbISKA1btTGzfDM7NExKx7n7X4H/IGiBK6Hrd8p/A7e0HkPMrNC6Xu3UmqhsCt/jrf1Tsq4j23Emx23Oui05ztuFThvtIXevAn6cZdKDwOVhk+Yigr4Rnb0MpMzsHwTnBZdkqdN5fS+Z2a+BF8OiO919iXXtDJyLm4DfmdnbwPNkT7Ay173JzP5AcOBq1dpsC0FmPc13rXd8XJUAP7Hg0swUwbnvGQStLN8BXjWzGqCe4PTBunC+1tcrP5yv9csBd19vZpcAP7DgCpkWggPn7+noJrLvt6+EB+o0sJygo+clwLVm1kzQT2tXWl5mAreb2csEx4pnCDoExkG3rzNwJ8FphpfCL/aNBP0N/hwmaBVm1kRwKvgbGcscAjwcJqZGx89Bq2nAHWECtBr4XE9sjLs/Gv44eD7MRWqAT7v7K2b2beDJ8MummWAfvRmu+1ag0szqw+28ptOi7wb+aGYVBB1mXw/LjyQ4bdQCNIXLHAb83swKCX74Xr2n8e/iy7CzdSwJj6WXAPfQvl2tffJ25nMEp1jrCK8CC/2UYJ++QvBemu7BFYZ7Eus7ZnY98FeC99Kj7v5wlnpNFpzenB2e2soDfkTwffL/hWUG/NDdt5rZH4EHwtNSXwpf930I3h+trUtzO61jq5n9giDZXUvwfQXBD9Fs6/hOluPMTjta72Bbcj3F2YHusCsiIiKxotNGIiIiEitKXkRERCRWlLyIiIhIrCh5ERERkVhR8iIiIiKxouRFRPqchc9a2tM6IjIwKHkRERGRWFHyIiK7zHJ/ivlIC568+7IFT9o9IiwfZcETfZeY2c/JuF16eFfh1qck/zy8A+mOYqk1s/+y4InAz1v7U7a7e/r2TWb2m3D9a83sQjO7xYIn/P7Z2u+8m/UJuCLS95S8iMjumkxwd+kjCG4R/mmCh9hdQ/udab9N8KTtI8Kyu8LyG+H/b+/+VaMKogCMf6dOIb6ATSoLEdHCP0ECEkEshJBX8CEECy0EbdLkCYQUgl1AhIg2phDEEAjkIUIqy4TkpJi57E0Im91lN2Hg+zW7DHu5w1aHu8t8bGXmPUof7BZARNym1Gaf1CbNMWePar/IHPAnM+9STgF+Xde3gIf1Hl8oR5x35indllfAOiWwd4dyIvLLOsCsUXov9ymnkn4Y/auRNEvmASRNapSK+QIl4kdm/qpPXG5QYpHLdf1bRHT15GeUCNzfGFSS9y/ZxyGDwvE/YKm+H1ZN/167N7uUY9C7yGK398sKuJKukcOLpEmNUtS9KACT5177AvicmW/G2MdRDjon/WLzGrCamRsRsUhpQp3Ze2aeRET/+m7vXQH30Rj7kHRF/NlI0iz1i7yLwEFm/j+3/gK4WT//E1ipYcruPzNDi8BDjFtN75tqAVfSdDm8SJqld8CDWqb+yGCIeA88jYht4Dmlgkxm7gFvgc16zQ9GKNYOuffXiPgNHIxzYWYeAivAp1oq3gEeT7gPSVNmVVqSJDXFJy+SJKkpDi+SJKkpDi+SJKkpDi+SJKkpDi+SJKkpDi+SJKkpDi+SJKkpDi+SJKkpp+I+/SiMcRmVAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 648x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "model name\n",
       "DecisionTreeClassifier    0.482481\n",
       "MultinomialNB             0.770846\n",
       "RandomForestClassifier    0.401546\n",
       "SGDClassifier             0.700938\n",
       "Name: accuracy, dtype: float64"
      ]
     },
     "execution_count": 280,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Answer\n",
    "models = [mnb, svm, dct, rf]\n",
    "from sklearn.model_selection import cross_val_score\n",
    "# Fitting the model into a plot\n",
    "CV = 5 # 5-fold cross validation\n",
    "cv_df_bow = pd.DataFrame(index=range(CV*len(models)))\n",
    "entries=[]\n",
    "for model in models:\n",
    "    model_name = model.__class__.__name__\n",
    "    accuracies = cross_val_score(model,bow_train_answer,train_labels,scoring='accuracy',cv=CV)\n",
    "    for fold_idx, accuracy in enumerate(accuracies):\n",
    "        entries.append((model_name,fold_idx,accuracy))\n",
    "    cv_df_bow=pd.DataFrame(entries, columns=['model name','fold_idx','accuracy'])\n",
    "\n",
    "# Build a plots to visualize the model selections\n",
    "plt.figure(figsize=(9,6))\n",
    "sns.boxplot(x='model name',y='accuracy',data=cv_df_bow)\n",
    "sns.stripplot(x='model name',y='accuracy',data=cv_df_bow,\n",
    "             size=9, jitter=True, edgecolor=\"gray\",linewidth=3)\n",
    "\n",
    "plt.show()\n",
    "cv_df_bow.groupby('model name').accuracy.mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 284,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 648x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "model name\n",
       "DecisionTreeClassifier    0.559033\n",
       "MultinomialNB             0.761082\n",
       "RandomForestClassifier    0.490455\n",
       "SGDClassifier             0.757722\n",
       "Name: accuracy, dtype: float64"
      ]
     },
     "execution_count": 284,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Over Sampling Answer BOW\n",
    "models = [mnb, svm, dct, rf]\n",
    "from sklearn.model_selection import cross_val_score\n",
    "# Fitting the model into a plot\n",
    "CV = 5 # 5-fold cross validation\n",
    "cv_df_bow = pd.DataFrame(index=range(CV*len(models)))\n",
    "entries=[]\n",
    "for model in models:\n",
    "    model_name = model.__class__.__name__\n",
    "    accuracies = cross_val_score(model,A,b,scoring='accuracy',cv=CV)\n",
    "    for fold_idx, accuracy in enumerate(accuracies):\n",
    "        entries.append((model_name,fold_idx,accuracy))\n",
    "    cv_df_bow=pd.DataFrame(entries, columns=['model name','fold_idx','accuracy'])\n",
    "\n",
    "# Build a plots to visualize the model selections\n",
    "plt.figure(figsize=(9,6))\n",
    "sns.boxplot(x='model name',y='accuracy',data=cv_df_bow)\n",
    "sns.stripplot(x='model name',y='accuracy',data=cv_df_bow,\n",
    "             size=9, jitter=True, edgecolor=\"gray\",linewidth=3)\n",
    "\n",
    "plt.show()\n",
    "cv_df_bow.groupby('model name').accuracy.mean()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## `Note` \n",
    "**In comparison with TFIDF and those 4 models, TFIDF still have better performance after being over sampling than BOW** "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# ___\n",
    "##  SVM {BOW | Unigram}\n",
    "\n",
    "- In the case of SVM with BOW - oversampling decreases the accuracy rate of the models - **OVER SAMPLING do not work** in this case\n",
    "- The accuracy rate on testing data for both answers and questions are roughly **65%**\n",
    "\n",
    "## `Question`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 241,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.64286\n",
      "Precision: 0.64205\n",
      "Recall: 0.64286\n",
      "F1 Score: 0.64197\n"
     ]
    }
   ],
   "source": [
    "# Question: Support Vector Machine | BOW\n",
    "svm_bow_predictions = train_predict_evaluate_model(classifier=svm,\n",
    "                                                 train_features=bow_train_question,\n",
    "                                                 train_labels=train_labels,\n",
    "                                                 test_features=bow_test_question,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 256,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.61555\n",
      "Precision: 0.63091\n",
      "Recall: 0.61555\n",
      "F1 Score: 0.61884\n"
     ]
    }
   ],
   "source": [
    "svm_bow_predictions = train_predict_evaluate_model(classifier=svm,\n",
    "                                                 train_features=X,\n",
    "                                                 train_labels=y,\n",
    "                                                 test_features=bow_test_question,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## `Answer`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 240,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.65336\n",
      "Precision: 0.65941\n",
      "Recall: 0.65336\n",
      "F1 Score: 0.65517\n"
     ]
    }
   ],
   "source": [
    "# Answer: Support Vector Machine BOW\n",
    "svm_bow_predictions = train_predict_evaluate_model(classifier=svm,\n",
    "                                                 train_features=bow_train_answer,\n",
    "                                                 train_labels=train_labels,\n",
    "                                                 test_features=bow_test_answer,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 261,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.60294\n",
      "Precision: 0.61012\n",
      "Recall: 0.60294\n",
      "F1 Score: 0.60397\n"
     ]
    }
   ],
   "source": [
    "# Oversampling\n",
    "# Answer: Support Vector Machine BOW\n",
    "svm_bow_predictions = train_predict_evaluate_model(classifier=svm,\n",
    "                                                 train_features=A,\n",
    "                                                 train_labels=b,\n",
    "                                                 test_features=bow_test_answer,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##  Decision Tree  {BOW | Unigram}\n",
    "\n",
    "- Decision Tree on BOW shows a very low accuracy rate - roughly 47% ~ 49% on testing data set\n",
    "- Oversampling do not work in this case since it decreases the accuracy of decision tree by 10% for both question and answer category classification\n",
    "\n",
    "### `Question`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 263,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.4916\n",
      "Precision: 0.48683\n",
      "Recall: 0.4916\n",
      "F1 Score: 0.48423\n"
     ]
    }
   ],
   "source": [
    "# Not over sampling\n",
    "# Question: Decision Tree | BOW\n",
    "dct_bow_predictions = train_predict_evaluate_model(classifier=dct,\n",
    "                                                 train_features=bow_train_question,\n",
    "                                                 train_labels=train_labels,\n",
    "                                                 test_features=bow_test_question,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 265,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.40966\n",
      "Precision: 0.43839\n",
      "Recall: 0.40966\n",
      "F1 Score: 0.41476\n"
     ]
    }
   ],
   "source": [
    "# Over sampling\n",
    "# Question: Decision Tree | BOW\n",
    "dct_bow_predictions = train_predict_evaluate_model(classifier=dct,\n",
    "                                                 train_features=X,\n",
    "                                                 train_labels=y,\n",
    "                                                 test_features=bow_test_question,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## `Answer`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 243,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.47479\n",
      "Precision: 0.46931\n",
      "Recall: 0.47479\n",
      "F1 Score: 0.47126\n"
     ]
    }
   ],
   "source": [
    "# Answer: Decision Tree | BOW\n",
    "dct_bow_predictions = train_predict_evaluate_model(classifier=dct,\n",
    "                                                 train_features=bow_train_answer,\n",
    "                                                 train_labels=train_labels,\n",
    "                                                 test_features=bow_test_answer,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 262,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.44538\n",
      "Precision: 0.4597\n",
      "Recall: 0.44538\n",
      "F1 Score: 0.44658\n"
     ]
    }
   ],
   "source": [
    "# Over Samppling\n",
    "# Answer: Decision Tree | BOW\n",
    "dct_bow_predictions = train_predict_evaluate_model(classifier=dct,\n",
    "                                                 train_features=A,\n",
    "                                                 train_labels=b,\n",
    "                                                 test_features=bow_test_answer,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##  Naive Bayes  {BOW | Unigram}\n",
    "\n",
    "- Naive Bayes works the best with BOW with 72% accuracy roughly on both question and answer\n",
    "- However, the over sampling does not work in this case because it decrease the accuracy of Naive Bayes\n",
    "\n",
    "## ` Question`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 270,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.72269\n",
      "Precision: 0.72444\n",
      "Recall: 0.72269\n",
      "F1 Score: 0.71833\n"
     ]
    }
   ],
   "source": [
    "# Question: Naive Bayes | BOW\n",
    "mnb_bow_predictions_question = train_predict_evaluate_model(classifier=mnb,\n",
    "                                            train_features=bow_train_question,\n",
    "                                            train_labels=train_labels,\n",
    "                                            test_features=bow_test_question,\n",
    "                                            test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 269,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.72899\n",
      "Precision: 0.73178\n",
      "Recall: 0.72899\n",
      "F1 Score: 0.72611\n"
     ]
    }
   ],
   "source": [
    "# Question - Oversampling\n",
    "mnb_bow_predictions_answer = train_predict_evaluate_model(classifier=mnb,\n",
    "                                            train_features=X,\n",
    "                                            train_labels=y,\n",
    "                                            test_features=bow_test_question,\n",
    "                                            test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## `Answer`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 252,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.73739\n",
      "Precision: 0.73872\n",
      "Recall: 0.73739\n",
      "F1 Score: 0.73071\n"
     ]
    }
   ],
   "source": [
    "# Answer: Naive Bayes | BOW\n",
    "mnb_bow_predictions_answer = train_predict_evaluate_model(classifier=mnb,\n",
    "                                            train_features=bow_train_answer,\n",
    "                                            train_labels=train_labels,\n",
    "                                            test_features=bow_test_answer,\n",
    "                                            test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 268,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.72479\n",
      "Precision: 0.72527\n",
      "Recall: 0.72479\n",
      "F1 Score: 0.72279\n"
     ]
    }
   ],
   "source": [
    "# Over Sampling\n",
    "# Answer: Naive Bayes | BOW\n",
    "mnb_bow_predictions_answer = train_predict_evaluate_model(classifier=mnb,\n",
    "                                            train_features=A,\n",
    "                                            train_labels=b,\n",
    "                                            test_features=bow_test_answer,\n",
    "                                            test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Random Forest {BOW | Unigram}\n",
    "\n",
    "- Oversampling works with Random Forest and increased the accuracy by 10%\n",
    "- Yet the accuracy of random forest is still roughly 50% to 49%"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## `Question`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 276,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.42857\n",
      "Precision: 0.7551\n",
      "Recall: 0.42857\n",
      "F1 Score: 0.25714\n"
     ]
    }
   ],
   "source": [
    "# Question\n",
    "rf_bow_predictions_question = train_predict_evaluate_model(classifier=rf,\n",
    "                                            train_features=bow_train_question,\n",
    "                                            train_labels=train_labels,\n",
    "                                            test_features=bow_test_question,\n",
    "                                            test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 275,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.53992\n",
      "Precision: 0.57838\n",
      "Recall: 0.53992\n",
      "F1 Score: 0.53754\n"
     ]
    }
   ],
   "source": [
    "# Oversampling quesiton\n",
    "rf_bow_predictions_question = train_predict_evaluate_model(classifier=rf,\n",
    "                                            train_features=X,\n",
    "                                            train_labels=y,\n",
    "                                            test_features=bow_test_question,\n",
    "                                            test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## `Answer`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 278,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.42857\n",
      "Precision: 0.7551\n",
      "Recall: 0.42857\n",
      "F1 Score: 0.25714\n"
     ]
    }
   ],
   "source": [
    "# Answer\n",
    "rf_tfidf_predictions_question = train_predict_evaluate_model(classifier=rf,\n",
    "                                            train_features=bow_train_answer,\n",
    "                                            train_labels=train_labels,\n",
    "                                            test_features=bow_test_answer,\n",
    "                                            test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 279,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.5105\n",
      "Precision: 0.5159\n",
      "Recall: 0.5105\n",
      "F1 Score: 0.50899\n"
     ]
    }
   ],
   "source": [
    "# Answer\n",
    "rf_tfidf_predictions_question = train_predict_evaluate_model(classifier=rf,\n",
    "                                            train_features=A,\n",
    "                                            train_labels=b,\n",
    "                                            test_features=bow_test_answer,\n",
    "                                            test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# ____\n",
    "# `TFIDF Features` - NOT OVERSAMPLING\n",
    "\n",
    "As observed in the plot below, the accuracy rate on TFIDF features without oversampling is lower than TFIDF with over sampling (in the main submission) - Therefore we don't have to consider TFIDF that are not over sampling\n",
    "\n",
    "As observed accuracy rate on testing dataset, SVM (in part 6) with TFIDF still performs with the highest accuracy rate on both testing and training dataset \n",
    "Therefore, we go with SVM \n",
    "\n",
    "\n",
    "\n",
    "### Question"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 327,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 648x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "model name\n",
       "DecisionTreeClassifier    0.799969\n",
       "MultinomialNB             0.564730\n",
       "RandomForestClassifier    0.401546\n",
       "SGDClassifier             0.901300\n",
       "Name: accuracy, dtype: float64"
      ]
     },
     "execution_count": 327,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "\n",
    "# Answer\n",
    "models = [mnb, svm, dct, rf]\n",
    "from sklearn.model_selection import cross_val_score\n",
    "# Fitting the model into a plot\n",
    "CV = 5 # 5-fold cross validation\n",
    "cv_df_tfidf = pd.DataFrame(index=range(CV*len(models)))\n",
    "entries=[]\n",
    "for model in models:\n",
    "    model_name = model.__class__.__name__\n",
    "    accuracies = cross_val_score(model,tfidf_train_question,train_labels,scoring='accuracy',cv=CV)\n",
    "    for fold_idx, accuracy in enumerate(accuracies):\n",
    "        entries.append((model_name,fold_idx,accuracy))\n",
    "    cv_df_tfidf=pd.DataFrame(entries, columns=['model name','fold_idx','accuracy'])\n",
    "\n",
    "# Build a plots to visualize the model selections\n",
    "plt.figure(figsize=(9,6))\n",
    "sns.boxplot(x='model name',y='accuracy',data=cv_df_tfidf)\n",
    "sns.stripplot(x='model name',y='accuracy',data=cv_df_tfidf,\n",
    "             size=9, jitter=True, edgecolor=\"gray\",linewidth=3)\n",
    "\n",
    "plt.show()\n",
    "cv_df_tfidf.groupby('model name').accuracy.mean()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Answer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 328,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 648x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "model name\n",
       "DecisionTreeClassifier    0.479028\n",
       "MultinomialNB             0.406481\n",
       "RandomForestClassifier    0.401546\n",
       "SGDClassifier             0.772987\n",
       "Name: accuracy, dtype: float64"
      ]
     },
     "execution_count": 328,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Answer\n",
    "models = [mnb, svm, dct, rf]\n",
    "from sklearn.model_selection import cross_val_score\n",
    "# Fitting the model into a plot\n",
    "CV = 5 # 5-fold cross validation\n",
    "cv_df_tfidf = pd.DataFrame(index=range(CV*len(models)))\n",
    "entries=[]\n",
    "for model in models:\n",
    "    model_name = model.__class__.__name__\n",
    "    accuracies = cross_val_score(model,tfidf_train_answer,train_labels,scoring='accuracy',cv=CV)\n",
    "    for fold_idx, accuracy in enumerate(accuracies):\n",
    "        entries.append((model_name,fold_idx,accuracy))\n",
    "    cv_df_tfidf=pd.DataFrame(entries, columns=['model name','fold_idx','accuracy'])\n",
    "\n",
    "# Build a plots to visualize the model selections\n",
    "plt.figure(figsize=(9,6))\n",
    "sns.boxplot(x='model name',y='accuracy',data=cv_df_tfidf)\n",
    "sns.stripplot(x='model name',y='accuracy',data=cv_df_tfidf,\n",
    "             size=9, jitter=True, edgecolor=\"gray\",linewidth=3)\n",
    "\n",
    "plt.show()\n",
    "cv_df_tfidf.groupby('model name').accuracy.mean()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Decision Tree {TFIDF | Oversampling |Unigram & Bigram}\n",
    "\n",
    "## `Question`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 290,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.43067\n",
      "Precision: 0.43184\n",
      "Recall: 0.43067\n",
      "F1 Score: 0.43004\n"
     ]
    }
   ],
   "source": [
    "# Question: Decision Tree | TFIDF Features\n",
    "dct_tfidf_predictions = train_predict_evaluate_model(classifier=dct,\n",
    "                                                  train_features=X_train_question,\n",
    "                                                 train_labels=y_train_question,\n",
    "                                                 test_features=tfidf_test_question,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## `Answer`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 291,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.48529\n",
      "Precision: 0.4785\n",
      "Recall: 0.48529\n",
      "F1 Score: 0.48072\n"
     ]
    }
   ],
   "source": [
    "# Answer Decision Tree | TFIDF Features\n",
    "dct_tfidf_predictions = train_predict_evaluate_model(classifier=dct,\n",
    "                                                  train_features=X_train_answer,\n",
    "                                                 train_labels=y_train_answer,\n",
    "                                                 test_features=tfidf_test_answer,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Random Forest {TFIDF| Oversampling| Unigram & Bigram}\n",
    "\n",
    "## `Question`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 293,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.57563\n",
      "Precision: 0.61671\n",
      "Recall: 0.57563\n",
      "F1 Score: 0.58112\n"
     ]
    }
   ],
   "source": [
    "# Question: Decision Tree | TFIDF Features\n",
    "rf_tfidf_predictions = train_predict_evaluate_model(classifier=rf,\n",
    "                                                  train_features=X_train_question,\n",
    "                                                 train_labels=y_train_question,\n",
    "                                                 test_features=tfidf_test_question,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## `Answer`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 294,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.5063\n",
      "Precision: 0.58209\n",
      "Recall: 0.5063\n",
      "F1 Score: 0.5108\n"
     ]
    }
   ],
   "source": [
    "# Answer Decision Tree | TFIDF Features\n",
    "rf_tfidf_predictions = train_predict_evaluate_model(classifier=rf,\n",
    "                                                  train_features=X_train_answer,\n",
    "                                                 train_labels=y_train_answer,\n",
    "                                                 test_features=tfidf_test_answer,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Naive Bayes {TFIDF | Oversampling |Unigram & Bigram}\n",
    "\n",
    "## `Question`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 296,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.69958\n",
      "Precision: 0.73842\n",
      "Recall: 0.69958\n",
      "F1 Score: 0.70293\n"
     ]
    }
   ],
   "source": [
    "# Question: Decision Tree | TFIDF Features\n",
    "mnb_tfidf_predictions = train_predict_evaluate_model(classifier=mnb,\n",
    "                                                  train_features=X_train_question,\n",
    "                                                 train_labels=y_train_question,\n",
    "                                                 test_features=tfidf_test_question,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## `Answer`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 295,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.69748\n",
      "Precision: 0.72507\n",
      "Recall: 0.69748\n",
      "F1 Score: 0.70154\n"
     ]
    }
   ],
   "source": [
    "# Answer Decision Tree | TFIDF Features\n",
    "mnb_tfidf_predictions = train_predict_evaluate_model(classifier=mnb,\n",
    "                                                  train_features=X_train_answer,\n",
    "                                                 train_labels=y_train_answer,\n",
    "                                                 test_features=tfidf_test_answer,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# __________\n",
    "# `Word2Vec`\n",
    "\n",
    "Eventhough oversampling does increased accuracy for those model but not much when comparing to other model. Word2Vec is underperform"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 329,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 648x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "model name\n",
       "DecisionTreeClassifier    0.551074\n",
       "RandomForestClassifier    0.595820\n",
       "SGDClassifier             0.615232\n",
       "Name: accuracy, dtype: float64"
      ]
     },
     "execution_count": 329,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Question\n",
    "models = [ svm, dct, rf]\n",
    "from sklearn.model_selection import cross_val_score\n",
    "# Fitting the model into a plot\n",
    "CV = 5 # 5-fold cross validation\n",
    "cv_df_w2v = pd.DataFrame(index=range(CV*len(models)))\n",
    "entries=[]\n",
    "for model in models:\n",
    "    model_name = model.__class__.__name__\n",
    "    accuracies = cross_val_score(model,avg_wv_traina_features,train_labels,scoring='accuracy',cv=CV)\n",
    "    for fold_idx, accuracy in enumerate(accuracies):\n",
    "        entries.append((model_name,fold_idx,accuracy))\n",
    "    cv_df_w2v=pd.DataFrame(entries, columns=['model name','fold_idx','accuracy'])\n",
    "\n",
    "# Build a plots to visualize the model selections\n",
    "plt.figure(figsize=(9,6))\n",
    "sns.boxplot(x='model name',y='accuracy',data=cv_df_w2v)\n",
    "sns.stripplot(x='model name',y='accuracy',data=cv_df_w2v,\n",
    "             size=9, jitter=True, edgecolor=\"gray\",linewidth=3)\n",
    "\n",
    "plt.show()\n",
    "cv_df_w2v.groupby('model name').accuracy.mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 330,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 648x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "model name\n",
       "DecisionTreeClassifier    0.818393\n",
       "RandomForestClassifier    0.598617\n",
       "SGDClassifier             0.631684\n",
       "Name: accuracy, dtype: float64"
      ]
     },
     "execution_count": 330,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Answer\n",
    "models = [ svm, dct, rf]\n",
    "from sklearn.model_selection import cross_val_score\n",
    "# Fitting the model into a plot\n",
    "CV = 5 # 5-fold cross validation\n",
    "cv_df_w2v = pd.DataFrame(index=range(CV*len(models)))\n",
    "entries=[]\n",
    "for model in models:\n",
    "    model_name = model.__class__.__name__\n",
    "    accuracies = cross_val_score(model,avg_wv_trainq_features,train_labels,scoring='accuracy',cv=CV)\n",
    "    for fold_idx, accuracy in enumerate(accuracies):\n",
    "        entries.append((model_name,fold_idx,accuracy))\n",
    "    cv_df_w2v=pd.DataFrame(entries, columns=['model name','fold_idx','accuracy'])\n",
    "\n",
    "# Build a plots to visualize the model selections\n",
    "plt.figure(figsize=(9,6))\n",
    "sns.boxplot(x='model name',y='accuracy',data=cv_df_w2v)\n",
    "sns.stripplot(x='model name',y='accuracy',data=cv_df_w2v,\n",
    "             size=9, jitter=True, edgecolor=\"gray\",linewidth=3)\n",
    "\n",
    "plt.show()\n",
    "cv_df_w2v.groupby('model name').accuracy.mean()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## SVM {word2vec}\n",
    "## `Question`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 311,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.58193\n",
      "Precision: 0.60266\n",
      "Recall: 0.58193\n",
      "F1 Score: 0.57298\n"
     ]
    }
   ],
   "source": [
    "# Question: Support Vector Machine | Word2vec\n",
    "svm_wv_predictions_q = train_predict_evaluate_model(classifier=svm,\n",
    "                                                 train_features=avg_wv_trainq_features,\n",
    "                                                 train_labels=train_labels,\n",
    "                                                 test_features=avg_wv_testq_features,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 312,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.58613\n",
      "Precision: 0.63231\n",
      "Recall: 0.58613\n",
      "F1 Score: 0.58528\n"
     ]
    }
   ],
   "source": [
    "# Question: Support Vector Machine | Word2vec #Oversampling\n",
    "svm_wv_predictions_q = train_predict_evaluate_model(classifier=svm,\n",
    "                                                 train_features=w2v_q,\n",
    "                                                 train_labels=q_labels,\n",
    "                                                 test_features=avg_wv_testq_features,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## `Answer`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 313,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.63866\n",
      "Precision: 0.6373\n",
      "Recall: 0.63866\n",
      "F1 Score: 0.63771\n"
     ]
    }
   ],
   "source": [
    "# Answer: Support Vector Machine | Word2vec\n",
    "svm_wv_predictions_a = train_predict_evaluate_model(classifier=svm,\n",
    "                                                 train_features=avg_wv_traina_features,\n",
    "                                                 train_labels=train_labels,\n",
    "                                                 test_features=avg_wv_testa_features,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 314,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.56513\n",
      "Precision: 0.59781\n",
      "Recall: 0.56513\n",
      "F1 Score: 0.56487\n"
     ]
    }
   ],
   "source": [
    "# Answer: Support Vector Machine | Word2vec\n",
    "svm_wv_predictions_a = train_predict_evaluate_model(classifier=svm,\n",
    "                                                 train_features=w2v_a,\n",
    "                                                 train_labels=a_labels,\n",
    "                                                 test_features=avg_wv_testa_features,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Decision Tree {word2vec}\n",
    "## `Question`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 317,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.51681\n",
      "Precision: 0.5237\n",
      "Recall: 0.51681\n",
      "F1 Score: 0.51831\n"
     ]
    }
   ],
   "source": [
    "# Question: Decision Tree | Word2vec\n",
    "dct_wv_predictions_q = train_predict_evaluate_model(classifier=dct,\n",
    "                                                 train_features=avg_wv_trainq_features,\n",
    "                                                 train_labels=train_labels,\n",
    "                                                 test_features=avg_wv_testq_features,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 318,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.4958\n",
      "Precision: 0.50449\n",
      "Recall: 0.4958\n",
      "F1 Score: 0.49897\n"
     ]
    }
   ],
   "source": [
    "# Question: Decision Tree | Word2vec # Oversampling\n",
    "dct_wv_predictions_q = train_predict_evaluate_model(classifier=dct,\n",
    "                                                 train_features=w2v_q,\n",
    "                                                 train_labels=q_labels,\n",
    "                                                 test_features=avg_wv_testq_features,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## `Answer`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 319,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.57143\n",
      "Precision: 0.56992\n",
      "Recall: 0.57143\n",
      "F1 Score: 0.56995\n"
     ]
    }
   ],
   "source": [
    "# Answer: Decision Tree | Word2vec \n",
    "dct_wv_predictions_a = train_predict_evaluate_model(classifier=dct,\n",
    "                                                 train_features=avg_wv_traina_features,\n",
    "                                                 train_labels=train_labels,\n",
    "                                                 test_features=avg_wv_testa_features,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 320,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.53151\n",
      "Precision: 0.55481\n",
      "Recall: 0.53151\n",
      "F1 Score: 0.53776\n"
     ]
    }
   ],
   "source": [
    "# Answer Decision Tree Oversampling\n",
    "dtc_wv_predictions_a = train_predict_evaluate_model(classifier=dct,\n",
    "                                                 train_features=w2v_a,\n",
    "                                                 train_labels=a_labels,\n",
    "                                                 test_features=avg_wv_testa_features,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Random Forest {word2vec}"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## `Question`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 304,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.59874\n",
      "Precision: 0.66562\n",
      "Recall: 0.59874\n",
      "F1 Score: 0.55853\n"
     ]
    }
   ],
   "source": [
    "# Question: Random Forest| Word2vec\n",
    "rf_wv_predictions_q = train_predict_evaluate_model(classifier=rf,\n",
    "                                                 train_features=avg_wv_trainq_features,\n",
    "                                                 train_labels=train_labels,\n",
    "                                                 test_features=avg_wv_testq_features,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 307,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.57773\n",
      "Precision: 0.6182\n",
      "Recall: 0.57773\n",
      "F1 Score: 0.5869\n"
     ]
    }
   ],
   "source": [
    "# Oversampling\n",
    "# Question: Random Forest | Word2vec\n",
    "rf_wv_predictions_q = train_predict_evaluate_model(classifier=rf,\n",
    "                                                 train_features=w2v_q,\n",
    "                                                 train_labels=q_labels,\n",
    "                                                 test_features=avg_wv_testq_features,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## `Answer`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 305,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.62185\n",
      "Precision: 0.68274\n",
      "Recall: 0.62185\n",
      "F1 Score: 0.57746\n"
     ]
    }
   ],
   "source": [
    "# Answer: Random Forest | Word2vec\n",
    "rf_wv_predictions_a = train_predict_evaluate_model(classifier=rf,\n",
    "                                                 train_features=avg_wv_traina_features,\n",
    "                                                 train_labels=train_labels,\n",
    "                                                 test_features=avg_wv_testa_features,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 321,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.55882\n",
      "Precision: 0.60534\n",
      "Recall: 0.55882\n",
      "F1 Score: 0.56437\n"
     ]
    }
   ],
   "source": [
    "# Answer Random Forest Oversampling\n",
    "rf_wv_predictions_a = train_predict_evaluate_model(classifier=rf,\n",
    "                                                 train_features=w2v_a,\n",
    "                                                 train_labels=a_labels,\n",
    "                                                 test_features=avg_wv_testa_features,\n",
    "                                                 test_labels=test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# References"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "1. Towards Data Science | Multiclass Text Classification with scikitlearn - https://towardsdatascience.com/multi-class-text-classification-with-scikit-learn-12f1e60e0a9f\n",
    "2. Towards Data Science | Machine Learning Multiclass Classification with imbalanced data set https://towardsdatascience.com/machine-learning-multiclass-classification-with-imbalanced-data-set-29f6a177c1a\n",
    "3. Machine Learning Mastery | A Gentle Introduction to Imbalanced Classification https://machinelearningmastery.com/what-is-imbalanced-classification/\n",
    "4. Kaggle | Best Algorithm for Classification https://www.kaggle.com/getting-started/42409\n",
    "5. Medium | Tips and Tricks for Multi-class classfification https://medium.com/@b.terryjack/tips-and-tricks-for-multi-class-classification-c184ae1c8ffc"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "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": 2
}