final_model.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Collecting textblob\n",
      "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/60/f0/1d9bfcc8ee6b83472ec571406bd0dd51c0e6330ff1a51b2d29861d389e85/textblob-0.15.3-py2.py3-none-any.whl (636kB)\n",
      "\u001b[K    100% |████████████████████████████████| 645kB 29.0MB/s \n",
      "\u001b[?25hRequirement already satisfied: nltk>=3.1 in /home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages (from textblob) (3.3)\n",
      "Requirement already satisfied: six in /home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages (from nltk>=3.1->textblob) (1.11.0)\n",
      "Installing collected packages: textblob\n",
      "Successfully installed textblob-0.15.3\n",
      "\u001b[33mYou are using pip version 10.0.1, however version 19.0.3 is available.\n",
      "You should consider upgrading via the 'pip install --upgrade pip' command.\u001b[0m\n",
      "Collecting gensim\n",
      "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/d7/b9/6c93685bed0026b6a1cce55ab173f6b617f6db0d1325d25489c2fd43e711/gensim-3.7.1-cp36-cp36m-manylinux1_x86_64.whl (24.2MB)\n",
      "\u001b[K    100% |████████████████████████████████| 24.2MB 1.9MB/s \n",
      "\u001b[?25hCollecting smart-open>=1.7.0 (from gensim)\n",
      "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/ff/c8/de7dcf34d4b5f2ae94fe1055e0d6418fb97a63c9dc3428edd264704983a2/smart_open-1.8.0.tar.gz (40kB)\n",
      "\u001b[K    100% |████████████████████████████████| 40kB 22.7MB/s \n",
      "\u001b[?25hRequirement already satisfied: six>=1.5.0 in /home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages (from gensim) (1.11.0)\n",
      "Requirement already satisfied: scipy>=0.18.1 in /home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages (from gensim) (1.1.0)\n",
      "Requirement already satisfied: numpy>=1.11.3 in /home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages (from gensim) (1.14.5)\n",
      "Requirement already satisfied: boto>=2.32 in /home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages (from smart-open>=1.7.0->gensim) (2.48.0)\n",
      "Collecting bz2file (from smart-open>=1.7.0->gensim)\n",
      "  Downloading https://files.pythonhosted.org/packages/61/39/122222b5e85cd41c391b68a99ee296584b2a2d1d233e7ee32b4532384f2d/bz2file-0.98.tar.gz\n",
      "Requirement already satisfied: requests in /home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages (from smart-open>=1.7.0->gensim) (2.20.1)\n",
      "Requirement already satisfied: boto3 in /home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages (from smart-open>=1.7.0->gensim) (1.9.116)\n",
      "Requirement already satisfied: urllib3<1.25,>=1.21.1 in /home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages (from requests->smart-open>=1.7.0->gensim) (1.22)\n",
      "Requirement already satisfied: certifi>=2017.4.17 in /home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages (from requests->smart-open>=1.7.0->gensim) (2018.11.29)\n",
      "Requirement already satisfied: idna<2.8,>=2.5 in /home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages (from requests->smart-open>=1.7.0->gensim) (2.6)\n",
      "Requirement already satisfied: chardet<3.1.0,>=3.0.2 in /home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages (from requests->smart-open>=1.7.0->gensim) (3.0.4)\n",
      "Requirement already satisfied: jmespath<1.0.0,>=0.7.1 in /home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages (from boto3->smart-open>=1.7.0->gensim) (0.9.3)\n",
      "Requirement already satisfied: s3transfer<0.3.0,>=0.2.0 in /home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages (from boto3->smart-open>=1.7.0->gensim) (0.2.0)\n",
      "Requirement already satisfied: botocore<1.13.0,>=1.12.116 in /home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages (from boto3->smart-open>=1.7.0->gensim) (1.12.116)\n",
      "Requirement already satisfied: python-dateutil<3.0.0,>=2.1; python_version >= \"2.7\" in /home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages (from botocore<1.13.0,>=1.12.116->boto3->smart-open>=1.7.0->gensim) (2.7.3)\n",
      "Requirement already satisfied: docutils>=0.10 in /home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages (from botocore<1.13.0,>=1.12.116->boto3->smart-open>=1.7.0->gensim) (0.14)\n",
      "Building wheels for collected packages: smart-open, bz2file\n",
      "  Running setup.py bdist_wheel for smart-open ... \u001b[?25ldone\n",
      "\u001b[?25h  Stored in directory: /home/ec2-user/.cache/pip/wheels/f7/a6/ff/9ab5842c14e50e95a06a4675b0b4a689c9cab6064dac2b01d0\n",
      "  Running setup.py bdist_wheel for bz2file ... \u001b[?25ldone\n",
      "\u001b[?25h  Stored in directory: /home/ec2-user/.cache/pip/wheels/81/75/d6/e1317bf09bf1af5a30befc2a007869fa6e1f516b8f7c591cb9\n",
      "Successfully built smart-open bz2file\n",
      "Installing collected packages: bz2file, smart-open, gensim\n",
      "Successfully installed bz2file-0.98 gensim-3.7.1 smart-open-1.8.0\n",
      "\u001b[33mYou are using pip version 10.0.1, however version 19.0.3 is available.\n",
      "You should consider upgrading via the 'pip install --upgrade pip' command.\u001b[0m\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[nltk_data] Downloading package punkt to /home/ec2-user/nltk_data...\n",
      "[nltk_data]   Unzipping tokenizers/punkt.zip.\n",
      "[nltk_data] Downloading package stopwords to\n",
      "[nltk_data]     /home/ec2-user/nltk_data...\n",
      "[nltk_data]   Unzipping corpora/stopwords.zip.\n"
     ]
    }
   ],
   "source": [
    "from __future__ import division, print_function\n",
    "!pip install textblob\n",
    "!pip install gensim\n",
    "from tensorflow import keras\n",
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "import numpy as np, pandas as pd\n",
    "from numpy import array\n",
    "from numpy import asarray\n",
    "from numpy import zeros\n",
    "import os\n",
    "\n",
    "from keras.preprocessing.text import Tokenizer\n",
    "from keras.preprocessing.sequence import pad_sequences\n",
    "from keras.models import Model, Sequential\n",
    "from keras.layers import Flatten\n",
    "from keras.layers import Embedding\n",
    "from keras.initializers import Constant\n",
    "from keras.layers import Bidirectional,Dense,Dropout,LSTM,Dropout, Activation,Add,BatchNormalization,dot,Input,concatenate,SpatialDropout1D,Lambda\n",
    "from keras import optimizers\n",
    "from keras import backend as K\n",
    "from keras.models import model_from_json\n",
    "from keras.preprocessing.sequence import pad_sequences\n",
    "from keras.callbacks import EarlyStopping\n",
    "\n",
    "import nltk\n",
    "nltk.download('punkt')\n",
    "nltk.download('stopwords')\n",
    "\n",
    "import numpy as np\n",
    "import collections\n",
    "import os\n",
    "import pickle\n",
    "from textblob import TextBlob\n",
    "import re\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "from nltk.corpus import stopwords\n",
    "\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Collecting pydot\n",
      "  Downloading https://files.pythonhosted.org/packages/33/d1/b1479a770f66d962f545c2101630ce1d5592d90cb4f083d38862e93d16d2/pydot-1.4.1-py2.py3-none-any.whl\n",
      "Requirement already satisfied: pyparsing>=2.1.4 in /home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages (from pydot) (2.2.0)\n",
      "Installing collected packages: pydot\n",
      "Successfully installed pydot-1.4.1\n",
      "\u001b[33mYou are using pip version 10.0.1, however version 19.0.3 is available.\n",
      "You should consider upgrading via the 'pip install --upgrade pip' command.\u001b[0m\n"
     ]
    }
   ],
   "source": [
    "!pip install pydot"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "DATA_DIR = \"InferSent/encoder/data\"\n",
    "EMBED_SIZE = 4096\n",
    "QA_TRAIN_FILE = \"train.tsv\"\n",
    "QA_EMBED_SIZE = 64\n",
    "BATCH_SIZE = 32\n",
    "NBR_EPOCHS = 20\n",
    "vocab_size = 5467\n",
    "stop = set(stopwords.words('english'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "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>ID</th>\n",
       "      <th>Question Stem</th>\n",
       "      <th>Choices</th>\n",
       "      <th>Complete Question</th>\n",
       "      <th>Answer Key</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>7-980</td>\n",
       "      <td>The sun is responsible for</td>\n",
       "      <td>(A) puppies learning new tricks (B) children g...</td>\n",
       "      <td>The sun is responsible for (A) puppies learnin...</td>\n",
       "      <td>D</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>7-584</td>\n",
       "      <td>When standing miles away from Mount Rushmore</td>\n",
       "      <td>(A) the mountains seem very close (B) the moun...</td>\n",
       "      <td>When standing miles away from Mount Rushmore (...</td>\n",
       "      <td>D</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>7-870</td>\n",
       "      <td>When food is reduced in the stomach</td>\n",
       "      <td>(A) the mind needs time to digest (B) take a s...</td>\n",
       "      <td>When food is reduced in the stomach (A) the mi...</td>\n",
       "      <td>C</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      ID                                 Question Stem  \\\n",
       "0  7-980                    The sun is responsible for   \n",
       "1  7-584  When standing miles away from Mount Rushmore   \n",
       "2  7-870           When food is reduced in the stomach   \n",
       "\n",
       "                                             Choices  \\\n",
       "0  (A) puppies learning new tricks (B) children g...   \n",
       "1  (A) the mountains seem very close (B) the moun...   \n",
       "2  (A) the mind needs time to digest (B) take a s...   \n",
       "\n",
       "                                   Complete Question Answer Key  \n",
       "0  The sun is responsible for (A) puppies learnin...          D  \n",
       "1  When standing miles away from Mount Rushmore (...          D  \n",
       "2  When food is reduced in the stomach (A) the mi...          C  "
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train = pd.read_csv(\"InferSent/encoder/data/train.tsv\",sep='\\t')\n",
    "train.head(3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages/ipykernel/__main__.py:5: FutureWarning: set_value is deprecated and will be removed in a future release. Please use .at[] or .iat[] accessors instead\n",
      "/home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages/ipykernel/__main__.py:6: FutureWarning: set_value is deprecated and will be removed in a future release. Please use .at[] or .iat[] accessors instead\n",
      "/home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages/ipykernel/__main__.py:7: FutureWarning: set_value is deprecated and will be removed in a future release. Please use .at[] or .iat[] accessors instead\n",
      "/home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages/ipykernel/__main__.py:8: FutureWarning: set_value is deprecated and will be removed in a future release. Please use .at[] or .iat[] accessors instead\n",
      "/home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages/ipykernel/__main__.py:21: FutureWarning: set_value is deprecated and will be removed in a future release. Please use .at[] or .iat[] accessors instead\n",
      "/home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages/ipykernel/__main__.py:19: FutureWarning: set_value is deprecated and will be removed in a future release. Please use .at[] or .iat[] accessors instead\n",
      "/home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages/ipykernel/__main__.py:15: FutureWarning: set_value is deprecated and will be removed in a future release. Please use .at[] or .iat[] accessors instead\n",
      "/home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages/ipykernel/__main__.py:17: FutureWarning: set_value is deprecated and will be removed in a future release. Please use .at[] or .iat[] accessors instead\n"
     ]
    }
   ],
   "source": [
    "#Create new features from the answer choices  \n",
    "for index, row in train.iterrows():\n",
    "    ans = re.split(\"(\\([A-Z]\\))\", row['Choices'])\n",
    "    ans.remove('')\n",
    "    train.set_value(index,'OptionA',ans[1])\n",
    "    train.set_value(index,'OptionB',ans[3])\n",
    "    train.set_value(index,'OptionC',ans[5])\n",
    "    train.set_value(index,'OptionD',ans[7])\n",
    "    \n",
    "#Find the answer key and replace it with the value   \n",
    "for index, row in train.iterrows():\n",
    "    ans = re.split(\"(\\([A-Z]\\))\", row['Choices'])\n",
    "    ans.remove('')\n",
    "    if row['Answer Key'] == 'A':\n",
    "        train.set_value(index,'Result',ans[1])\n",
    "    elif row['Answer Key'] == 'B':\n",
    "        train.set_value(index,'Result',ans[3])\n",
    "    elif row['Answer Key'] == 'C':\n",
    "        train.set_value(index,'Result',ans[5])\n",
    "    elif row['Answer Key'] == 'D':\n",
    "        train.set_value(index,'Result',ans[7])     "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "num_of_categories = 4957\n",
    "shuffled = train.reindex(np.random.permutation(train.index))\n",
    "a = shuffled[shuffled['Answer Key'] == 'A'][:num_of_categories]\n",
    "b = shuffled[shuffled['Answer Key'] == 'B'][:num_of_categories]\n",
    "c = shuffled[shuffled['Answer Key'] == 'C'][:num_of_categories]\n",
    "d = shuffled[shuffled['Answer Key'] == 'D'][:num_of_categories]\n",
    "concated = pd.concat([a,b,c,d], ignore_index=True)\n",
    "#Shuffle the dataset\n",
    "concated = concated.reindex(np.random.permutation(concated.index))\n",
    "concated['LABEL'] = 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "4074    3\n",
      "3977    3\n",
      "1283    0\n",
      "1723    1\n",
      "1560    1\n",
      "2124    1\n",
      "61      0\n",
      "395     0\n",
      "3051    2\n",
      "2967    2\n",
      "Name: LABEL, dtype: int64\n",
      "[[0. 0. 0. 1.]\n",
      " [0. 0. 0. 1.]\n",
      " [1. 0. 0. 0.]\n",
      " [0. 1. 0. 0.]\n",
      " [0. 1. 0. 0.]\n",
      " [0. 1. 0. 0.]\n",
      " [1. 0. 0. 0.]\n",
      " [1. 0. 0. 0.]\n",
      " [0. 0. 1. 0.]\n",
      " [0. 0. 1. 0.]]\n"
     ]
    }
   ],
   "source": [
    "from keras.utils.np_utils import to_categorical\n",
    "#One-hot encode the labels\n",
    "concated.loc[concated['Answer Key'] == 'A', 'LABEL'] = 0\n",
    "concated.loc[concated['Answer Key'] == 'B', 'LABEL'] = 1\n",
    "concated.loc[concated['Answer Key'] == 'C','LABEL'] = 2\n",
    "concated.loc[concated['Answer Key'] == 'D','LABEL'] = 3\n",
    "print(concated['LABEL'][:10])\n",
    "labels = to_categorical(concated['LABEL'], num_classes=4)\n",
    "print(labels[:10])\n",
    "if 'CATEGORY' in concated.keys():\n",
    "    concated.drop(['CATEGORY'], axis=1)    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "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>ID</th>\n",
       "      <th>Question Stem</th>\n",
       "      <th>Choices</th>\n",
       "      <th>Complete Question</th>\n",
       "      <th>Answer Key</th>\n",
       "      <th>OptionA</th>\n",
       "      <th>OptionB</th>\n",
       "      <th>OptionC</th>\n",
       "      <th>OptionD</th>\n",
       "      <th>Result</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>7-980</td>\n",
       "      <td>The sun is responsible for</td>\n",
       "      <td>(A) puppies learning new tricks (B) children g...</td>\n",
       "      <td>The sun is responsible for (A) puppies learnin...</td>\n",
       "      <td>D</td>\n",
       "      <td>puppies learning new tricks</td>\n",
       "      <td>children growing up and getting old</td>\n",
       "      <td>flowers wilting in a vase</td>\n",
       "      <td>plants sprouting, blooming and wilting</td>\n",
       "      <td>plants sprouting, blooming and wilting</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>7-584</td>\n",
       "      <td>When standing miles away from Mount Rushmore</td>\n",
       "      <td>(A) the mountains seem very close (B) the moun...</td>\n",
       "      <td>When standing miles away from Mount Rushmore (...</td>\n",
       "      <td>D</td>\n",
       "      <td>the mountains seem very close</td>\n",
       "      <td>the mountains are boring</td>\n",
       "      <td>the mountains look the same as from up close</td>\n",
       "      <td>the mountains seem smaller than in photographs</td>\n",
       "      <td>the mountains seem smaller than in photographs</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>7-870</td>\n",
       "      <td>When food is reduced in the stomach</td>\n",
       "      <td>(A) the mind needs time to digest (B) take a s...</td>\n",
       "      <td>When food is reduced in the stomach (A) the mi...</td>\n",
       "      <td>C</td>\n",
       "      <td>the mind needs time to digest</td>\n",
       "      <td>take a second to digest what I said</td>\n",
       "      <td>nutrients are being deconstructed</td>\n",
       "      <td>reader's digest is a body of works</td>\n",
       "      <td>nutrients are being deconstructed</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      ID                                 Question Stem  \\\n",
       "0  7-980                    The sun is responsible for   \n",
       "1  7-584  When standing miles away from Mount Rushmore   \n",
       "2  7-870           When food is reduced in the stomach   \n",
       "\n",
       "                                             Choices  \\\n",
       "0  (A) puppies learning new tricks (B) children g...   \n",
       "1  (A) the mountains seem very close (B) the moun...   \n",
       "2  (A) the mind needs time to digest (B) take a s...   \n",
       "\n",
       "                                   Complete Question Answer Key  \\\n",
       "0  The sun is responsible for (A) puppies learnin...          D   \n",
       "1  When standing miles away from Mount Rushmore (...          D   \n",
       "2  When food is reduced in the stomach (A) the mi...          C   \n",
       "\n",
       "                           OptionA                                OptionB  \\\n",
       "0     puppies learning new tricks    children growing up and getting old    \n",
       "1   the mountains seem very close               the mountains are boring    \n",
       "2   the mind needs time to digest    take a second to digest what I said    \n",
       "\n",
       "                                          OptionC  \\\n",
       "0                      flowers wilting in a vase    \n",
       "1   the mountains look the same as from up close    \n",
       "2              nutrients are being deconstructed    \n",
       "\n",
       "                                           OptionD  \\\n",
       "0           plants sprouting, blooming and wilting   \n",
       "1   the mountains seem smaller than in photographs   \n",
       "2               reader's digest is a body of works   \n",
       "\n",
       "                                            Result  \n",
       "0           plants sprouting, blooming and wilting  \n",
       "1   the mountains seem smaller than in photographs  \n",
       "2               nutrients are being deconstructed   "
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#train = train.drop(['ID', 'Choices', 'Complete Question','Answer Key'], axis=1)\n",
    "train.head(3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "GloVe data loaded\n"
     ]
    }
   ],
   "source": [
    "embeddings_index = dict()\n",
    "f = open('InferSent/dataset/GloVe/glove.840B.300d.txt')\n",
    "for line in f:\n",
    "    values = line.split(' ')\n",
    "    word = values[0] ## The first entry is the word\n",
    "    coefs = np.asarray(values[1:], dtype='float32') ## These are the vecotrs representing the embedding for the word\n",
    "    embeddings_index[word] = coefs\n",
    "f.close()\n",
    "\n",
    "print('GloVe data loaded')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "\n",
    "#Split question, candidate answers and right answer to test and train data \n",
    "X_Question = train['Question Stem']\n",
    "X_Answers = train[['OptionA','OptionB','OptionC','OptionD',]]\n",
    "Y = train['Result']\n",
    "\n",
    "#Global variable initilisation \n",
    "MAX_NUM_WORDS = 10000\n",
    "MAX_SEQUENCE_LENGTH = 50"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "## Iterate over the questions to remove stopwords\n",
    "lines_without_stopwords_questions=[] \n",
    "for line in X_Question.values: \n",
    "    line = line.lower()\n",
    "    line_by_words = re.findall(r'(?:\\w+)', line, flags = re.UNICODE) # remove punctuation ans split\n",
    "    new_line=[]\n",
    "    for word in line_by_words:\n",
    "        if word not in stop:\n",
    "            new_line.append(word)\n",
    "    lines_without_stopwords_questions.append(new_line)\n",
    "question_texts = lines_without_stopwords_questions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 5249 unique tokens in questions.\n"
     ]
    }
   ],
   "source": [
    "#Generate tokenised text for questions\n",
    "#Generate question Embedding Matrix\n",
    "t = Tokenizer(num_words=MAX_NUM_WORDS)\n",
    "t.fit_on_texts(question_texts)\n",
    "\n",
    "word_index = t.word_index\n",
    "print('Found %s unique tokens in questions.' % len(word_index)) \n",
    "\n",
    "# integer encode the questions\n",
    "encoded_docs = t.texts_to_sequences(question_texts)\n",
    "# pad documents to a max length of 100 words\n",
    "padded_docs = pad_sequences(encoded_docs, maxlen=MAX_SEQUENCE_LENGTH, padding='post')\n",
    "#\n",
    "padded_docs_reshaped = np.delete(padded_docs,1,1)\n",
    "\n",
    "## EMBEDDING_DIM =  ## seems to need to match the embeddings_index dimension\n",
    "EMBEDDING_DIM = embeddings_index.get('a').shape[0]\n",
    "num_words = min(MAX_NUM_WORDS, len(word_index)) + 1\n",
    "embedding_matrix = np.zeros((num_words, EMBEDDING_DIM))\n",
    "\n",
    "\n",
    "for word, i in word_index.items():\n",
    "    if i > MAX_NUM_WORDS:\n",
    "        continue\n",
    "    embedding_vector = embeddings_index.get(word) ## This references the loaded embeddings dictionary\n",
    "    if embedding_vector is not None:\n",
    "        # words not found in embedding index will be all-zeros.\n",
    "        embedding_matrix[i] = embedding_vector"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "## Iterate over the candidate answers to remove stopwords\n",
    "lines_without_stopwords_candidate_answers=[] \n",
    "for answers in X_Answers.values: \n",
    "    for answer in answers:\n",
    "        answer = answer.lower()\n",
    "        line_by_words = re.findall(r'(?:\\w+)', answer, flags = re.UNICODE) # remove punctuation ans split\n",
    "        new_line=[]\n",
    "        for word in line_by_words:\n",
    "            if word not in stop:\n",
    "                new_line.append(word)\n",
    "    lines_without_stopwords_candidate_answers.append(new_line)\n",
    "answer_texts = lines_without_stopwords_candidate_answers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 3951 unique tokens in answers.\n",
      "[[ 0.          0.          0.         ...  0.          0.\n",
      "   0.        ]\n",
      " [-0.036665    0.20106     0.28510001 ... -0.34485     0.16798\n",
      "  -0.16952001]\n",
      " [-0.43511999  0.028351    0.49110001 ... -0.47150999  0.43009999\n",
      "   0.36572999]\n",
      " ...\n",
      " [ 0.23746     0.050147    0.032209   ...  0.57862002  0.29216\n",
      "  -0.21355   ]\n",
      " [ 0.92971998  0.48074999  0.60821998 ... -1.06930006  0.38532001\n",
      "   0.68383002]\n",
      " [ 0.80149001 -0.55936998 -0.0053804  ...  0.14056     0.2823\n",
      "  -0.12672   ]] 3952\n"
     ]
    }
   ],
   "source": [
    "#Generate tokenised text for candidate answers\n",
    "#Generate candidate answer Embedding Matrix\n",
    "t_candidate_ans = Tokenizer(num_words=MAX_NUM_WORDS)\n",
    "t_candidate_ans.fit_on_texts(answer_texts)\n",
    "\n",
    "word_index_candidate_ans = t_candidate_ans.word_index\n",
    "print('Found %s unique tokens in answers.' % len(word_index_candidate_ans)) \n",
    "\n",
    "# integer encode the questions\n",
    "encoded_docs_candidate_ans = t_candidate_ans.texts_to_sequences(answer_texts)\n",
    "\n",
    "# pad documents to a max length of 100 words\n",
    "padded_docs_candidate_ans = pad_sequences(encoded_docs_candidate_ans, maxlen=MAX_SEQUENCE_LENGTH, padding='post')\n",
    "\n",
    "# EMBEDDING_DIM =  ## seems to need to match the embeddings_index dimension\n",
    "EMBEDDING_DIM_CANDIDATE_ANS = embeddings_index.get('a').shape[0]\n",
    "num_words_candidate_ans = min(MAX_NUM_WORDS, len(word_index_candidate_ans)) + 1\n",
    "embedding_matrix_candidate_ans = np.zeros((num_words_candidate_ans, EMBEDDING_DIM_CANDIDATE_ANS))\n",
    "\n",
    "\n",
    "for word, i in word_index_candidate_ans.items():\n",
    "    if i > MAX_NUM_WORDS:\n",
    "        continue\n",
    "    embedding_vector_candidate_ans = embeddings_index.get(word) ## This references the loaded embeddings dictionary\n",
    "    if embedding_vector_candidate_ans is not None:\n",
    "        # words not found in embedding index will be all-zeros.\n",
    "        embedding_matrix_candidate_ans[i] = embedding_vector_candidate_ans\n",
    "        \n",
    "print(embedding_matrix_candidate_ans, num_words_candidate_ans )    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "## Iterate over the predicted answers to remove stopwords\n",
    "lines_without_stopwords_predicted_answers=[] \n",
    "for line in Y.values: \n",
    "    line = line.lower()\n",
    "    line_by_words = re.findall(r'(?:\\w+)', line, flags = re.UNICODE) # remove punctuation ans split\n",
    "    new_line=[]\n",
    "    for word in line_by_words:\n",
    "        if word not in stop:\n",
    "            new_line.append(word)\n",
    "    lines_without_stopwords_predicted_answers.append(new_line)\n",
    "pred_answer_texts = lines_without_stopwords_predicted_answers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 4525 unique tokens in answers.\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(4526, 300)"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Generate tokenised text for predicted answers\n",
    "#Generate predicated answer Embedding Matrix\n",
    "\n",
    "t_ans = Tokenizer(num_words=MAX_NUM_WORDS)\n",
    "t_ans.fit_on_texts(pred_answer_texts)\n",
    "word_index_ans = t_ans.word_index\n",
    "print('Found %s unique tokens in answers.' % len(word_index_ans)) \n",
    "\n",
    "# integer encode the questions\n",
    "encoded_docs_ans = t_ans.texts_to_sequences(pred_answer_texts)\n",
    "\n",
    "# pad documents to a max length of 100 words\n",
    "padded_docs_ans = pad_sequences(encoded_docs_ans, maxlen=MAX_SEQUENCE_LENGTH, padding='post')\n",
    "\n",
    "\n",
    "# EMBEDDING_DIM =  ## seems to need to match the embeddings_index dimension\n",
    "EMBEDDING_DIM_ANS = embeddings_index.get('a').shape[0]\n",
    "num_words_ans = min(MAX_NUM_WORDS, len(word_index_ans)) + 1\n",
    "embedding_matrix_ans = np.zeros((num_words_ans, EMBEDDING_DIM_ANS))\n",
    "\n",
    "\n",
    "for word, i in word_index_ans.items():\n",
    "    if i > MAX_NUM_WORDS:\n",
    "        continue\n",
    "    embedding_vector_ans = embeddings_index.get(word) ## This references the loaded embeddings dictionary\n",
    "    if embedding_vector_ans is not None:\n",
    "        # words not found in embedding index will be all-zeros.\n",
    "        embedding_matrix_ans[i] = embedding_vector_ans\n",
    "\n",
    "embedding_matrix_ans.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1326\n",
      "Found 1734 unique tokens in answers.\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(1735, 300)"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Get the knowledge text\n",
    "# Generate word embeddings from knowledge text\n",
    "sentences = []\n",
    "with open('InferSent/encoder/data/openbook.txt') as f:\n",
    "    for line in f:\n",
    "        sentences.append(line.strip())\n",
    "print(len(sentences))\n",
    "\n",
    "\n",
    "lines_without_stopwords_knowledge_text=[] \n",
    "for line in sentences : \n",
    "    line = line.lower()\n",
    "    line_by_words = re.findall(r'(?:\\w+)', line, flags = re.UNICODE) # remove punctuation ans split\n",
    "    new_line=[]\n",
    "    for word in line_by_words:\n",
    "        if word not in stop:\n",
    "            new_line.append(word)\n",
    "    lines_without_stopwords_knowledge_text.append(new_line)\n",
    "knowledge_texts = lines_without_stopwords_knowledge_text\n",
    "\n",
    "\n",
    "t_knowledge_text = Tokenizer(num_words=MAX_NUM_WORDS)\n",
    "t_knowledge_text.fit_on_texts(knowledge_texts)\n",
    "word_index_knowledge_text = t_knowledge_text.word_index\n",
    "print('Found %s unique tokens in answers.' % len(word_index_knowledge_text)) \n",
    "\n",
    "# integer encode the questions\n",
    "encoded_docs_knowledge_text= t_knowledge_text.texts_to_sequences(knowledge_texts)\n",
    "\n",
    "# pad documents to a max length of 100 words\n",
    "padded_docs_knowledge_text = pad_sequences(encoded_docs_knowledge_text, maxlen=MAX_SEQUENCE_LENGTH, padding='post')\n",
    "\n",
    "\n",
    "\n",
    "# EMBEDDING_DIM =  ## seems to need to match the embeddings_index dimension\n",
    "EMBEDDING_DIM_KT = embeddings_index.get('a').shape[0]\n",
    "num_words_knowledge_text = min(MAX_NUM_WORDS, len(word_index_knowledge_text)) + 1\n",
    "embedding_matrix_knowledge_text = np.zeros((num_words_knowledge_text, EMBEDDING_DIM_KT))\n",
    "\n",
    "\n",
    "for word, i in word_index_knowledge_text.items():\n",
    "    if i > MAX_NUM_WORDS:\n",
    "        continue\n",
    "    embedding_vector_knowledge_text = embeddings_index.get(word) ## This references the loaded embeddings dictionary\n",
    "    if  embedding_vector_knowledge_text is not None:\n",
    "        # words not found in embedding index will be all-zeros.\n",
    "        embedding_matrix_knowledge_text[i] = embedding_vector_knowledge_text\n",
    "\n",
    "embedding_matrix_knowledge_text.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##CREATE TRAINING,TESTING AND VALIDATION DATASET FROM TOKENISED QUESTION(X1),CONTEXT TEXT (X2) AND LABELS(Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(3717, 50) (1240, 50) (3717, 50)\n"
     ]
    }
   ],
   "source": [
    "X_question = padded_docs\n",
    "X_context = padded_docs_knowledge_text\n",
    "X_answers = padded_docs_candidate_ans\n",
    "\n",
    "X_Question_train,X_Question_test,X_Answer_train,X_Answer_test,Y_train, Y_test = \\\n",
    "train_test_split(X_question,X_answers,labels,test_size=0.25, random_state=42)\n",
    "print(X_Question_train.shape, X_Question_test.shape,X_Answer_train.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "# X_Question_train, X_Question_test,X_Answers_train,X_Answers_test,Y_train, Y_test = train_test_split(X_Question,X_Answers,Y, test_size=0.30)\n",
    "# print(X_Question_train.shape ,X_Question_test.shape, X_Answers_train.shape, X_Answers_test.shape,Y_train.shape,Y_test.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Create an Embedding layer for both question and candidate answers \n",
    "# note that we set trainable = False so as to keep the embeddings fixed\n",
    "# embedding_layer_questions = Embedding(num_words,\n",
    "#                             EMBEDDING_DIM,\n",
    "#                             embeddings_initializer=Constant(embedding_matrix),\n",
    "#                             input_length=MAX_SEQUENCE_LENGTH,\n",
    "#                             trainable=False)\n",
    "\n",
    "\n",
    "# embedding_layer_candidate_answers = Embedding(num_words,\n",
    "#                             EMBEDDING_DIM,\n",
    "#                             embeddings_initializer=Constant(embedding_matrix_candidate_ans),\n",
    "#                             input_length=MAX_SEQUENCE_LENGTH,mask_zero=True,\n",
    "#                             trainable=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(5250, 300)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "Text(0.5,1,'example vectors')"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXYAAAEICAYAAABLdt/UAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMi4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvhp/UCwAAIABJREFUeJzsvXeYJUd57/+p7hPnTN6d2SStJJTIQQiwsQ3YwLWJxgF8sa/tB67BxjbG/pGMMSB8AdtgsjEIRDKSkJBQAoQEQmElAStpV6vV5hwmp5NPn471+6PqpJkzO7va2V3vUX2eZ545oU91dXf1t95637eqhZQSg8FgMHQO1pmugMFgMBiWFyPsBoPB0GEYYTcYDIYOwwi7wWAwdBhG2A0Gg6HDMMJuMBgMHYYRdsOTCiHEFUKIq890PQyGU4kRdoPhDCKEOCSEeMWZroehszDCbjCcpQghYme6Dob/mRhhNywbQoi1QojvCyGmhRAHhRB/1/Td7UKITze9v04I8Q39+kIhxN1CiFkhxIwQ4hohRH/TtoeEEO8VQmwVQpSFEF8XQqwSQvxYCFEUQtwlhBjQ254vhJBCiLcLIcaEEONCiPcco86/IoT4uRAiJ4R4TAjxskW2e78Q4sZ5n31eCPEF/bpP12tcCDEqhPiYEMJu2vZtQoidur47hBCXCSG+A6wHfiCEKAkh3qe3fb0QYruu071CiKfNOxfvF0JsBcpCiJh+P6rL3i2EePnxXTFDxyKlNH/m76T/UEbCJuDDQAJ4CnAA+G39/WpgCvgt4E/0dz36u4uAVwJJYAjYAHyuqexDwC+BVcA6Xc5m4HlACrgb+Ije9nxAAt8FMsCzgGngFfr7K4Cr9et1wCzwal3/V+r3Q22O7zyg0lRnGxgHfkW/vxm4Uu9zGHgI+Ev93RuBUeAFgNDHe17Tsb2iaT+XAGVdlzjwPmAfkGjafgtwLpAGLgWOAmubjv/CM90ezN+Z/TvjFTB/nfEHvAg4Mu+zDwDfbHr/B1qEZoBfP0ZZbwAebXp/CPiTpvffB77c9P6dwC36dU3Yn9r0/SeBr+vXzcL+fuA78/Z9J/Dni9TrAeDP9OtXAvv161WAC6Sbtn0zcE9Tme9apMz5wv4h4HtN7y3dKbysafu3Nn1/EaqjewUQP9PtwPz9z/gzrhjDcnEesFa7D3JCiBzwTyjRq/EDlKW7W0r5QO1D7Va5TrsTCsDVwMp55U82vXbavO+et/3RpteHgbWL1PmN8+r868CaRY7xWpRgA/yxfl8rJw6MN5VzJcpyB2Vd71+kzPms1fUFQEoZ6WNZ1+7YpJT7gL9HdVhT+jy2O1bDkwgj7Ibl4ihwUErZ3/TXI6V8ddM2Hwd2AmuEEG9u+vwTKCv7WVLKXuD/oFwWJ8O5Ta/XA2OL1Pk78+qckVL+2yJl3gC8TAhxDvB7NIT9KMpiX9lUTq+U8hlN31+4SJnzl1cdQ3UUAAghhD6W0cV+I6W8Vkr56/p3Evj3RfZleJJghN2wXDwEFHUgLy2EsIUQzxRCvABACPES4C3AnwF/DnxRCFGzQnuAEpDXn713GerzISFElxDiGXq/17fZ5mrgdUKI39b1TQkhasK9ACnlNHAv8E1UJ7ZTfz4O/AT4tBCiVwhh6YDwS/VPrwLeI4R4vlBcJISoifckKh5R43vAa4QQLxdCxIF3ozqNn7erkxDiUiHEbwkhkkAVNXqJju8UGToVI+yGZUFKGQKvBZ4LHET50a8C+oQQvcB/A38rpRyVUt4PfB34prZIPwpcBuSBHwE3LUOV7kMFHX8G/IeU8idt6nwU+F2Uy2gaZVm/l2PfF9ei/NnXzvv8z1BB4x1AFrgR7dKRUt6AGq1cCxSBW4BB/bt/Bf5Zu3DeI6XcjRqxfBF1Dl8HvE5K6S1SnyTwb3rbCZT75wPHqL/hSYCQ0jxow9A5CCHOR3UscSllcGZrYzCcGYzFbjAYDB2GEXaDwWDoME7aFSOESKEmlCSBGHCjlPIjy1A3g8FgMDwBlkPYBZCRUpZ0FP8B1GSMXy5HBQ0Gg8FwYpz0IkJS9Qwl/Tau/47ZW6xcuVKef/75J7trg8FgeFKxadOmGSnl0FLbLcvqcHqxo02o6c1fklJubLPN24G3A6xfv55HHnlkOXZtMBgMTxqEEIeX3mqZgqdSylBK+VzgHOCFQohnttnmq1LKy6WUlw8NLdnhGAwGg+EJsqxZMVLKHHAP8DvLWa7BYDAYjp+TFnYhxFBt7WwhRBq16t2uky3XYDAYDE+M5fCxrwG+rf3sFmrJ0R8uQ7kGg8FgeAIsR1bMVtQDDwwGg8HwPwAz89RgMBg6DCPsBoPB0GF0nrBX5mDbcqz6ajAYDGcnnSfs22+CG98CTvZM18RgMBjOCJ0n7IF+HkFoluI2GAxPTjpP2KV+KpgMz2w9DAaD4QzRwcJuHvtoMBienHSgsGtL3Qi7wWB4ktKBwm4sdoPB8OSm84Q9MsJuMBie3HSesNcEPTLBU4PB8OSkc4X9JB/5ZzAYDGcrHSjsJnhqMBie3HSgsBsfu8FgeHLTecIeHdtir3gBn/nJbrzACL/BYOhMOk/Yl5h5+tDBOb5w9z4eH82fxkoZDAbD6aODhb29RR5GKqgameCqwWDoUJ68wh4ZYTcYDJ1Jxwn7SFDiP/v7kIvksdcsdaPrBoOhU+k4Yb+vOsGVA31kvULb78N6mrtRdoPB0Jl0nLCH2gUTRe3XYw+1oIdG2A0GQ4fSccIutbAv6oqJjCvGYDB0Nh0n7CHaYl8k3dFkxRgMhk6n44Q9WsJir7lgjI/dYDB0Kh0o7DUf+rFdMaGZeGowGDqUjhP2evBUHjt4alwxBoOhU+k4YT/e4KlxxRgMhk6l44S9ETxdakmB01Ylg8FgOK10nLA3ZpYuIuyydTuDwWDoNDpP2Dn2BKVG8NQIu8Fg6Ew6T9jrM0+XSnc8bVUyGAyG00oHCrv6L80EJYPB8CSl84R9iZmnZkkBg8HQ6Zy0sAshzhVC3COE2CGE2C6EeNdyVOyJ0lgEbLHgqVmP3WAwdDaxZSgjAN4tpdwshOgBNgkhfiql3LEMZZ8wETUf+lIWuxF2g8HQmZy0xS6lHJdSbtavi8BOYN3JlvtEWWpJgdA8aMNgMHQ4y+pjF0KcDzwP2Njmu7cLIR4RQjwyPT29nLttoeZjl4tOUNLbGYvdYDB0KMsm7EKIbuD7wN9LKRc8vkhK+VUp5eVSysuHhoaWa7cLqE9QWvLReEbYDQZDZ7Iswi6EiKNE/Rop5U3LUeYTJWSJmafmYdYGg6HDWY6sGAF8HdgppfzMyVfp5JDy2MFTs1aMwWDodJbDYv814E+B3xJCbNF/r16Gcp8QNYs9NK4Yg8HwJOWk0x2llA8AYhnqsiw0hNvMPDUYDE9OOnDmac0V097HHpl0R4PB0OF0rLAvvR67UXaDwdCZdJywh0tY7LU8dqPrBoOhU+k4Ya8J9qIPszZrxRgMhg6n44T9ePPYQ2OyGwyGDqXjhH2pRcDMWjEGg6HT6VhhX8xir7lgpLHYDQZDh9KBwq7/m6wYg8HwJKXzhL0eHD12HnvY/muDwWA46+k4Ya951pey2I0rxmAwdCodJ+z14CmLPRpPb2eE3WAwdCgdKOz6/xLBU5MVYzAYOpUOFPbjzGM3ym4wGDqUDhR2/X+JPHbjYzcYDJ1KBwr78eWxG4PdYDB0Kh0n7DU7fdFFwMyDNgwGQ4fTccJek+uI9sJtLHaDwdDpdJywNx6Nt4TFbpTdYDB0KB0o7Ao3DNp/r/XeuGIMBkOn0nHCXnfFLLakgHHFGAyGDqfjhH3JJQVMuqPBYOhwOk7YpVD/F3uQRmQetGEwGDqcjhP247HYRSxrXDEGg6Fj6Sxhl5IIZbIv9sxT1z5E98X/TikaPZ01MxgMhtNGhwl7RKRdMYtlvUQUAfBk8XTVymAwGE4rnSXsUbj0E5S0syaM2lv0BoPBcLbTWcIuI0Ltill0rRj9+WKLhBkMBsPZTocJe1jPilnMFSO1xR5hhN1gMHQmHSbs0ZJZMTVLfbHvDQaD4Wyn44R9qUXApHHFGAyGDqezhD0KCUXNx75YVowOnhphNxgMHUpnCbuUSz/ztO6KMcJuMBg6kw4T9qZ0x8VcMahVH6URdoPB0KEsi7ALIb4hhJgSQmxbjvKeKDIKkeLY6Y7rGQNgRTB22uplMBgMp5Plsti/BfzOMpX1hAmjxhrs7Sx2KSVx4QNgSfe01ctgMBhOJ8si7FLKDcDccpR1UvWI/PrrdsHTSILQwVMh/QXfGwwGQydw2nzsQoi3CyEeEUI8Mj09fUr20WyxyzYWexhJhKi5aNo/YclgMBjOdk6bsEspvyqlvFxKefnQ0NAp2UfUtP5Le4tdIkRtGyPsBoOhM+morJioxWJfGDwNI1l3xWCyYgwGQ4fSUcIeyoawt5PtUBpXjMFg6HyWK93xu8AvgEuFECNCiP+7HOWeKDJsstjbuWKafezSCLvBYOhMYstRiJTyzctRzslyPMHThi1vXDEGg6Ez6ShXTCSPncduXDEGg+HJQGcJ+xIWexQBwljsBoOhs+lYYV/UYq9/boTdYDB0Jp0l7LI5j73N95FE1l0xRtgNncO+qSJ+aB4eY1B0lrCHDbFeNHgqGus/GgydQKHq86rP389tW8zCdgZFRwl7cx57W2GXkpqgS2Oxt+XbPz/E4dnyma6G4QQouwF+KJktm4XtDIqOEvaWJQXaBk8bFrswFvsCSm7AR27bzi2PGsvvbMILVFt2vCdZm3ayEJ7Z7DYppU6jXpqtIzmedcWdPLB35hTXqtOEvSXdcSHKYlcXod2SA0928o5a8bLimVTQs4mab93xn0Sj0CiEL1wGm791Rqtx1f0HufCfbqdQXXq12IoXUqwGWOLU16ujhD2MTsDHLp5EN8Fxkq+oxlk2wr4sFKo++6dLp3w/XqDaevU0Cvumw3N868GDp21/C/DK4MxBfuTM1QH4z3v2AVCsLn3P1K5PMm6f0jpBhwm7bM5jb9MrRlFD8I3FvpCGxW46veXgy/fu5w+//PNTvh+vZrGfxut246YRPnvX3tO2vwV4Og7kV89cHWjcM36wtJ7UhD0VP/Wy21HCHspjL9sbyka649ko7NWgyv7c/lNWfl3YXSPsy8FI1iFb8XGDU3s+z4QrpupHBGcyvdKvqP+Bc+bq0IR7XMKutkkbi/3EkNGxG3YYSaSoCf7ZJ+yf2fQZ3nDrG5gsT56S8gs1YT9egRh/DGZPXUcDMFNy+eLP9qrA9/9gfrh1jC/+rNWCnSmqLJVah3mqqFmLp9MVU/VD/PAMXhNPubhcZ14G150fhF23L8supopVto7kjq86J2SxG2E/IZot9lC0f9BGLXgatfn+hNn0bZjccfLlHCeHC4cB2Dqz9ZSU37DYj9PHfstfw8/+5ZTUpcZdOyb59E/3cGSuckr3c7LcvHmU6x4+2vLZTEkLe+XUCntQzXMo9cc8I3/fKd1PM1U/xAujtquoPlEqfoUPPvBB5qpLP2XTc4oA/HLPaONDKWHjlbDjlmWpz1fuPcBbv/Xwot83X9fjGZUZYX+CNM88bdfcltVilxJ+9P/B5m+fXDknwMX9FwPw+Mzjp6T8mrCXj9dX6+SgenwWzROl5l74n+73ny65dV93jZqw55bDYvfKcMvfQHl2wVep2V0AvKrwvZPfz3FScysEyziS2p3dzW37b2PL1JYlty0WcnxysJ/9FOqfffi6DRD5UFqeEW3e8Zkte4uOFpuNjeOy2PU2xsd+grQuAtbue1nPb2+XNXNC+BWIAnCLJ1fOCSCEigg/NvXYKSm/lrLlHG9WjFcC7xiW9B0fgGv/6KTqVBMQx29TJylh7FH1/wwzXXRbpvT7YURWW3TLYrGPb4UtV8ORhcFY4ajONUfPye/nOKlqCzVYzB1TnIRHrz6BAgtU7vk4AF7kLbn5XH6W7/T1sjNTrZ/3zdt3Nfa9CPtz+zmUP3TMsvdNldiwZ5pqECLl4hkvh+cabqDj87Friz1mLPYTojZBScjFZ57WLHZ5sq6YqrYUqvmTKwfYOV7gnl1TS27nharBb5/djh8t//C+ZrEPuCPgtw9KRTIikroRe+VGEKsds/tgeveJV+TQg/DNV0Po12+GtpNvRjfDV18GY5tPfB/LSBRJZkpui9U2V26IUzuL/ef7ZnjNF+4/LksPaJznNoZEzFFClpPdJ1Drk6PW4c4fpdR57Ltwa/sRRlvGHqVy+AEA/HDptp0vZ9X+7YDdE2qdnP5IfXYsi/0Nt76B193yurbfzZRcDs+W+fK9+/nB967i1ZNXArTNUT84U+ZjP9xZf388wu74IQnbwjoNiewdJew1H7stj+GKqfnYT9YVUxN0faP9yw928KOt40+oqCvv288/37Jtye1qlowbuuSqOYIoqPvdFzC7Hz62Gmb2HXc98o6PIOLm4G/h+v/Tdps///Gf85+P/icEnhr2esfI0w6qjbQ0UNb98dzoIw/B4QfBydUtw7aTpiq6rEp26TJPITnHxw9li0hPF11ebG3jH2PfbRs8fWwkz/axAjlnaesUaBL2hec7WVbtLi+7TrzyTxBXd7iLLjxWvzazx9e5u0UqlpKjmgFzLGrC7loRjx7NUXFDhtHtwJlT7fME+cTtO3n7f28iW/F4qf8ALy8oX3276/edXxwmW/H44pufx4VilGf88t2Uqzlu3nvzonEH149OixsGOkzYaz52Gwjb5bFLWQ+anrTF7hZa/t+6ZZTUho/BjttOuKiSG1I6joBlc4Mv+2W+t/t7vPbm1/LYdBvXzPabVSrYluMfDucdnyS6Ee+7q+02hwuHOZg/2BB0r8INjxzl9sfbdGqB2yrs934Cvq2tpamd8Nh17StSE6/Qw/WPkcpXS3ULj2+NlAdGH6DgFZbe8ASZ1tkvQSTr/tiZksub7Hv5q9gPKJUWWtllfb1rx7ckNZeXu7D+yYo699ZxuDCOl7ybJ1tdvMOsLiXsjv7tI9+AL7946Q7dLeBoV6N7HNezrEfMVStiJFuh7AUMi6Z4T/nYI+C8u3CkfXSuwnjeIVfx6JYlUrJKCretsG88OMtl6wd47rn9vMTayrkjP+TuXTfw4Z9/mKPFowu2B3XOTkfgFDpI2L0gIueoBmFL0dZir/pRwxVzsj72eRa7F0a8fOZq+N6fnnBRjh8c1zT+5iFq2S8z7UwDcOu+W5FSMlVomqyR7Gmp32LMlb26hVET9u93Zzgaa//URCdwKPrFhmB7Za66/yBff6DNLMSgCn5ZP+EEyI82briNX4Hb/q69f7xWdug1uWLaCHttckqwtBAUvSLvuOsdvPNn71xy2xOlJuzQcE3MlDwuFHrNnTazI2sd+fEM4QF1HqHtCCldVa6HeKjOh5SSL235ErvnnoAbTPPRX3yU9294f8tnh2fLbDqsBLsWCPSDRe6jmrBP7VCxqMpSwl6kol0UI7niktk2Fd2uq7ak7AaU3XnCvkQA9VDh0ILPJgsuhWrAbNmjD3WeV1CopwHXyDs+O8YLvOgpgyRjFt0oA8OpqDVgyn77RfQKXg76NjBaGm37/XLSMcJ+46YRrn/oEADWIq4YxwvrDpjFLPYoknzhZ3uZLR1bLLbObidnWXVfe/BEFyNyi8Sc2QVD+XY0B5VKfolEoBrcfSP3ccPWzbzs2j9g95QWzkSmXv5iZMsel/2/n/KpO5UAFByfpOVwxdAKbu3OLNg+khHVsMpspUA2p29cv0K+4jGabeOTrwluzY3gV6A26iiMKUvba3MT1MSrycfeNiumbrEvbalGunPZPPUoubIL17wJdt+x5O+Oh6lio0OtCfVM0eFCoSzpRBsLrm6xH+/kpVrMo8317HImAIjLqurgK1N85bGv8O773n3cxzCf6co04+XWUdjnf7aXd39PZazUOlo/WsJizx3R9V5ipOQWqAglR1c9sJcH9x27I3B81UYqFpTdkLIXtgi7bBNAbTaM5gdQpZRMasNoNOvUhX1QFBdY7I8cmkNKeNEFK0jGbHqEujaeo9I0K0H7uNOcfwSn5+ZFLfrlpGOEfbJQrQdPLURbi9zxw7orJoK2VsGjR7N85qd7eN+Ni+eKSyn5iz3f5prenvqNlgif4Ay4b76Kb8/+sarfEil9buhi6cZf8kuUp5RffqoyxR0H78BOH2HL+N5aJdX/sS1wRR8c3LCgvAMzqvH+1737iSJJ3vEZ7lZi7LQJ8FQD1fAPzs1w9Ybt+lOJWy0zWawuHJbPF3av3FiNr6CtWW3J7Zks8vm79qpr4jVcMdVjumLc1v/HIO/WtpFsvO1K2HsnHFyevO9mi712DuZmtrA1rT5Llxda7LX1eI47eOot4mOXkm5XiVgaDzeI2JdTcZWu2BPzuT+4b4aZcnGB2ypf8Sm5IVLKeuxjSVdMbbSylLBXC3WLHREwVzl2Z+1qq7hsqdFPxQ0YEjnGrdVI4AUP/RP//tC/t/zmy/c34ljzLfaCE9Q75SCS9AttsYv8AmHfNaHu+Wef00cybtGDujaePkZnkdmwpVBdp3O6zznmsS0HZ6WwT+SrbNgz3fJZNQhVOgzKYm/X3Jot9kjItl4AS/v5JrUVNlGeIIharXEncHBkQNGyIHAIfY8e+QQXe5pQOekDPQ/xpce+2HRA+QVBJz/0GUgOAGoyR6mpAU1V1PmYqeh6aBFmRpdx/6eV2G+/BfbfA+VZDs82LIu7d03hh5LhjLqhHCEaLpSm4wYIcSiXGj5Kyy8jpbouAH7kq7/AxRU0hNqvqIAr1IX9/375TgA+ecduPnvXHu7fO9PqigmO5Yo5hsXuV+H6P62fw7BJgLon9aST8vIsn9riitHi8JPyl3jbmlVUhKCnOm8Z5Kmd/O7EFxFEuIHKMpooTxx7J1rI5HyBrMwRi9T+08Kl6ofszarOfXVm9RM6nj+5aiNHcjkKbqHF+Cm5AW59YpKu1jxXTBRJlUVS0ZOMate7ulDY3SDkh1vHVFzCLdYt9nfEb6J78hF1vFLyrW3fWjBpyYvUtS9bglI1oOQGDJNlr3UBB+IxXBly9c7W+NJtWw/UX89fmqN51CWI6EOd7xUstNhr17grYZOwLXqEuo9c7bdfTNjL0SRI6wlflxPhrBN2xwv5069v5M++8VBLFkrVCxF6HRhbiraLgCmLXb2W1JbxbaW2tnLZDSn7ZV5782u589CdLdvULBlXl+VXcvWh2wkTVy6P/t6H+cHB7zc+/+Zr4EsvbNnUizwGUkrYS36JctOQL+uqhj9X0UP1+emKU7tU+uENfw7feQN86kKi/ffWv946qhplJqUaeFVYC/yi1ZoP13IInMaNmhZKWEZzap+v+v6r+NVrf5U39gkuP399kz9eu2J8R2UuAEFphnzFZ11/Sh32gwebhH0pV8wxfOzTu2DnbbD3p+p0ND1dy65od9VSft/jZKqNsAeRujYbugYY9Me5+8jd/MWdf6GEcs8d/HbxZgYo4QYRdx+5m1d9/1XMOIt3NLNZZQHnc60C5xdGeefwSrYnEqTwcPyQvTkl7DGrfZykmVtv+AN+fmcbl43wCGTQIlJlT1m11aaA7/x0xy/evY9nX3En0pkXeG3jQvrA9x/nb699lIcOzSlXjLbYPSFYd/QHAIyWRvn0pk8vuAf9SF37smXhuBUqXsgKUWBCDnJP7woAzus9r+U3lUC1q9WZ1dw/cj/vuP7H/P53P8LHfvkxJgsu68UkzxH76MHB1kbiCpFfkO7ohREJ20IIgWUJ+rQrxtfHuJiwO3KaBCuP67qcLGedsH/j/r0cnZrl/BVd/PMtjRmYjh/WH55h6VefuH1ny6wxxw8b6Y5C4ocRV963v2WNjTknT+bCT1GU+yh6RdzQZdZpFYCipy6gqy2MwMnTJ47x1KHAVelXO25bOGljpZpN2mPlKfvFRo74pD62ps7HCz0GU4OACtCUmrIHyoHyL+adHDz89YXCXppQbhmAF/8dILFndjGYSajfa59vMq5+V7WE+k0Tji5TiBC/yXLsQtVjTAv7ZGUSN3TZX8sAqK/Ep//nGj7GAYocnivj6+t0z+5pCnktCi2umDYxjJqwt8uiyOt9aFeA3zTqynh6tFdZKKR/cedfcNXjVy0s7xi0C56eH8QBuLtniKFggk2Tm9g4sZEH9o9x68PKWswIh6NzFb6z6WECGTBeWjxdtlxS5zt0WgXy6OwO7s108Terhknj4ngNi71Sc4E99DU1wakNXyju4prxhW46YalRULM7puwqa7159FTxnZYO6f6906TwEPOvyTxhnyt73PToaP01brGeFeMJgWspN1ItEDnfYg9lo3zXm6Ps+mSoUohS/DShhDNlp1p+Uw1VWe+5/D2kYinumbqandlH2Ti+kclClX+I3cjn4l+qu2Gg5mNXbSfv+Dx2NIcfRMTthuVY87G72u/vLDIHxBPTpMVQ2++Wm7NO2M/f9TXuSn+A37/sHLIVv77CnONHdYvdkhYS+OqGA+yebDQoxwvraZAh8L2Hj/Kvd2zlv+5p5HqPFEewErNUrYP1YMv8yUANYddlVfL1oVtbPrEOvvh8uOltatLG4zfWv5JayFNWCYmk5Jdaff9NguRFHt3xbmIiRtkvUw5dBrUl6uup1YnsY2qpg8MPNsoQSmALB7QL4ikvAwR+aZaLhtSklpKeXZeIK7GsCLEgs6DZEvHDhismQyPo1Jam1EgAco3c+0FR5PBsBd8p8+LuSZ62uqcRmD3urJg2rpiab1cLfPNTbvqlntxSmF4wzN4+u51Nk5vaH8ciTJdcEraOfbgOuWqOmLbYH01GrJFT5PQw/d03bmRsVnXC3VT50dZxfnlYZRTVxcstwv2faVmS1nPKuAI+nypyKH+I3/7ub7BtcjOBHnXMxizSeBRdlwN55XIo+2UIffjx++Chr+IETotb0XOLTFuCbDh/6VsJlp41W0t7PbiBlzjqdfM5+9zj/8hvfu836++fMpRhoN3odZ4L6fubGnGHmZLbmscuBI5QAYpam5tzGsIehBEBjTr4XpaqU8YWkkKY4LDOFc+5eX566G62zSjf+spQnef1Pet5xuDzsBLTCNuh6BWZKrr0UmFI5OkqPw12AAAgAElEQVSnxJFYjJ+nUqwUBWacGTZNbuK/f36IN135C7ww4pn2YfjhP4CU9NaCp7oTWsxiD6xpMtZw2++Wm7NO2PvLBzlHTjAglLjWViJUAqB97C0PvpN1C6bqN/vYYaJyhO5Lr+BAYU9966yrbjqPfD0LxZ98vOEzpEnYdUMMnUKjl4+1WglqZz7kj8DKS9T7jV9p1E6LnW3pBaPcPP/xkybfetAcmPNJ2AkyiQwlr0Qp8lipfdDCVnUKPF3PZv/x8NMBuOMRPeU6kYH0AFTmuGBlBktASQfzbKvmihFq9cYm0aw2CUAQNjrMmitmLL/IbFW3iBM4ZGsB5uyh+nf9osiRuQovnLmJa4N/4MrEZ0jLmu/crwe0Kl5IFEm2HG1KaVskj93xQn78wEPqjRb4ZkFLWkpkotI037j/ALMlFzcIiWRE2S8zXhpnIl897lUZp4sua7Qr6fp9V/GWO9+C1J3dpHDpocScnvbfnQ5Ioc5pBoec4yNi6lweLhzmqsevItxyLfzso/DAZ+v7CN0yOxMJbs7A1Y98ljEvx+MbPo7T1PmmhMuW6UfqeeDloAylSW7sTnPX3s288JoX8nd3/119+4mpbUghmCNojaeIhluz8NCX1Gcbr+SvgmsByDUFNvcUHgVAhiF4ZYrVoMXirTPPYr9lyyjPWNuLECo1tDl46gmBK9WIp5Zh0myx5xyfyGpcmyDM4elVHrNhjLK2pieKc7zvniv4xrZvIKXkb2JqtNxtp3G9JMJ2ELZDyS8xWajSY3l0iyqrRJYr+3t57/AKBilwNLydd9z1DmZKLm4QUXZDXi3uZ/zRbyOdHLb2sXu6I3bu+tCCFNeCV0BaFXrsVQvPzSngrBP2pK9ukFWeOnG1tcOVsNdcMZb2sUvuHrmTP7jtDxgpjuD4DYs9EjDr70eIiKx/pF5+XlsWVqxUnxDkb7upZbGvuo/dVkO+0Gmy2OM6FWLsUe564EH+5Gu/aKq9thqbUvykbgye9ukVvALf/nnDovU9pz493Qs9EnaC7ni3stgjn6GwJuy6rr62jJrdDMNPBWi4i2IpovQA6SDH+hVdJGJW3RUjLZ2Tawm1cuODnwPgv+7dx+YjDdeMbTf8+y+ydvGc5DhjufYPPShvuYY/vP7lvGSt8n3WLPaqSDFIkcOzZVa66hrM5R/kw6vjyh4L3XqHXfRzbNg7zRu+9CAHZ/Rx1LNi1LFLKfnuQ0fYNVFA1i32miumYfHnYqoRpIRPrpDndV98gK/ce4CyX0YiGSuP8Y5v3sunfrj0YlRVPyTv+PyNfTPfjv8bM9Vpxktj+Pph6VJAPibIaouzKxWQ1MLeLarkKx5WTF2zG/bcwOc3f55dtc55+031/UivTF4bEjv1bOPp7AEqlUYSQcUO2Dh5H+lYmlee90oqfoUgN8JHV67gAyvUPu4fvb++/dismhI/Z1lqfkFxEsqzWFZDuAva3SWrhXocKef4xAnoptEGqre8HT6xloLjLSLsDYt931SJ7WMFfv+ycxjsStQt9porxhcCoTvtusVenSMII/78Gw/xk+2TyKYnoIWyiKdjSwVLjU57QoGwPXyRp+yXcYOIku44MvvvoVSJI6wKll3FDV0mi0V69D10gRhnPBajYNtkrDyOnMIJHAquOt6S6zPevZ/Xn7OGzz72JV57fjd5S+Bpw8cRFsw0lnDOOz6jReV26ouf+sApnIXC3hWoIe1KLey1tDGnKXiqLHaBQLJtdjMSyWhplIoXUmsOIVAMVbZCOWxYgQVPvRaxIhVfzygUtPiF68KurfOoWqC/7mPXPcdXX8Yr7no1O/YfalR+fvqffr0vWltPL8xv+z4Xr2qs+fH9jft5xWfuY9tonrlKhbiIk4lnKPklSjKoC3uNaTHHH61dxVzzrMGhS4FWYS/bffRT4uLhbhK2VXfFSG2xF+J6gtPoZqSUfO6uvWzY18jusOyGdf6u2E1cEft2vXOYz+ShezkSqBtPghLbRA+zsWEGtCum11cd0YdXruDBrjQH4vF68DTWs5Vt9ns4qK9BfR0Wv9Vi3zle5AM3Pc5tj42xTuiOrTIDvkPYZJHO2o1m7+WnGMtXOTxbpqRdRk7g8N7CB3jJ4aYspTbsntvNfYfVyOBNhW/zUnsr1aBCJXCoWAKhI/gztk1OX49kwicllLWZQY0KahZ7LQVvqhbcndkDBeV3F36FvK1Ea29JnYfZoITTZMkeSEi2zD7Ab6z7DQZTg5T9MjN6klK1TfrqmM4McSwLZ2wzfPoSuPaNZJKNc1XQbsioWqBLuPWZmO+M3cQtiQ/Xt6voTqjsOAzZqp256SbrtCkrZtPhObDL7PCuIjZ8C4+PzVAqzNVdMa4QCC2StTjBXHWOubLHfXumuXXLKIHVaPdxq0y5pGNMuozhQP0XIqLklamUSyqLDejadiu5UlwZQ3qkXPCKdImasE8wrhfqEvESvl6qIKcnJZbdEC+Wo2pZfHPP9QBMxLpxteFWsQTYasRx5/YJnvPRn/DwUdUZ9yVWLLgOp4KzStillPREqoH0O+pE1Xyv1SCsW0IWNpEAC8m+gvKvTTvTyhWj23cgBAVf3TROmFPpd1f0YedV0FLEisyWVaPyhWgZWtVdMfriyWqeXm3NyHmpd68ZbAqI1YW9KbUqcNgmz1e9PFB45Gtk8w2LZzZXYK7s8XfXPUo1qBKM7yITyyiLnZAVYUjzLbsvXmRHMsl+bZX6wKM9K5DQyNyJJZkOuhgURV70lBUkYhYlN+BiMUIo1U2Zj2Xgaa+D2X3kHR8viMi7TXEEq9U6X8+YCk63yTS6rWmyU9ayVCpnooui6GWQEkfmKvSFs1SsDMO6ozoaj0GosjzifVtAhOwtqKUTHC9k72SxNSvGqzCw4UOsZpYjc3P8cEWJ0YTuIPOjLRb7nN2Y1l3OqlHIL5yPtQRNU/YEK9xjTyT53ObP8fktKlf6pu4Mnxrsx/fySCRZ26YLdRNP2zZFPZJKJPyGK0Y4lL0AEWv1P09Xm4L1eoEzK3DqouVoF+G0kDhOwyh5qCtJMcjyojUvIhNXbeSoXiso1ua6jBUbbTp76zv0h4/iyca1LUg9GtLCPECJfMXnUjHCkNUYLdQME9cp8dR+da4L6XWNnblFJsuTfPyXH2fvVJ5UzyF+cvQ2KqkN7Jh7lHRUrqc7egIy7hR85uk408p9OFudrc/W3TVRJBSNzse2KpSLqn4lfWlXNd2GuWoRNztCybJIRRF2eZaqm2g5F2W/REonAZxvjTGpZ16HsRKRrc5x3aCrlvHmtf9DibV1r78jRD2edPNmZakfnFOGRne8l9PBWSXsFS9ke9rlk4P99FbU0L1mJTpeyLmW8jdGkU0EWFaFCd0BTFemVfBUlxUCWU+JrisbOePdeRVgtGMFZpYUdpv3DK3g36Z/Qlz7uAmqKmCleUv8J40DCOdZmlJiBQ5H5HDdv1ggYl2pKdvHUXUY6EoQEy79Y78gUy0wV50jBHqiiARNQmWrGzhrWVy97mJefNHF/NnWz/JIKkmv9gUSS3HESTEcq9DnT5O0IHDL/DDxQcirG8mTvooJZA8ymVUdQtFtjDSk7ZGTGaZti5xlsSKcIfQqbZdcvb27MVFmOmYrV5Sd5OaM4Ir1RSYKVVaEM4ykLmVYxwwOxOPIwKMaOtgZNaw9UlYPNbnu4SO88rMbmMsXGuf1wL2s2fUtbkhewWPhx7iuP8Vt/Reo7/NHW3zsM03C7uSnQASUxG4eHGsEnGfjkkxw7JU7Z51Z8p6y5u7KdHFnpgtRVAG6nGWRstfU91fWqXa27daFvZsqWA7CmjfqqubATuo3u/HDiISsUrBab9cZ28Zpyth6NKV+syazhkw8gx/5bBpR8aN4O2GvNNZTyWpfthy4AL9J2PO1ZxxoH/mAKJJzPFaJLLPxRpmPJZN8o68H3ylzabpIKAXT9hp2JOK8c3gIp5rjgdEHuG73deyY2cvK3sb1SMWnsYVs8bGvdA5BYRQnr+7folck76h6FR0Xz5L0SXU+LNuhWlZtoWKpOq32G3UremVCLexdEQhnjkq1VdidoExKH/eK+KS654GJmFWPXw07m/lU7CsMOofJ2a1rvuyLr8PV9Xcsq54wcDRbYQV5hvd/E4De5OlZWvmsEvbZfIlfZmxu7Ommq6QueC146noB5wrVUKNIrRVjpxu+6mlnmorvI/UFixAUqyp7wJV5iKmbIgjU0FbYZWbL6uL4zBN27S91hcWd3Rl+5h3mjiH9uyiAcsOSuTDf8LE/LnweTyQYjVy1kFZQxRGCnIg1LHbL4lfFVkZiNnd3pZmojBDr20RfOk4oJAkp6Z7Zy6Se0JKJJCm58DLe15Xm3xNuff7tjB2vxwEcmWBfOclQNA2ffw6vZQMxN09S+FT0xI9AerDiYogCCno2a6nJhXSBfYgZ2ce7hof46EqVgjnojjVS7JqYjMXqwjJl22r2ZCzBwbhgLCGIyyoDFJjIPLU+ojqQiBP4LrHMHoQVIMIeHsvfSfq8r/Do1GYyF36SvbXhfeDCuPKHV5IFXFu5jKYSSljJHmp5IMT+RLz+uifM8+KEukbNU+jHYjF6o2MLe87NUQ6KgGTatqkIi0jHgCIh6LaVP/VoPEbNpPAip0nYHazYwvzuab8I/edCzxqY2cNkoUoXLjM1sdfM2BaOTkntClLsSiixqgk7wL6CshhjTbpey/Ia83Kk9OeztgXnvgiqOYRo8rETKQNEW6sDokiu4qtZnk3rCd3Y081nBwcIvCLn+Qc4ItYxGfVwV6aLezNpNrs5sq7qBEfyM/R3N4Q9EZ/Fh7qY+gi69X1YaVp3Zbyk7qsBlD9+SKczCtsh707gCEHFVpb8uqYU5opfIcyPURKCWBRDVHOUndZzWQlLJLWwe03xo23JRN27utbfxhtjG7jE3aZGnk3st4fx9IYVIepxtKNzFX7dehyrrDrY3kQfp4OzStjzc5MULYFjWSQKh4GQiZIS87X+IRK1BimVj93uOozAYjg9zExlBsdvWNJF2yIvGqmC+/L7kVD3Y0oBO2d0RoUA3Hx94a+itnQqNBrP5u5ArR0DbB/5OUdjNhNyoP79JF388You/njdav5toJt/3PB+8B1ee84abrjoofqEqmw8ya9bj/P5gX7+fnglj1g/Ir32BiYrY0gBCSnJuBWKOmc2E0Vthf1QXInXZ1/2WX28cbqFnk1biZiLtJsi9HiW3F2/ccv1dFCPjz2kzpc3oaz4ih5pWFLi2JIsXexJxNmlhXI4GGtZJ6N5+P8MV7sPbFtZM3aSou7MVsbUSKuQOY+yDkgfiMcJPBcroW5mWbxMldl1iGr37ViJOW5LlhmL2bzKeZwjoxsJrQRjTWIzZfdCqh/GHiVoikXsTCTwpbrOg6LIu1PfAGjMIQBGYzH6KbZdyyWMQsIoJFvNEkofYfnMxmwcS7QsxZC2u+mJdbEv3uhInLBMSnhsTCX58bnbEAklYOc0tc2psKKyllZeDNO7mSy4pHHJxlszruZsux4Q7I/6CLUwrs6sri8nMGerdhI0TYqZKM4wUhhje1jmeagys7YN5/2aauPNwVNLgFfC0nGMQYoUKi7D5BhremDEaI/qRFOizCpnDyOpi5jyEuzRnc3jqDRQgMnyHJm0T8yK0WMPY8dnG8sJoCz2bj0vw2kyFA7mDwKSYZHDsSy64xkucT1y3SNsPP+H/L8Vg1S1sJ/TJOyedKAwRsmyIEph+yVE0Gqxu0GJhM6NH286ri2phhsx1LPLX+hvImvPGz2JBF5MlelYStjdIKRQDbjIGqNgWViRTXeiTdbcKeCsEvZydoqiPqFBUKZ7/df4xLY3c/uB29l97hf5uLYcV3WniQSI9FFWJi7gnJ5zmHKmqPoL09fiYRzXGuH3Hv0kP+jO1P2YAHJKrSXiC4EHPO97L+Xmq36F4owSupyO2L866icQ8KdrVnFHpov//fBHefW56/hx2Jg5en1f44JOxmKMlkaZKY4yNW8VxdnkAE+3DnB/VxopBLMJ1blMR3qKdRQj0yRA3VKSbLM0wkhclbu2ey0AJStOzUaacwVZlLCPxGyeEh0gHWp3S9PxX39QlXF49xbszB5kfJwYCbqjiJKw6IlP4VoWo7EYVSFYHYy1TM5IRwuFfTKmhT2WoFwTIltZlV7Xakp65HQwHsPzqoh4DsJu3OzlrE48Q20XVyOIu9IhD6TTjODz+NxOxvovZ6JpiFyUEax7PoxuItAuhYs8j32JOCP0E0hLzVac90SbIbubfYk4XcIll19otf/zg//M39z9N/X0z+FMmVnLIhCipf2kYmlWJgc40DRCqIYVUvjc1p1hIl3BTh8C4BKv0TanIxdS/RQGn8IXvKPMlQt04VKIN8RISEkoBOOxGJYU9NIPQJIU3YnuusU+k9BPxWrySf/Vd+7gHbf/CxLJu9MXApB9ybsh3Y+IApKWEtNUFCn3T6nhshkQRaLyDHERtnSiM4FqPyutaXqqE2R7LmVvtZ89ulPbZkd1iz0SFXqsaXoDn5XxQUQ8Xx+xggqe6ryoFkPhK3veT6xvM7n132djOkU60c3vlCuUU8ooeaArhavTINc3pelKAsLiCDkrhgzVfdgz73F3kSxwJGbzs650/biioIc9iUa9fJ2v/gK2qY6wiSIRjq2FXShXzL4pdU4uEqMULItkaJ9d67ELIX5HCLFbCLFPCPGPy1FmO5zCdF14dibiiMwhAH508EdEVqPhdsVj+AhEepSh+KUMdQ0x48zgeKpHTjRd01634fOatG1ylsVQoCSwp6omqvhCsDcRJ5Ah/yXyFHQqYe0xe8/0JX+Yr3AoEednXel6eT+Knl9/vUHP8MxEkjnbIu8V+OTG7y44xiPxNDf2dFO2rLr7QkqLSkwFDsejYbqbMjy6o4h0G2GvBQjXZJQltT8e53kXrOemTC+HZqtkZQ85y+L156xlb2qqnjFTarKcSpbNvwwOsdfdRXrtdcR7t2NLm55IUrQtcnF1PqUQfHBoFeN9Oyg0LSubbuqA1gUBA2GoLfYy2AkqNbHp38JbVw8TdA3VLfaqZTHnFbDiWWLRIF51iJf0fJjQVRM8oiBDxYLre1QHNR46XNMzyJF4DFuq81KWoRL2qR1Eemj8dC8kEIJd8RQVVLrl5Dxhf44YZEcyQQQUZuetEliZY+uUmq1YP7aBSSLdSTX7XlN2mqH0Co40WezVoEwCj4fTSmBimQMg4emeEqJVXauYIoR0P9fZDl/rSbNt9IdYQlJuEtLz9EzcI7EYtoyTQY0Ou/Xj8WrCPhpfmA1TKeznqLuZ11aq7DuSwiLOnCUgpdwEaUu5h1YHoWpHTbnyg6KIXVbvH0mlSOm26GshflpMBWvdlU/nv51nMx6PISRsTTYyg4RdwY5m6A0C1tpprHi2brHbEhzRFAMJqvWF7wDs9BFKXaqjSSX7+J1yq+svsH1SocXKeZMK/co4s3YMEar7s2/eOkiRzPMfKwb4+1VD/McKdS6jamtqYqBHvJbwKVkWq4KgLqAlIsp6FOhoV8zROdURXCiUxZ6OLAa98ZYY3KnipIVdCGEDXwJeBTwdeLMQ4uknW247vMJMXdhv6GkI8qbJzayuJnjPlEtQvoCMlVSWveXTb13CUHqI6co0ge7Fm4V9yGucghiSgm2xXgdeaj5ED8H2pA5MBcGCIFaPW+QjczNcUpUtPfmBlM5lpmFBly3BrN7mkenbFxzj4yLLv64cJBNF/FU2T8Lvws9djkiroNxkNESmaSGcTBTVb675pKVFV7yLuBWvWx4fGe7n8FyFrOxhMmbjC0HWjni2peINRcsiUbNmLJcbe1Mc7M4iYuoGikWCnihiX3wd16/9k/q+fpJJsLV/lpFce7/0UBgyHGhhD6pgJ6hqYZ/uHufhdIrZVIxy07kteBXi8VkuCUtYROQdn6iqRiBB4blYUrInqTrMOzJdXGNt5YaebgYDdV4qMoRzLgcZEc8qH+elKBfFjlSCCkkGRYFJuyGYMRHjmb6qx6F4jEpOi9q9/w6//DLBTW9nrDTaEox96rr2x5yOpViRbp1C7kYVcnG/3ras1Cj9gcU5WqgvW3UZcxa4yR5uKKiR4UhWBfQLFvTomdaX6O23pdcQyi66hBqtZiIl6DVh99qkOa6LP06Iw4VVh4NOFyLqVhOAUsrqT1uqE3ya53EgHiMoNtJcByiSrE4xa1k8lkzw6nnCeqF1CIDe8y8jjCsBXlMeJGvb7JxVwW9hVwhw6I0intrdhxPzmBtSozE/7GkR9iP5AnYwzOvOUyugCtE47+nUAOcGAZdNXsjF+SGytk2YnCMZ2qTx69Y5gF+dYs4WyEAZAgPzxFVQZlLfl79ZrrDKuRwZqAyWvjAkKboJbPWbWuf91lyBHwZDpKOIMiG+/rzmiilWfWxCzhcT5C2LoajKy+54RX39olPJcljsLwT2SSkPSCk94Drgd5eh3AWE5dm6sE81WVllv8TqIOI5xW6cI39JvKlh9EarGcqPUwkqCD2VPN6UIPjMoJFqlrMs8pZFxLn19wBFkuzQApKSkuK8m6Vfr5URC+Mt/rlKZpTRv9zB4cv+krJlcaG2ygJt3U3H2i/T+7pimWvHJnhbvsCvHng5kdvIBy7IHp4vGgGY7kjStchT1Lu0BZGJZ5htqvOu2b1M9B2uP0yjYln8irUTCZQsi5W6PCueRwpBPtGY2ZkIJd1RRNc5F7Dq0tUkZD22xExcsnu88RzI2nECrApUzn0tjYxYEldng+Tj6oaZFrNUBPVlEop+BRHP8QL3EBeJUXIVj7CqUujCynmc3+RH3a2vj2tZrA4CuiKJIwNY9Uy1O70w1qrYCs73fO7qlZRkiiGRb3HFdIsYLx1V2TGPJZNUc5NqvZ6NX4at32P/+E7mZ+vbqfZP60nHUgx3r6m/j4IMvqzyWNNqukJErA4k/6tc4cq+F/KC4csB+IksMuFmSUSSQ47q1IuEdcv+mdrT4JJFyARdtupAurSYdcUXX7J3OKNSgM/1A8b8bkK/Wz20RVvsSS3sl1VdXMviwGxjJvSgKNLlzrBBuwpfVWpdSqPfziGFzYueeSlWUgV2h8uqM66lcfZ3+5TCCn1RxHmVOaSAvec9DwDp96qYlsaRPlUvznniTcREEhFrGhEmVX2fmR/i6UVl6FldB0lGMRIECL+3qS1l8Szwgz59HBWSVhdSWnTFerHtChMxmzcVinxhaoa+8C14sy/hnX3P55vjU1gyjWupK1/zrw+FIedW8nRHEVV8qpaeLSssJeyOz3likoQIKdgWq/TCZfSdHcv2rgOaE35H9GctCCHeLoR4RAjxyPT09Pyvj4tBUagL+6xtkQoS1GTl3MBnSiqLo3no9uzZhxh6TE0i+EbmQ4AKQNY4P2w0lPFYjFAIZhKqkdUCqVUR4/GkumGmbZuSZZFusZqVMFlhqsXvKFKT9A4MM5tR1v6z3YWpgM+pLlzA6nneZTzFDxDAejHDU6OGcBSjXtYmz+Gf/AzrfZ+VYcigTsuzae1wUmGM3/r0vSSsNLNN3oaD1Q3MrbmXa/qUReIIwdPFISZtm0AI1unpuSKuOsLZeMO6SciAuIxTCsqMlI5y/uAlrO1eh6Wb0s6xhjXiNgl7b2izKgzVKAEoWTF8PVGndipn/UOUCVmlA5ZHgzyRFbEmCFglsuQqPkHpqYTOOYSVp9T99vM5J/DokhEufn0msKzlvMe7eVu+wEgy4sGuBCsoMBmzsXWb6HaLXOD7CCn58NAKrp25DeYOqPXF8yNMhwsfF3eg1D7fvSue5vcu+cP6e+n3E1Dkuv4kT3ddhnUPsd73iQMv9kOGde79Y3rJhhc7DodkFh8oRh7PqbpcNzrO7/kN904UJcjoqeoZPwljW+huap+D82Y7btGd4PrAZyLswXf7OVoc5bMjd/K1vl7ievbx83TbvGG7yqWvyjgDFOn1Z9iYTrEyCHnuvGuQtMrIZC/9mSRWYhoZxTkv8+yWbXoyPvlAWezrciq+skW7OSJ/sEXYQxEiowQjWQebJMJu3K8Fv4QUFmnhco6n26yIiIcxEvg8c/IVfHBGBacn9SxZN1Buln5RoifRiwxTpEiRis2Ss23WDD8LfvOfSSUTRN4q3jr8Ai72fUSYpGxHTNi2mqkLDEQRVGbpjiQ+Xn22sWMJ9h/awK27XkPhoi8zYdvkLZv+2ijvLBH240JK+VUp5eVSysuHhp7YCmfPXSXrKVFztk0mskkJdaEuDBym68LeULE1zmGGtFDUAmtx0bgp3lie5orpWdb4IYe1LzQRUzdCLRBWsaz6SoUH4nEiIRj0GmV0aVdIFHa1WKkiVqI7GWNCr33ybHehiD+36bN04Y9g6k/YesFHmHybSt97X/x6/tH6QX2bvOylmFrDm6eO8qORcbqkJKP3v4JWX7Hl2RyYLiNksp5GCJAP1RyA7d3q3BXsBL3CYaP2+77I024CLezZWOPHru1jyxQlr8RIcYRze87lr5/71/z+ue/CkpJ9bmPN62Zhj/lpVgcBc7bNFwb6+aNgP6HVOhweq+7FkSGlUFm5B3WHtc4PlbA7PtJbyeDhN0HYxbNcJQbzXVFrQp+uSOJKH3QnH9XmEMS7eXWpjCVhR8pmQBQYj8V4ig6s90QSG3jBlMqBP+iPwOhmru7t4WZRZiLOAvZX2q+l3h3v4sL+C3n3XIVnVyKeFU1B6hATMZt3zeVZrdcyv6gWICxOMiSUEXDfnCrzNwMbT0i2pJJESHqtBM/wfPpTGZJ6BUTfj9OVWMW/Tc3wjMIK+PoryWy7uV6PZw28pP56RRAyHYshpGRdEDAre5H+AOOlMW4b/zn3dqWJW1ViWFzo+ySjiIOeckcdlcMMihKrRJYdiSTP8DySUtL8MLLACuuW/znDFSJvBZee/4KWDKlEokoh8ugNI9ZktbDn1SxY6Q/gN6wvh2sAACAASURBVBXoWxExUhzNVrBIYscbI4SR0igi0cMfPXuQtz5rFRfp8xnYYAvJC+zZ+oS3UT2/w5MrAVgXL9ET74EozYCbpyul7ok1570UXvpe0vEYQoCdUOc48hI80pXglevXMaVddwNhCE6OjIRIOIRStaFACO6VWQ4kYrgxnx3JBHnLpi+KkLG0yng6xSyHsI+C9l0oztGfLTvFnoblkbUsuqQkibpQlwalusXu96mbMhNFXFJ+rD7tvuYGsGKNKftdIuQPSmVWhYGaxg70xFZiEa+LYdkWhELQE0b1IFm82t8oQzdaN2r4/Vf6AjtWRgjBUZ2L+6w2Fubzmiz22Zk1FGefxfoVGdLpRppVf9Oa17moj2xybcuzL2uumBV+a/nxSB2PkI2cXVtCVajL42hByet84I2pFINhyNP09nYix3wKtgR6KHpFsm6WwdQgr7/w9fzG2tewyouRtRqOilonFwVdVGSGtbqD/VkmzRHpgtUqyAeKamLWhFRLIBzS6WVrgoCniSO8sXQNb7Ae5IHk3/NcsZ9fc6qs8wNeUdHLCWvX0+ogpEtKtQKgTvOrzwiOZ4gByciiErPZmLGYs23+l/YVd0cRf+R+iIfLf83vFsrMyhJy5GGu7O/lxp5ujsZidZGS0sImTnmRh0hnEmq08AYnxjWTI5wn1flc5wf8arXKWh2kf4peuoLiOMO6c54UOeIizQtSKlh8f1oH/WpL0Sa6WJFWbR+ZQMbSvKZcYY2bhdCjq+nRcH3JwcY18V8KKJdiUsIsfUTeIIH0mXGzZG0L26qSJEYMuNTzyabUufl6Xw8fWhewaXCcw3Gbi101qrSjJvejZSFSaiSY7prjN85/GmvXX1o/VoBQFClJn94oYnVxCiElY5VJRNSNjBJIQT2R2BOSgXQ3I1kHIRPQZLGf23MuJLsZSvgkwgqfctW5Waut8meKg/V744iOcQkxTCAt/oFrGC6OIYMMvYHHqP6+lmyQTtik4zZCj/iabZDPxX8FgIEwAiQZKZBWlVD6SH3PPZxKkdYGx65EH64FvVEEfetAtI6sTwXLIewPAxcLIS4QQiSA/w3ctgzlLqD4jNfXX0sh6I88VuoTvi4ImNIpXyKuhDsTRaz3D7BSC3vNlxoT6iImmnSlL4rq62n0xlcSE43UMkfPZmtunHuql9VfZ3TjcYKG2K/04mCrJXiPBkWGg4DVTTnRmSjCkrJF7KNID5FXdJHuavhI+5umw1eibmZjTUPrZC9pWRN2JRBdep2MWKjKk6ES6nQUMRCAtFvdCSU7jiMEG9NJXuBU6dITYeLJhUHBSAi8/7+9Mw+z6yjv9FtnvfvWi1rdrVar1ZJl7bLkXbaRbOEFGy+xgQDBEBKDMQQTSAbChIQQQiDDsGSG8ZBJMiFhhiUJMYknYZ8wSVhC2Aw4XjC2bMnat1Zvdzk1f1Sd5failtytXut9Hqm77z333KpTdX7nq+/7qspu43TtNCdHT1LyVZ1Trk1xNDfueIDGUB8nZJ7l+vo94443ezMNQVXvBOSihOgZqQRlWaPBa52/53XBp/mw9zEAXtbyBD31Ov/w7H526BmJ1VOb8APYUK2SCiAQo6AXhQrCYbB2dbgNh2HH5o9KBXpqNS49pbNKgoCfyk5cx6G1YXNKVHl0/7c4Yds859g86zh01+rIeo5cI26b9AQB7JzedzaV1UKn2+mykREEgm4dzF9Rr3PKLsPpg4wcH8KWEmkFBPU8XfkVlBsN/q/OtiqGvnM3w/KsEn0ZeDQ89R3LG+qh7Z56jnuPn+S2vevIuHGm1qHDWwA9OxI4IgsEtdiCPG7bWNYoaS0N3fU6Jx21wflDrUOccODLLacJhGClTjywZex+fNj3eCAlqQU19p3ex/q21ZSLOdpq6t6SjRSHh59DooTOBdq14eLKEuhzVbX4jVqhsA+B9EAHT9+T38y7r3i3as/RU1AbYrVT4PRjv8muugrErg2eJKMzs0Jhz7hlHB20f8e+n2EfuIEV1ThFtzOv3CQ53ybj2dFqrdKP3ceHssew8SjqgUVG2gh7CEmAbKj2+XY6xc7hETJBwA/TyuArNAJEYZyX+rwwbWGXUtaBNwJfAB4BPiOl/PGZP/X8GLsHY3twmhed+jaucOis1yOL3Q4X+9EmdyGQ+EHAAW2xO9pV4yf8kMXEBJay14ZrxVbuiA7yhcLkCJdGrTV6PxMEVKXNcCMOauarGRB1Hjv+GN8+/RQra3XyQRD5wdePVllRr+u1XsIlJ5UQr2zJ4HrxzVhu2qXG4aCVWFwpU4lcQS26DoWaKrujBb3e0CORICDfGN/kA5bLL3W0c9i2edHgEBndmUNXzFieHCgjkTRkI9rRKe3a2NXxQ8zG4WsY2fdKTpBtejCOpWs4rm/aUuc5QBUhtaUzhk11Zd1LJ80yXe/64Br+8GcZNo9W8QKBFKM0dBeXoVXtFQhsHytwGbIFP/E8rh8c4kl3C14g8QOLIxQ4PlSlGKib9PNDyod+2HF4wvPortexa3lagjpCqjrtGRlft9BiT2VVv9ir+9+lwyOQaeGK4dOkRiqsqtXZ6/VDbYinH/9J1I6jI1mO2svYNFrlZ56LIxy2Odp4cDO0ZXQfDFwcz2fEKbBK6hnSB37IG06cZHRkJRlHX1vpUh1UwnX96UHqToYRfIJabNGftiwa1igZ7cKqNBoM2JKvZLP0OAU+8dwBhH5AddVUf7WJjaBPF/J8TJzkG/u/QV3W6cn3UMl45GqqT62pjUQzX8N27dSjFI8yMgiFXZ1vxIJyKsdILSCdEOBLWjaS83Lg59TotTqI8DK45Enph19X/ZnI6NrrOCBh35F4dLG6OsJN9ae5PbHNY1teCe9rd67i/T+3GcJzDcZ92/KOUnZ6sfV1zeBEvn/RUIZDTQg2j4zSWa/zhBfqUADFpHPj/DEjPnYp5f+RUq6VUq6WUr53Js45EeEaLSGZQPKKUwN8IH8zvoRDeqZnGDy1hGoU4eVpbQQ8p33snvaxu4kZm2Enkw2Pgp/Ds+JUqbr2+XVpi7vF70AGcWdOS8lx8tQbsfskG6ib5Q1feQODjVF+5fgJLKDkZEgFAb9x9DjvPXwUG8hrK7JP58/2VDJg2dR0h08n/JMZx2efSCzWn7DYV9dqCAmVUVUOoTMkqjoeUAgCckGzHx7goN3ghymfNx0/ya6h4cjCa1jxOiS2lLz01ADvP3iU5csviF4PLfaM51Ad7Rx37jrqu7/Y2MG3Ri/BmmijWaB1OOHaskrYUlIXklwgok7akILHgy6+H/RxwYjaEUikS2wZGeWq03Uap9dS1Rs0OIGNsEYZ1oFgqS12y05x+Of/geONVo7ZAVIIyo2An+W2c8vgIL1DWUAQSMillUvvC9n4If+k59JXq1E+tp7XnDzBEZ0hcVujeSYjQN6PrWsAp67KdvHwCGTbuHhklF17LyYtJU+6/erQ5/4t8gs7ssDXD3ps0nGYyzsvp6QfpLhpWrUrRgY+rm0x4lVYKbQLRm8wckBWovZUo1BB56O/xPsOH2XIVX20LbWM5F6SQ+4oGWBUOlQaAUOWYJ/r0pNupave4IrhEbKBTV7nhNs0T88H+Jsn/gaA3mIvlZyHV1N9fF099pGHueSdenSctipNFrtEBfYr6SwlBlhffyr6bPqiV6lfvJxaoqJ6Grwcac/G0Q9Ulxqu3jj6lG3jyAy/eGV/dA6ZLvML9pfYnnCH2p6yrvvb81x74TLQs32vPtrKN556Bruu6trq9UXLkGSEi7DVOazhTbjaoNxYrdJdq3NQLxgWuWJmgQU18/R0rXmd54wMyEjJBce1RSWVZeTo4be0ygRS8A17u061C4OnqvO4MpkWqQOg9SKNYPy2WhC7YopuB2jLImWpJbiOyQK1huoUjpT0lXoAtQTrrcsujbIHym6OchDQX6uxRb8WiuM1azrpa8uS0Vt71bQoJj1yWS/FgXohXiSqOhhZ7FcNjfDR2g4KI+o6hJMxRnQwNBdIcsF4N8hxEbsEALLh0D0RxCoGAf/x6HFuyK3iVVdsil4Py552bU6M9o47N/oafy64it+t3R2JVkhOW1SZ4Xg505STJa1vjtZEXvunGrvZU/0DvtjYgR1umZIqkpeStx71kI08UouohwNWLVoRMAgzlyybTPcmGkGGo3rqfCEIOFrcyFsPD9N2qif6vvZ2NcHskONgJXKiV9Vq7BqxuPP0adbpNrwo3xe9b2vDoahdMeGib5cc6uPP9x+gNQggpx7OvVqI/91Wm7B0n/ouFX2JLmzv5qvP+WzVwnNT301RYBIvmxB2D88W1FIt0V6dIc/JFnI6AJiyVbseCDpwgVOWartXXtpPUI/jQ8fdGjkpOSpKlLWr6WnHolVvy/ieI0f5tcYWRkRO1zcebYV87ZmvAWrf0bzvsGKwzFVDw2xOiGhBj0Q7dcwr67QgE8I+IgRSCJZlfLZaT5CVcd9Jp3V/8fPaYh8CL8t7btvIFRfEBoZb6o1iIo4s8K5b1sMb/w3e9F3E6t1stJ5CAJ/cf4CPHTgE3pg0UW2VL+cEOSmxq+qaLfP7IjdNRiRjWHn+w9HNvO/QES4cEZExCKrfsFBcMbPJWIv98ZyK9vsHv0tdWhxBdXpb++dG/G72VD/Ag6f6aWs0ovRFR+ebOglhL+pOJmtFTo/WSbvjO2tXKHzWMqTe4SUVrslBgZoehhWCgBdt2xZ9rj+XEItUS5SlE1Lyy6SdNG+/cT0P3ndl9HpNjBfhrOdzarQB+sHB6ADXDA/z5mMn6KvVaMl1IwMtQo20Wmt9WN0s+SAgO4GwB1oMwgdB3otHHuFDsCCBci9WzyVNK9SFrpiUZ3FgdPW4cwssHCtc3MkZ547pCLSrpBG7trJOhpTW7XIiR/8xqayvbwfr4hNooauHAXEt7D4OQkhu+7tr+bNiMRJ223bIeg4pO8OIXiuoEARk8yVeVX07H6zfFZ26/8Ld0e/Vw9dGv/dV61zhq+yfP33uIP/49LPYnVtUcA/IaXdJIaVFQm9f91hjTZweqNdW6dEL1/076sHQGRygiGqHjR1d/ECupn2oxKbBX+CmVTeBr5d9ddO0ZXR2mXQ5PdqgkY6vYcgBWSGvxSp0DZ0ky6D0OYY618sv7SE4vZn6abX/7qgFKwcO84i1loq+LxpCRJOt2hoBq7O9DOmsHMfOMpZ6UGfPyj1UUhWEEAyJC/nYwcNsHLVolzZtgWSlzkTq9FQfyjstTRb7gB6xLs+k2G4/SUr3BYHADw2byGIfBC/Di7d00tUaj/7sSm+UNeYI/XprP7SshlXXRMdtHq1y1fBI1H8itMXen9bLctfUNehI9ccWe8Jt6wiXgl/h5sEhPm9dx8HcLnWa0TIr6g31vbPAghL2sT72mq+efu2jT/OMbKOO6hS2ttg9K81PZRdfaOyg7q2JPheKlZUI+oTDQlkvMjhabwo4hbQ0Grx4QNDpXgFa2D0rTUPYnBBFqg11oxQaARWdmQPQn9gt/dfXvYbfOtLsuy74BdJOGs+xyKdi4a3pIepn6nEHzPkpBkbqsbDf8D4ellvZau3EAvKtXXjVMrkgwK5W2NBViHz3+SAgpzNeRH28P7yuH4y+n8Gq653eM8qfX8KG1/w97PkdlSamSbpihhh/g6ccj0Ja1akhbLpqDdJBgF9X175Dt8G9114WfSbrZvG1xV5JxBdCYf+OjF1BodVU1wFzJ6V++tqKGqyf5j9VijyiV7Z0bBfLEtyxNb7BCo2AfK7A9+QaDlLhd2/byOffeCWda3dExwyfiMu3qlZjm3iMBhY5KakEAXRsxtIuiYxdQkqLvJ77EAr7w0Fs1aMzvLrFIYakz75GmYa2Dlv1mt1d+XaelW3srn4I279GuRhDi93N0pqKXTH7TwxDtlnYa06WATKRoJdSYXBb8LeNy/mG2EY+5dCa83ndxrfgnrw9+mxXzy4+nH8bxUb8YG1NTLby8q2MWKq9PWd8u6edNG/Z/pbo73/K3cgVIx/leO56vvLsAb66/6gauQArMmr0UvbaoxFeFcEzlqprUVjszOxlVM+qTTtpRJhZ4uehOqCEXc+2JbEKptOyKjLo2uw44QGAVcowrCUMvDDYHp9AXbuLyjrPfrSPoJ6nI7MymiNRTjzYbMvDS6vr/HS9wkBRrRd1+uQO/uGyP1cLrc0CC0rYB6oDuJYbrVx3YjQb7Y34UxkPv2wdHE3r445T4Jl0fEFdSwmdmEDYg1qRYtqNhq9JfCl57WARr74qCvJYpHg2vZ5HrNWMaldMIQgot/RHvv7Vxd7oHO3pFfTUmy97V66LtvT43P6GFvaqE5cl76cYGK1BWT8sOjZzr3wHJ9JK6Ht7enn/i3bxjaefhVqZK1e3IgPV0fNBQFav5ufV+3CEQ3sm9tdXpRpqCzfNO3e+AYCTeqGzouVBoRP8PHlvAovdUXV62+FBLh9OLATmuuRT6lqVcll+8eQp3nf4KGntq7xVVrj/2HHWrYxFL+tlSWlLvlSrIlddw1eDbfwgCMVY8Hcb9H6gHcotFOgyZfNKDFrSze33qJ50Y4VbpyXqULB8KvlYDH7+kh42d5cQjs9HRvr5Q/sFKv21nkc00pSDgNbqPqzWNRAGMDs2YemHZtoqQuDi62sSCsDTMhEb0cLeIw4xILKMNAIG0urB1aGzXVYU4yB5Ss+jIBVb7K1R8NTDEiByzRslD6XU5wteuMxAmvt2reaeq/t4e/0e/rR6LUX90L3/urX88pWxi23l5peTSqfJJvpqayEO/KWLbZxw2jggy6S0oGb0QPSuyha+fNeXVTqippJPsZ9WjlU2q71qa0PRJusXF9fwwWs+yKrslsgVUxOCnziqjTrdHOsaj3EyUH2taVatn1O7M1UHIKODwE4i3lHsZuvIKO01yRu3v7rp+lDu5YBo5zlZYVLCdMfTB0DYBINXMfj4O0g5fmSxtztx0oQj3CgLan+9wIbiNfx839uoHn0B1opLZiXVERagsOe9PGn9FN3Z38UpV3XuI35PtFN89DC3Y6s76UO0tcVOIgAa+thv37ye/3DjOkqpiYV92MpwbLCKrzvPvmMB1xx7B592bqWBTzoIVPZLbjklv0RntpNcKu44g4HNkGwOtN1/0f08sOcBxtLQDyDpxmmEhbEWu5tipB4wrG9icm3kS+r73vuSy3jFZT2RsBeCgJz2h/qymwdve5Db+m+Lzn3I0oElx+e2/ttYU17DW7bciyVlnD8NKhsBmh6yjm3h2RZfOPY61tXimYae40TCvn+gRn+txrVDw3g6Y2ez28ZrTw5AusLbL1Hrxzmk8bSwl+ujiK7tvFm8gyHiMhxavgt+6wS0qDKLlGrfAR3EzI4ZUtd02putd70K6wBQcHwqGT2q8R3sxPILu1/3OV7wyj+kp5IhqLbg1mMDQrRdAOVeZdW19GMLVb7e9KXUTm2JxfhVDzLywg8wnCg/OdVeFXGaISvPSK3BUVdZxFta1lJJVdjUdkHkxsp4+lyhK8bLsKqwilv6buH+nTfyW7dswCkoYQ/7zaDfrusUW7q/dv06btms6nDg1AilTDxCXJ4rI3UQta/YS2vOp60Ru85a8ito6Pdz5XYeKr6UO6u/HcVkivoh0OKXKXjNOwWV9fWtt1wYv6j9zVa2lRf2vpCs70Wxq1EheFz3rZUnD5Kun+RI0BrVI8LLQeh7z2rjKLmhvF/gD/aP8J69WfasH7PfqBD8Zern+FRjN5MSjtyHjkC6hO86gOrr4fcs8+L727U8snk1ij1EiVI6zWs2v4RNXRU2d5fGnv28saCE/W073sZnbv4MKX1Bt3V30NbZC8DLbtyNpy2kus6HrtZii1yOxMPIhg68JTNb+qs1djktvOnyF5HxnObOo0lJyWkynBiqsaql1HQOz7b48buvp9IIlPvAdliRX8GG1g1gx9/z6OHRWNjX3wY9V5DzclEgLEkj9N35eaQWrHxKC/uG2+Gy+5CFLqr1gKfadsGe98Dybcp3eNt/Y+Xmq1leTPPS7crSzQcBWR3wSllFego9kVtFSsE/pV+kvs9J4dkef/3iv+aWC17CptEq6xOdN7R2S34pHhKjJnX8S7CRR5bF8w0c4ZL3XVKuxZX9rVT1sNerp5DSxg23Csu08IoLX8HDdz/M0dNBJOyleh38HCmveYic8Wz1BNdWU2uLun6tFWXVZZ1mYa/rNrd1fCWXeFgW7TQVvfpmMTM+BgHKih/Z/xKO7b0zms1K+4Ww9gbYeAdYNo4W9lX+LkYP3BELe9sFiEt+qfmE+bg/Dtt5RmoBzwkl9huWbeUfX/qPLM8tp6zLFQl75IrJ4Nouv3fV73HfzsspZz38gvr8iUwvAAOeEvaCXlM8vG+yfiK2lI7r25JNIxsZpBT0FHrY1F2kQjVaZbQl28ZxvXpkobKMhpvjWdkWjW6zDT0yS8eB8Pjcqh5Oe5yVEk2tz6jjc76d8LHDXtclHQS0/UTNvF7Rp4LZTfdmckeicMSSuN9IFWjpWsuOrdub+mrIV/O38LHGGZa2Sj4kUsXIePQcK+p7RT92azqWS+fa7TwrW3k86CLnOywvpvnbN+2kozg+IeN84Ux9yPwh42bIuJmoYbNuNr5BWtbg6k10R4Jw67lELvpQB6HBFOgF9ethilq2nczgIT5avgT0BIWUM74RvEAygLLYlxVzPEss7K4tyPoOv3/4CC3aL/zhXR/GtVxI5N9+c+8gndJXqS53/sl4n16CQHdQO5Vn6OnX84prj5IdzDMwcgRZWom44fcY1QthWak8XPkr+pMWbH15dJ4bN/Ty+ee0j92qMHLgVloryn8cXkvZSFN3i8oCyibcQpbFX9TL0B1bNb7t41kepVSzBZJ2bU4O12jJ5UBX2bFsCimHYtrlgVdux/mgD7UhWgeW82TQg1hZUUPydHyuwwMjtOi0zHKjAX4hEraUazFSC8j4uuvqa9TR1s6/vvM62h49CA8rdw7xZaeul1gOA+fh6odeIEl5eVpyWtjTEwv7PVf3cXSwSnveh3/Oqw0p2i6AjT8XHeOKFDRcHj88iGuLJsvfG7MxA7l2pN5wveoWGBlu8LOgjZ3QtJZIJeNxeGCUtJ5gE7tixo8o02VlkR5MraI09DOeCpTQF3XaZdjWOT++7UvpWASLGRfZyELg4dkeW7tLCFQbHHIcWtOtHJR5WsUp/EIbrt7Q/LJlu/m/j5zCSf+LLvN4t2JFp4y2FHJqNnBQh8oq2Psv0egl4zmRK+a/lks8KxusrNUR+78HXTvYuWEdX/nGBBZ7SGSxJ9Iv/TzuKz/TLPbJa6b71Z3Wh/jLV0+wKG2TsJfw9aQs17Yi/7uVeLi4lkf72ovpHf0oQFPMbDZZUBZ7SJiKmHbSyu8L0LoGR988I+EGGBtW8srLlMtiYKTO6iH1ueGGegBUtTsgupESAdOJhD0lJQMyzYmhKpWMEob2nLrRXP3dW/XEI4DWdCtFv9jU0f75qVNUhYf0C2cUdQCpg0B2Ok8w2snrN7+ZQtql1pCM1gOCQHLktHpIRf7cCQhFLN8IsNw0teOXk9UWa+SvDNJYrg/3/gtsv7v5BK//Z7jizU0v5bwcZb85AJv2bFxbUMnE1861XO64qJvX7lxFPuViaVdIcbCb0QO3cXrzq+Eln2i6Fndc1I0TCnsQqPxkbf2GQpQNLVgnHtW05f0ogFbUfs6MpazBmhZ2W6/zEVrshSAAN2GxTyLsQgh+46YL+aWr+mKruW1d80EyRRC4PPTD56g15LjPe8k2cjNUdSZPwy8yWg/4WnU9T6Q2QMfG6LCwXOlww4dUPEFpLHZeWaz7GiVuHf4t3rb3UoSAtONhCzvq05mEsF/QkYiXZDyCkQ4awyp+s7FbxysaAa7lkHfzVNq0IZUu49rqwXXRsq1Uj+zB1m6UcjYejcT1UNe1JevHKX/rboaXfhJWqABj1reRtTI+JX6Q8jmaPsXKikoD5cJbIkHPOGN87CGhsNvNwk6m0nxcghVlda7vDC2DnsvGH2DFLhfSpagNXTseLVp+IotMGw6rWtVryYfobLIwhT3soG4GNt0FV70Vsm306Ys5rIW9NZvnTbtVNszJ4Rruwbv45RMnWdmiplUfbrTx5fQNsPZ6deKEsKft8a4YSPGcrHBsqEpLNoUjnKiThcJOpQ+2vqL5Y4mO9uNDVYSXQfjNPsiJaCnqm04fm3HtaEj7/x4/wpbf+SI736/yhaNh/wSsLa/l2vJGto1WETp9K9zJJSy/bGRUHcormy0eANtRHTzBsswyOrLNPsu0a9OW8/ESK1y6ts31Gzq452od+NSWUzXMYCqvgAtvbjrP3Vf0sm25cv2UGwH4+ciyCn3C4d/RtQ0tWR0obNFB3czoToBoYalI2L2ksGcoaUEvTeKKacIvqlFGS3/Ty/2V5aTtEus68mzoHN++vmMxHLrhHB8nq8oY/vx/J1r4i/V/FD84iIU9nNtA6xpYuRO6xmR4AOTaaWDxk9NZfiT7GJAZPNvCsiy6cl105ZSgZhJ95eLe2MVWSruM7P95Rva/RF0bbW2WGw1aU60IIWhp61D90XajPh9eO6FTacv58RPVrlnbzh3buljdno1nX9qeansRxhEcZCPHzeUHqJ9WmU/LV14Dr34ILr8vEvapLfaEdT7Fffare9ae8X11Pi3sqWIk7MoVo2do+zmCuh4BavfpnduVsRjGl2abBeWKCYl8hU4WWnqhcysA//0XtvPNJ4/x0MF/AJRohZbsULXBk9ltfPbNe/nevr18+rGHGTm+k/+xciXXFdXKcmey2G1h84FVf8rX91mM1OoU0y6Xd17O8PF1/DvghtbYr3xvfIG1UDakoIGNcDNEuwifgXxOdVqhrY20Z9NZUmX85LeeVr52zVQW+4c3vg6++3+wQmF3wsyh2BUzzl1wBj6y6yPjrlHGs3Ftn2RRHGtMF9MWIkwJ6QAAFllJREFUTVVPvpqs3Dl9g5SCBvi5yBUTCl0oOtFNHA6Htb+3I93Gi4oP8OC/ncLpfzBaQ93RD5Ymi93L4tgW7Xmfttz4WZTjSBXVA3zMA/Aj17+DodoQy7LLaEywRr7vWLxr+NX8vv+n2H4eO12Ck3tp6A2Oaw3J8jF+2LK2dNNuwsf+mocmLdf7Oj7E/3oqtsJDIfrsLZ/F03W3Ei6irStiF5hKSxUwZvnnW08PcvAq7drr2q42SiE2ZsLAaJgCXCmMnzbf05LhP790q/7Sl8PT/wSFZss+q61bz7GoHr8UJ/coXflu6FUP50jY3Ql87G4mtsqT/dKb2FIPaS+k+Kt7r2CoOvlyF7hpGDkBqVLsY7djH7uTzqsNPJxBbL3n6htesJpr1raxsWt2Nq8ey4IU9mhINmY4Wsp43LCxg8/uG4reT1qyoYgsz7UzvPceQHfOsPET5xsrWp7tMeJ3sn9Q7SZTSLt87PKP8aEvPcbXeJwJNqqJsWyksKPA4Q+6XsaFm88iQq5vxEsuWMnbWpaRcmNh/9efHWs61D+DxQ5End0O1y/RQhleQ9nIxA+ns2B5bvxw+77dyoJ94kS82cY4YdeumNFI2Ccu9650F0f3/yvL6w1lsbtKKPesX8ad27tji7jQraznsp43EOYyexkuXbmST33zB5Rk6IoR41wxRW2xA3zitZfQejbCftWvQmP8io55L04FtSfoEJ5t8dnGC9j54vu51XajuELgxzf/8lLzSDHM1sl4U7SvZqD1IoaeiteHD6/vZBtvpBPnnajMlHu56fhTsPE16u+rflX9Ixb2fEotcftY7QIcjlDKtI8/T5Jtr1Cj5DF595u7i7x1z1p2rWvn419fz6tWfpA718axnQkt9lDYk3Gh0J/upJqt90nYvnKKZXQncsU4VmQI2n6O4X2vxG/9KoWSstSFEHMm6rBAhT30sWeciTtruMlw2mm2QkORT24o61giIewJi11/h2u51IIaKTtF2rWp1tUTORxihRbkcHXi3ZDiL0pRrSpxGlx5LWxadebj9WcAejs7eONmJZqdRVXGwWqDdR15/v2Amo17JotdHaCE0NJpgRNZ7P45WOwTsesCdUM//Z0notc8e4xrQ/9dlbF1NhHdXpH7j+vFmbw8aU/FRbK+wx0XJTYqaO2Htz89zmLHzbKlU6/PD+jFBWOLPXTFNILoM+s6pnaPAdB/7dTHTED48I3cdtpf3tIeu7Q6x1nsoY/97IS9vdD8YJqsX7x1z1r62s5szQLwhm+qQOcEeLYg7dpYlsCzLY4e383L19+CO7bNJyI7PgvMtS3edK1ynX73N/dE91bIhD728N5NCnsi7jIjhLqQdMUkLfZMDlltY2T/S/Hbpn6QzAYL2sc+UUoiwHuvfC93rb2LDS0bVKdL+sVo9kc7thUP4ZI+djeReQP4jt/0ubzf7JMN1ySZFMeL3A8dhbNMe4o6aHwDpj2bsv7OtcvijjulsLdfCC/5BCe69HrcoY89stjTk4rsueIlrHRvEldMjTMLe1MWQ7oc+YUnrGfyBm67EK68H/qvY1Vrliv7W7BkvEFLlMfeFDwdP3PyfBAaGbGwK4tuVVcXv3f7JnpbMqxpbxajKHg61YhM05ZvFvbJru+brl3DizaPH3WNw01PKpC3buvirS9UPmrfsSBI0Z3tn/DYc2WsqEMs6M0Wu743kpOz7PMl7KXmNtQ65KUTWTH27ExAmooFabFXUhUqqUq0dMBYeou9vOvyd0V/+45FtR5Ewuw7FkKorSxdW0BlNZRWQnuc7hQGT7NulhOjJ0jZqWZh1xZ76F8cnELYhe0zqj29HcWzGO5DLOxj/ISdpTTHh2qsqMQdfDKXRlwAAetvJfVjtTNPWJcoj72RjQVnmiSDp449sSsmfMhNOPyHWNgLXZBrI+2pNVWmrKftwJ53A8pT/OvXr+OXvwSj+nsc/f2u7bJ7xW4uGfq6CkjOAr7bbGBEKZ7pEi9f28PLL+0Z95kWnSZ4ttkV7fmzs9gn4+/etBPrLGdHXtRT5qIe5cbwHBuoT90+0yA0tpqDp6ErJjECsCyVUnkWCQpnRfh9CVeM58RZMW4qEdOYoXtouixIYX/1hldzc9/NUx+o8R2bAeqRlSqEwHdUPrRjWZBrg/t/2PSZsaMCz/aahsNhfmos7FO5YnxG9ay9juLEI41xuGnlP/aaLcrOUpof7z9FTyUT5XUn3UtnIsyuiNIHUyV+fdvv85ufGpk5iz0xFPcnEXbb9WHijYeajguzP8Jrf65C1VPJYCWCgXbiofOR3R+BM0w6nGlia0+XJ0xdTE0eb7msr8J7btvIxavOMO09wdla7JPxfP3CfjQaPn/CVvJLXNdzHRd3XBy/aDtqw/LOMVlCTmoGLfYJsmJsW7l/bA+hBb9aD2bMOJouC1LYc16uaUr4VISdLWlNpFxbCfskQ6co80YP00Mfe0i0/ol2i1SbNsOYAMenqmc/jrWqJuWiu9VaKGMsqC4dYFtRydCa83n2+PBZ38Bpb3yc4Yrl10DwjzMm7H5C2Mf5W7Ul/sevvZLHq2cQq1MqSE2Xmm14RlfMGXBs0bQTrGuf5bU/DzT5Z0EvR5BqdiOMwbEtfuGylZO+P5b2/Nig/+wITdgu59NidyyHD+360Pg37v3n8a/Z3gxa7KGwl6J6uo6AzS+FFZeBn8e3jbDPOhNZEyp4WMO1Jm6IsQFa3/GbhD3K8Z3AFzghtk9Vu2LOuvErq9S/MYTC3lPJ0J5Xwj5Sm2LEoAmzK5JZNON8v9PEs5PnHutj16s6Vop05M+wqblOqWOFmjQSPpDO9eHjWFbTZsvOWWRJnC8iUQiv84Y7VCpfeubWEGnL++RTDoWUy74Tw5H753zjzYLFfk64mab5ANM7V8IVY6uMOzcMnrbpGINrMTCqBX8esCSEPfatJy121QEns9ibli1AuWKS65XktMWePctsBRyf0TP6Hs6eO7d305r36C5n+MjLtvGRrzzOpq6zE4eucpr+9hzrl8fWTGxtzZSwx+I53hXjNf+cjD2/o0Svp1nYz9UidGzRlCHgOnPX5b2xwm5Z0SqPM0XKtfn6r+3ib76/j3f/7U8WlcV+Trz4o01LM0yLZPDUUQbHWAMjrLfxsc8iE1rsY1PPxlD0i7iWG82uTLpisp4dBf2EEPS357h92xQ7o3TvYGWpxmcvvnxadQE1Srh9m+q0KyoZ/tNdW876s4WUy5d/9Zqm18a5CKaJnxDPydIdx81uHUumAlteFv2Z1bGBc7VAHUtgSwHhWjFTfe95JLr5z7NVV856kYtwptxrU+HNsHEwbZ5nSuqEhMFTv4DnqCD+2Htl3EN7jlkSwj6RxR66IpxJsjKKfpGHbn+Iw8OH+YtH/gLf9qMHQ2HMeiJjhXJCbnw/y4BlUx44+5QyHu+86UJu3DQz1uMZhT1Mf5zKYh/D7gvb+Y2b1tF/NrnXCYQQTfMoLWtuFmWC2b35w3Tc2bKgJ0olXjQs3wI9V4Dt0J5PUUg549pwnJttjlkSwj6RxZ6OXDGTN8Ty3PJo16aUE1vsc7X+w/nkl6/um7Fz+UlXzFjXh602VGZsfvsUFFJuvN7MOWJpi11IiTibyTPnibAfzoYVHfbR2bLYwwfIvLHYZ5Jtr1D/UMs337ipY1yabtPiYPOARdgK4wk7XdKaiF0xZ26IMKvDs7w493uOluJcKCTz2FNjg5W2q9wws7STDBClO9oQr6U+B8yqxZ6aZVeMvYgt9gSeY43LPILZfWifDfOjFOeZON1xbFaMypo4E+EynCknFQXwFqPFPpOkHDfaiWe8xe42L6s6C1g6K0bAOY8UZpLZFXZtsc9W8HSCe2wpERqP88UVMz9KcZ6Z2GI/c1ZMSCjsvu0nXDHGYj8Tapga3uhjLPZsm5oQNotEFruUU66Bfz6ZzcyJMB131tIddZ1m6/vmGyZ4OgdMaLGfpSsmXJ1wsfvYZxLXFiAtEI3x6Y5X/xpceu+slicMn1oQbaA8F/S2ZFhW8GdluJ5LOXiOFS9vfJ6JgqfzJd1xlvHnmY99SShUmAHjT+Bjn8oVk3EyuJZLyS8ZV8xZYlta2IG0O0ZYvOy4JRLON5ZMCPscumJu39bF7du6Jtx7c6axLcFnX3c5va2zc62j4OkSt9hNHvss4k+QihV2wKmesBk3w2du/gwrCivwLIsXXNDGZavGb9ZriHFtC/Ta8yln7t1WoSvGmmNXzGwIepItK2ZuRutUxHnsS91iN8I+a6QmWGckdKucKd0xpL8cL0X6P19zyQyXbvHhWAKp5dQfa7HPAaIpK2Z+DJUXG5f2Vdh7bHDy1ToXOU0bcMwDloSwT2Sxp6aYoGR4/oSuGCkFqTmcwh8SWuympc8fuy5ojzZaWYrEWTHzo5fNj8fLecafwGJPzbOh02JCCKFcMdKaF9fX0t3cmnqbWYPheeHPMx/7/CjFeeaMFvs8ecIuNtTSW/Z5XxflbLCSWTEGw3lgvqU7zo9SnGdCf3r6eWTFGJ4vymJPLuE7V0QW+xyXw7B4MROU5oDd69p594s3sHZZvIBU6iyzYgzPD4GNlPa8WJ86DJ4K44oxnCeatsybB0xL2IUQdwkhfiyECIQQO2aqUDNN1ne4+4repnQz/xyyYgznjnLFWPPC52gsdsP5Zr6lO063FD8C7gC+PgNlmVXacmq9knJm7tPxFiNCu2LmQ/qXJUJhnx/WlGHx0d+eozXnUznbHdXOM9PKRZNSPgKzP/FiJtjYVeRLb7maNctmaMNbQxNK2O15ZbEbV4zhfHFZXwvf+Y/XzXUxIub+rptDjKifPyKLfR4J+9yXxGCYHaa02IUQXwYm2lrnnVLKB8/2i4QQ9wD3APT09Jx1AQ0LEwtb/T8PJoCFrhhhXDGGJcKUwi6lnJHxhZTy48DHAXbs2GEGxYscISyEnPtURwBLr+hoLHbDUmHu53sbFiWubGO0Pk+EHWOxG5YW0013vF0I8SxwOfCQEOILM1Msw0JnWfXl+Md+ca6LASSyYsw40bBEmG5WzOeAz81QWQyLCMcS8yJwCqG/31jshqXD/LjzDIsOxxbzZmNf2+SxG5YY8+POMyw6XHt+pDoCCBFa7AbD0mB+3HmGRYc9j1wxtt4Oz1jshqWCyYoxnBfuvryXUyO1uS4GoF0x0vjYDUsHI+yG88KudfNnNx1LOCCNxW5YOsyPsbLBcB6xhMmKMSwtjLAbFj22UANTI+yGpYIRdsOix7LCJQWMsBuWBkbYDYsexzIWu2FpYYTdsOixLeNjNywtjLAbFj2hj90y3d2wRDA93bDosS21/eFC3OnLYHg+GGE3LHrsyMduurthaWB6umHR49hmSQHD0sIIu2HR4xiL3bDEMD3dsOhxbONjNywtjLAbFj3x6o6muxuWBqanGxY9dmixm+5uWCKYnm5Y9Hi2B8R7nxoMix3T0w2LnjArxljshqWC6emGRY+jLXZhLHbDEsH0dMOix3WUj924YgxLBdPTDYseN7TYTXc3LBFMTzcselzHBE8NSwvT0w2Lnshi11vkGQyLHSPshkVP7GM3wm5YGhhhNyx6XDe02M2SAoalgRF2w6Inl6sA4GRa5rgkBsPsYITdsOjJpnMAdPdvneOSGAyzgxF2w6In3PM0/GkwLHaMsBsWPbYOmtomeGpYIhhhNyx6QkvdBE8NSwUj7IZFj7HYDUsNI+yGRU/JL3Hf1vvY1bNrrotiMMwKznQ+LIT4A+AWoAr8FHiNlPLETBTMYJgphBC8fsvr57oYBsOsMV2L/UvARinlZuAx4B3TL5LBYDAYpsO0hF1K+UUpZV3/+U2ge/pFMhgMBsN0mEkf+y8Cfz/Zm0KIe4QQ3xFCfOfw4cMz+LUGg8FgSDKlj10I8WWgY4K33imlfFAf806gDnxysvNIKT8OfBxgx44d8nmV1mAwGAxTMqWwSymvO9P7QohXAzcD10opjWAbDAbDHDPdrJgbgF8HrpFSDs1MkQwGg8EwHabrY/8vQB74khDi+0KIB2agTAaDwWCYBtOy2KWU/TNVEIPBYDDMDGIu3OJCiMPA08/z463AkRkszlxi6jI/MXWZn5i6wEopZdtUB82JsE8HIcR3pJQ75rocM4Gpy/zE1GV+Yupy9pi1YgwGg2GRYYTdYDAYFhkLUdg/PtcFmEFMXeYnpi7zE1OXs2TB+dgNBoPBcGYWosVuMBgMhjNghN1gMBgWGQtK2IUQNwghHhVCPCGEePtcl+dcEUI8JYR4WM/S/Y5+rSKE+JIQ4nH9szzX5ZwIIcSfCCEOCSF+lHhtwrILxUd1O/1QCHHR3JW8mUnq8dtCiH26Xb4vhLgp8d47dD0eFUJcPzelnhghxAohxNeEED8RQvxYCPFm/fpCbJfJ6rLg2kYIkRJCfFsI8QNdl3fr11cJIb6ly/xpIYSnX/f130/o93unXQgp5YL4B9ioXZr6AA/4AbB+rst1jnV4Cmgd89oHgLfr398OvH+uyzlJ2a8GLgJ+NFXZgZtQSzgL4DLgW3Nd/inq8dvA2yY4dr3uZz6wSvc/e67rkCjfcuAi/XsetdnN+gXaLpPVZcG1jb6+Of27C3xLX+/PAC/Trz8A3Kt/fwPwgP79ZcCnp1uGhWSxXwI8IaV8UkpZBT4F3DrHZZoJbgX+TP/+Z8Btc1iWSZFSfh04Nublycp+K/AJqfgmUBJCLJ+dkp6ZSeoxGbcCn5JSjkopfwY8geqH8wIp5XNSyu/q3weAR4AuFma7TFaXyZi3baOv72n9p6v/SWA38Jf69bHtErbXXwLXCiHEdMqwkIS9C3gm8feznLnh5yMS+KIQ4t+EEPfo15ZJKZ/Tvx8Als1N0Z4Xk5V9IbbVG7V74k8S7rAFUw89fN+Gsg4XdLuMqQsswLYRQthCiO8Dh1BbiP4UOCHjHeeS5Y3qot8/CbRM5/sXkrAvBnZKKS8CbgTuE0JcnXxTqrHYgsw/XchlB/4bsBrYCjwHfHBui3NuCCFywF8B90spTyXfW2jtMkFdFmTbSCkbUsqtqO1CLwHWzeb3LyRh3wesSPzdrV9bMEgp9+mfh4DPoRr8YDgc1j8PzV0Jz5nJyr6g2kpKeVDfiAHwR8RD+nlfDyGEixLCT0op/1q/vCDbZaK6LOS2AZBSngC+BlyOcn2FK+omyxvVRb9fBI5O53sXkrD/K7BGR5Y9VJDh83NcprNGCJEVQuTD34EXAj9C1eFufdjdwINzU8LnxWRl/zzwKp2FcRlwMuEamHeM8TPfjmoXUPV4mc5aWAWsAb492+WbDO2H/WPgESnlf068teDaZbK6LMS2EUK0CSFK+vc0sAcVM/gacKc+bGy7hO11J/BVPdJ6/sx1BPkco803oaLlP0XtuTrnZTqHsvehovg/AH4clh/lS/sK8DjwZaAy12WdpPz/GzUUrqH8g6+drOyorID/qtvpYWDHXJd/inr8uS7nD/VNtjxx/Dt1PR4Fbpzr8o+py06Um+WHwPf1v5sWaLtMVpcF1zbAZuB7usw/At6lX+9DPXyeAD4L+Pr1lP77Cf1+33TLYJYUMBgMhkXGQnLFGAwGg+EsMMJuMBgMiwwj7AaDwbDIMMJuMBgMiwwj7AaDwbDIMMJuMBgMiwwj7AaDwbDI+P8lFRswitw1jAAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "print(embedding_matrix.shape)\n",
    "plt.plot(embedding_matrix[16])\n",
    "plt.plot(embedding_matrix[37])\n",
    "plt.plot(embedding_matrix[18])\n",
    "plt.title('example vectors')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(3952, 300)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "Text(0.5,1,'example vectors')"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "print(embedding_matrix_candidate_ans.shape)\n",
    "plt.plot(embedding_matrix[15])\n",
    "plt.plot(embedding_matrix[36])\n",
    "plt.plot(embedding_matrix[20])\n",
    "plt.title('example vectors')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(3569, 50) (3569, 50) (3569, 50)\n"
     ]
    }
   ],
   "source": [
    "training_question_idx = np.random.randint(padded_docs.shape[0], size=3569)\n",
    "test_question_idx = np.random.randint(padded_docs.shape[0], size=892)\n",
    "validate_question_idx = np.random.randint(padded_docs.shape[0], size=496)\n",
    "\n",
    "training_candidate_answers_idx = np.random.randint(padded_docs_candidate_ans.shape[0], size=3569)\n",
    "test_candidate_answers_idx= np.random.randint(padded_docs_candidate_ans.shape[0], size=892)\n",
    "validate_candidate_answers_idx= np.random.randint(padded_docs_candidate_ans.shape[0], size=496)\n",
    "\n",
    "training_answers_idx = np.random.randint(padded_docs_ans.shape[0], size=3569)\n",
    "test_answers_idx= np.random.randint(padded_docs_ans.shape[0], size=892)\n",
    "validate_answers_idx= np.random.randint(padded_docs_ans.shape[0], size=496)\n",
    "\n",
    "\n",
    "training_Q, test_Q, val_Q = padded_docs[training_question_idx,:], padded_docs[test_question_idx,:], padded_docs[validate_question_idx,:] \n",
    "training_A, test_A, val_A = padded_docs_ans[training_answers_idx,:], padded_docs_ans[test_answers_idx,:], padded_docs_ans[validate_answers_idx,:] \n",
    "training_C_A, test_C_A, val_C_A = padded_docs_candidate_ans[training_candidate_answers_idx,:], padded_docs_candidate_ans[test_candidate_answers_idx,:], padded_docs_candidate_ans[validate_candidate_answers_idx,:]\n",
    "\n",
    "print(training_Q.shape,training_A.shape,training_C_A.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 4171 unique tokens in questions.\n",
      "Found 4088 unique tokens in questions.\n",
      "Found 4108 unique tokens in questions.\n",
      "Found 4089 unique tokens in questions.\n"
     ]
    }
   ],
   "source": [
    "#####################Option A ##########################################\n",
    "t_optionA = Tokenizer()\n",
    "t_optionA.fit_on_texts(X_Answers['OptionA'])\n",
    "\n",
    "word_index_optionA = t_optionA.word_index\n",
    "print('Found %s unique tokens in questions.' % len(word_index_optionA)) \n",
    "\n",
    "# integer encode the questions\n",
    "encoded_docs_optionA = t_optionA.texts_to_sequences(X_Answers['OptionA'])\n",
    "# pad documents to a max length of 100 words\n",
    "padded_docs__optionA = pad_sequences(encoded_docs_optionA, maxlen=MAX_SEQUENCE_LENGTH, padding='post')\n",
    "\n",
    "\n",
    "# EMBEDDING_DIM =  ## seems to need to match the embeddings_index dimension\n",
    "EMBEDDING_DIM_ANS_OPTIONA = embeddings_index.get('a').shape[0]\n",
    "num_words_ans_optionA = min(MAX_NUM_WORDS, len(word_index_optionA)) + 1\n",
    "embedding_matrix_ans_option_A = np.zeros((num_words_ans_optionA, EMBEDDING_DIM_ANS_OPTIONA))\n",
    "\n",
    "\n",
    "for word, i in word_index_optionA.items():\n",
    "    if i > MAX_NUM_WORDS:\n",
    "        continue\n",
    "    embedding_vector_ans_optionA = embeddings_index.get(word) ## This references the loaded embeddings dictionary\n",
    "    if embedding_vector_ans_optionA is not None:\n",
    "        # words not found in embedding index will be all-zeros.\n",
    "        embedding_matrix_ans_option_A[i] = embedding_vector_ans_optionA\n",
    "\n",
    "\n",
    "#####################Option B ##########################################\n",
    "t_optionB = Tokenizer(num_words=MAX_NUM_WORDS)\n",
    "t_optionB.fit_on_texts(X_Answers['OptionB'])\n",
    "\n",
    "word_index_optionB = t_optionB.word_index\n",
    "print('Found %s unique tokens in questions.' % len(word_index_optionB)) \n",
    "\n",
    "# integer encode the questions\n",
    "encoded_docs_optionB = t_optionB.texts_to_sequences(X_Answers['OptionB'])\n",
    "# pad documents to a max length of 100 words\n",
    "padded_docs__optionB = pad_sequences(encoded_docs_optionB, maxlen=MAX_SEQUENCE_LENGTH, padding='post')\n",
    "\n",
    "\n",
    "# EMBEDDING_DIM =  ## seems to need to match the embeddings_index dimension\n",
    "EMBEDDING_DIM_ANS_OPTIONB = embeddings_index.get('a').shape[0]\n",
    "num_words_ans_optionB = min(MAX_NUM_WORDS, len(word_index_optionB)) + 1\n",
    "embedding_matrix_ans_option_B = np.zeros((num_words_ans_optionB, EMBEDDING_DIM_ANS_OPTIONB))\n",
    "\n",
    "\n",
    "for word, i in word_index_optionB.items():\n",
    "    if i > MAX_NUM_WORDS:\n",
    "        continue\n",
    "    embedding_vector_ans_optionB = embeddings_index.get(word) ## This references the loaded embeddings dictionary\n",
    "    if embedding_vector_ans_optionB is not None:\n",
    "        # words not found in embedding index will be all-zeros.\n",
    "        embedding_matrix_ans_option_B[i] = embedding_vector_ans_optionB\n",
    "\n",
    "        \n",
    "#####################Option C ##########################################\n",
    "t_optionC = Tokenizer(num_words=MAX_NUM_WORDS)\n",
    "t_optionC.fit_on_texts(X_Answers['OptionC'])\n",
    "\n",
    "word_index_optionC = t_optionC.word_index\n",
    "print('Found %s unique tokens in questions.' % len(word_index_optionC)) \n",
    "\n",
    "# integer encode the questions\n",
    "encoded_docs_optionC = t_optionC.texts_to_sequences(X_Answers['OptionC'])\n",
    "# pad documents to a max length of 100 words\n",
    "padded_docs__optionC = pad_sequences(encoded_docs_optionC, maxlen=MAX_SEQUENCE_LENGTH, padding='post')\n",
    "\n",
    "\n",
    "# EMBEDDING_DIM =  ## seems to need to match the embeddings_index dimension\n",
    "EMBEDDING_DIM_ANS_OPTIONC = embeddings_index.get('a').shape[0]\n",
    "num_words_ans_optionC = min(MAX_NUM_WORDS, len(word_index_optionC)) + 1\n",
    "embedding_matrix_ans_option_C = np.zeros((num_words_ans_optionC, EMBEDDING_DIM_ANS_OPTIONC))\n",
    "\n",
    "\n",
    "for word, i in word_index_optionC.items():\n",
    "    if i > MAX_NUM_WORDS:\n",
    "        continue\n",
    "    embedding_vector_ans_optionC = embeddings_index.get(word) ## This references the loaded embeddings dictionary\n",
    "    if embedding_vector_ans_optionC is not None:\n",
    "        # words not found in embedding index will be all-zeros.\n",
    "        embedding_matrix_ans_option_C[i] = embedding_vector_ans_optionC\n",
    "\n",
    "#####################Option D ##########################################\n",
    "t_optionD = Tokenizer(num_words=MAX_NUM_WORDS)\n",
    "t_optionD.fit_on_texts(X_Answers['OptionD'])\n",
    "\n",
    "word_index_optionD = t_optionD.word_index\n",
    "print('Found %s unique tokens in questions.' % len(word_index_optionD)) \n",
    "\n",
    "# integer encode the questions\n",
    "encoded_docs_optionD = t_optionD.texts_to_sequences(X_Answers['OptionD'])\n",
    "# pad documents to a max length of 100 words\n",
    "padded_docs__optionD = pad_sequences(encoded_docs_optionD, maxlen=MAX_SEQUENCE_LENGTH, padding='post')\n",
    "\n",
    "# EMBEDDING_DIM =  ## seems to need to match the embeddings_index dimension\n",
    "EMBEDDING_DIM_ANS_OPTIOND = embeddings_index.get('a').shape[0]\n",
    "num_words_ans_optionD = min(MAX_NUM_WORDS, len(word_index_optionD)) + 1\n",
    "embedding_matrix_ans_option_D = np.zeros((num_words_ans_optionD, EMBEDDING_DIM_ANS_OPTIOND))\n",
    "\n",
    "\n",
    "for word, i in word_index_optionD.items():\n",
    "    if i > MAX_NUM_WORDS:\n",
    "        continue\n",
    "    embedding_vector_ans_optionD = embeddings_index.get(word) ## This references the loaded embeddings dictionary\n",
    "    if embedding_vector_ans_optionD is not None:\n",
    "        # words not found in embedding index will be all-zeros.\n",
    "        embedding_matrix_ans_option_D[i] = embedding_vector_ans_optionD"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(3569, 50) (3569, 50) (3569, 50)\n"
     ]
    }
   ],
   "source": [
    "training__OptionA_idx = np.random.randint(padded_docs__optionA.shape[0], size=3569)\n",
    "testing__OptionA_idx = np.random.randint(padded_docs__optionA.shape[0], size=892)\n",
    "validate_OptionA_idx = np.random.randint(padded_docs__optionA.shape[0], size=3569)\n",
    "\n",
    "training__OptionB_idx = np.random.randint(padded_docs__optionB.shape[0], size=3569)\n",
    "testing__OptionB_idx = np.random.randint(padded_docs__optionB.shape[0], size=892)\n",
    "validate_OptionB_idx = np.random.randint(padded_docs__optionB.shape[0], size=496)\n",
    "\n",
    "training__OptionC_idx = np.random.randint(padded_docs__optionC.shape[0], size=3569)\n",
    "testing__OptionC_idx = np.random.randint(padded_docs__optionC.shape[0], size=892)\n",
    "validate_OptionC_idx = np.random.randint(padded_docs__optionC.shape[0], size=496)\n",
    "\n",
    "training__OptionD_idx = np.random.randint(padded_docs__optionD.shape[0], size=3569)\n",
    "testing__OptionD_idx = np.random.randint(padded_docs__optionD.shape[0], size=892)\n",
    "validate_OptionD_idx = np.random.randint(padded_docs__optionD.shape[0], size=496)\n",
    "\n",
    "train_OptionA, test_OptionA = padded_docs__optionA[training__OptionA_idx,:], padded_docs__optionA[testing__OptionA_idx,:]\n",
    "train_OptionB, test_OptionB = padded_docs__optionB[training__OptionB_idx,:], padded_docs__optionB[testing__OptionB_idx,:] \n",
    "train_OptionC, test_OptionC = padded_docs__optionC[training__OptionC_idx,:], padded_docs__optionC[testing__OptionC_idx,:]\n",
    "train_OptionD, test_OptionD = padded_docs__optionD[training__OptionD_idx,:], padded_docs__optionD[testing__OptionD_idx,:]\n",
    "\n",
    "val_OptionA, val_OptionB, val_OptionC, val_OptionD = padded_docs__optionA[validate_OptionA_idx,:], padded_docs__optionB[validate_OptionB_idx,:], padded_docs__optionC[validate_OptionC_idx,:], padded_docs__optionD[validate_OptionD_idx,:]\n",
    "\n",
    "print(val_OptionA.shape,train_OptionA.shape, train_OptionB.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1735, 300)"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len((t_knowledge_text.word_index))\n",
    "embedding_matrix_knowledge_text.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "epochs = 10\n",
    "emb_dim = 128\n",
    "batch_size = 256\n",
    "batch_size_lstm = 32\n",
    "epochs_lstm = 50"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_2 (Embedding)      (None, 50, 100)           1000000   \n",
      "_________________________________________________________________\n",
      "spatial_dropout1d_2 (Spatial (None, 50, 100)           0         \n",
      "_________________________________________________________________\n",
      "lstm_2 (LSTM)                (None, 64)                42240     \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 4)                 260       \n",
      "=================================================================\n",
      "Total params: 1,042,500\n",
      "Trainable params: 1,042,500\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "None\n",
      "Train on 2973 samples, validate on 744 samples\n",
      "Epoch 1/10\n",
      "2973/2973 [==============================] - 3s 1ms/step - loss: 1.3871 - categorical_accuracy: 0.2688 - acc: 0.2688 - val_loss: 1.3844 - val_categorical_accuracy: 0.2823 - val_acc: 0.2823\n",
      "Epoch 2/10\n",
      "2973/2973 [==============================] - 1s 467us/step - loss: 1.3907 - categorical_accuracy: 0.2577 - acc: 0.2577 - val_loss: 1.3842 - val_categorical_accuracy: 0.2823 - val_acc: 0.2823\n",
      "Epoch 3/10\n",
      "2973/2973 [==============================] - 1s 455us/step - loss: 1.3887 - categorical_accuracy: 0.2540 - acc: 0.2540 - val_loss: 1.3840 - val_categorical_accuracy: 0.2823 - val_acc: 0.2823\n",
      "Epoch 4/10\n",
      "2973/2973 [==============================] - 1s 453us/step - loss: 1.3880 - categorical_accuracy: 0.2654 - acc: 0.2654 - val_loss: 1.3842 - val_categorical_accuracy: 0.2823 - val_acc: 0.2823\n",
      "Epoch 5/10\n",
      "2973/2973 [==============================] - 1s 452us/step - loss: 1.3880 - categorical_accuracy: 0.2627 - acc: 0.2627 - val_loss: 1.3844 - val_categorical_accuracy: 0.2823 - val_acc: 0.2823\n",
      "Epoch 6/10\n",
      "2973/2973 [==============================] - 1s 452us/step - loss: 1.3861 - categorical_accuracy: 0.2667 - acc: 0.2667 - val_loss: 1.3841 - val_categorical_accuracy: 0.2823 - val_acc: 0.2823\n",
      "Epoch 7/10\n",
      "2973/2973 [==============================] - 1s 456us/step - loss: 1.3865 - categorical_accuracy: 0.2661 - acc: 0.2661 - val_loss: 1.3839 - val_categorical_accuracy: 0.2823 - val_acc: 0.2823\n",
      "Epoch 8/10\n",
      "2973/2973 [==============================] - 1s 460us/step - loss: 1.3872 - categorical_accuracy: 0.2704 - acc: 0.2704 - val_loss: 1.3840 - val_categorical_accuracy: 0.2823 - val_acc: 0.2823\n",
      "Epoch 9/10\n",
      "2973/2973 [==============================] - 1s 456us/step - loss: 1.3864 - categorical_accuracy: 0.2624 - acc: 0.2624 - val_loss: 1.3843 - val_categorical_accuracy: 0.2823 - val_acc: 0.2823\n",
      "Epoch 10/10\n",
      "2973/2973 [==============================] - 1s 458us/step - loss: 1.3858 - categorical_accuracy: 0.2640 - acc: 0.2640 - val_loss: 1.3842 - val_categorical_accuracy: 0.2823 - val_acc: 0.2823\n"
     ]
    }
   ],
   "source": [
    "#Simple model 1 \n",
    "model = Sequential()\n",
    "model.add(Embedding(MAX_NUM_WORDS,100, input_length=X_Question_train.shape[1]))\n",
    "model.add(SpatialDropout1D(0.7))\n",
    "model.add(LSTM(64, dropout=0.7, recurrent_dropout=0.7))\n",
    "model.add(Dense(4, activation='softmax'))\n",
    "model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['categorical_accuracy', 'acc'])\n",
    "print(model.summary())\n",
    "history = model.fit(X_Question_train, Y_train, epochs=epochs, \n",
    "                    batch_size=batch_size,validation_split=0.2,callbacks=[EarlyStopping(monitor='val_loss',patience=7, min_delta=0.0001)])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1240/1240 [==============================] - 2s 1ms/step\n",
      "Test set\n",
      "  Loss: 1.384\n",
      "  Accuracy: 0.286\n"
     ]
    }
   ],
   "source": [
    "accr = model.evaluate(X_Question_test,Y_test)\n",
    "print('Test set\\n  Loss: {:0.3f}\\n  Accuracy: {:0.3f}'.format(accr[0],accr[1]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['loss', 'categorical_accuracy', 'acc']\n"
     ]
    }
   ],
   "source": [
    "print(model.metrics_names)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [],
   "source": [
    "#QA LSTM 1\n",
    "question_input = Input(shape=(MAX_SEQUENCE_LENGTH,), dtype='int32')\n",
    "embedded_question = Embedding(input_dim=len(t.word_index)+1, output_dim=300, input_length=MAX_SEQUENCE_LENGTH, weights=[embedding_matrix] ,trainable=False,mask_zero=True)(question_input)\n",
    "encoded_question = (LSTM(300,dropout=0.7, recurrent_dropout=0.7))(embedded_question)\n",
    "\n",
    "kt_input = Input(shape=(MAX_SEQUENCE_LENGTH,), dtype='int32')\n",
    "xembedded_kt = Embedding(input_dim=len(t_knowledge_text.word_index)+1 , output_dim=300, input_length=MAX_SEQUENCE_LENGTH, weights=[embedding_matrix_knowledge_text],trainable=False, mask_zero=True)(kt_input)\n",
    "xencoded_kt = (LSTM(300,dropout=0.7, recurrent_dropout=0.7))(xembedded_kt)\n",
    "\n",
    "\n",
    "merged = concatenate([encoded_question,xencoded_kt])\n",
    "output = Dense(4, activation='softmax')(merged)\n",
    "\n",
    "# This is our final model:\n",
    "qa_model = Model(inputs=[question_input,kt_input], outputs=output)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_13 (InputLayer)           (None, 50)           0                                            \n",
      "__________________________________________________________________________________________________\n",
      "input_14 (InputLayer)           (None, 50)           0                                            \n",
      "__________________________________________________________________________________________________\n",
      "embedding_20 (Embedding)        (None, 50, 300)      1575000     input_13[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "embedding_21 (Embedding)        (None, 50, 300)      520500      input_14[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "lstm_17 (LSTM)                  (None, 300)          721200      embedding_20[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "lstm_18 (LSTM)                  (None, 300)          721200      embedding_21[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_7 (Concatenate)     (None, 600)          0           lstm_17[0][0]                    \n",
      "                                                                 lstm_18[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dense_11 (Dense)                (None, 4)            2404        concatenate_7[0][0]              \n",
      "==================================================================================================\n",
      "Total params: 3,540,304\n",
      "Trainable params: 1,444,804\n",
      "Non-trainable params: 2,095,500\n",
      "__________________________________________________________________________________________________\n",
      "None\n"
     ]
    }
   ],
   "source": [
    "qa_model.compile(loss = 'categorical_crossentropy', optimizer='adam',metrics = ['accuracy'])\n",
    "print(qa_model.summary())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 2973 samples, validate on 744 samples\n",
      "Epoch 1/50\n",
      "2973/2973 [==============================] - 27s 9ms/step - loss: 1.4228 - acc: 0.2418 - val_loss: 1.4087 - val_acc: 0.2433\n",
      "Epoch 2/50\n",
      "2973/2973 [==============================] - 24s 8ms/step - loss: 1.3867 - acc: 0.2990 - val_loss: 1.4102 - val_acc: 0.2661\n",
      "Epoch 3/50\n",
      "2973/2973 [==============================] - 24s 8ms/step - loss: 1.3665 - acc: 0.3280 - val_loss: 1.4083 - val_acc: 0.2634\n",
      "Epoch 4/50\n",
      "2973/2973 [==============================] - 24s 8ms/step - loss: 1.3451 - acc: 0.3451 - val_loss: 1.4252 - val_acc: 0.2473\n",
      "Epoch 5/50\n",
      "2973/2973 [==============================] - 24s 8ms/step - loss: 1.3340 - acc: 0.3616 - val_loss: 1.4228 - val_acc: 0.2567\n",
      "Epoch 6/50\n",
      "2973/2973 [==============================] - 24s 8ms/step - loss: 1.3261 - acc: 0.3619 - val_loss: 1.4302 - val_acc: 0.2661\n",
      "Epoch 7/50\n",
      "2973/2973 [==============================] - 24s 8ms/step - loss: 1.3177 - acc: 0.3700 - val_loss: 1.4321 - val_acc: 0.2540\n",
      "Epoch 8/50\n",
      "2973/2973 [==============================] - 24s 8ms/step - loss: 1.2893 - acc: 0.3962 - val_loss: 1.4471 - val_acc: 0.2688\n",
      "Epoch 9/50\n",
      "2973/2973 [==============================] - 24s 8ms/step - loss: 1.2795 - acc: 0.4097 - val_loss: 1.4512 - val_acc: 0.2554\n",
      "Epoch 10/50\n",
      "2973/2973 [==============================] - 24s 8ms/step - loss: 1.2749 - acc: 0.4023 - val_loss: 1.4773 - val_acc: 0.2433\n"
     ]
    }
   ],
   "source": [
    "#qa_model.fit([X_Question_train,X_Answer_train],Y_train, epochs=2)\n",
    "\n",
    "history2 = qa_model.fit([X_Question_train,X_Answer_train], Y_train, epochs=epochs_lstm, \n",
    "                    batch_size=batch_size_lstm, validation_split=0.2,callbacks=[EarlyStopping(monitor='val_loss',patience=7, min_delta=0.0001)])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1240/1240 [==============================] - 4s 3ms/step\n",
      "Test set\n",
      "  Loss: 1.476\n",
      "  Accuracy: 0.254\n"
     ]
    }
   ],
   "source": [
    "#loss,acc= qa_model.evaluate([X_Question_test,X_Answer_test],Y_test, verbose=0)\n",
    "#print(acc*100,loss)\n",
    "\n",
    "accr = qa_model.evaluate([X_Question_test,X_Answer_test],Y_test)\n",
    "print('Test set\\n  Loss: {:0.3f}\\n  Accuracy: {:0.3f}'.format(accr[0],accr[1]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [],
   "source": [
    "#QA LSTM 2\n",
    "question_input = Input(shape=(MAX_SEQUENCE_LENGTH,), dtype='int32')\n",
    "embedded_question = Embedding(input_dim=len(t.word_index)+1, output_dim=300, input_length=MAX_SEQUENCE_LENGTH, weights=[embedding_matrix] ,trainable=False,mask_zero=True)(question_input)\n",
    "encoded_question = (Bidirectional(LSTM(300,return_sequences=False),merge_mode=\"ave\"))(embedded_question)\n",
    "encoded_question = Dropout(0.3)(encoded_question)\n",
    "encoded_question = Dropout(0.2)(encoded_question)\n",
    "\n",
    "# xanswer_input = Input(shape=(MAX_SEQUENCE_LENGTH,), dtype='int32')\n",
    "# # xembedded_answer = Embedding(input_dim=len(t_candidate_ans.word_index)+1 , output_dim=300, input_length=MAX_SEQUENCE_LENGTH, weights=[embedding_matrix_candidate_ans],trainable=False, mask_zero=True)(xanswer_input)\n",
    "# # xencoded_answer = (Bidirectional(LSTM(300,return_sequences=False),merge_mode=\"ave\"))(xembedded_answer)\n",
    "# # xencoded_answer = Dropout(0.3)(xencoded_answer)\n",
    "# # xencoded_answer = Dropout(0.2)(xencoded_answer)\n",
    "\n",
    "kt_input = Input(shape=(MAX_SEQUENCE_LENGTH,), dtype='int32')\n",
    "xembedded_kt = Embedding(input_dim=len(t_knowledge_text.word_index)+1 , output_dim=300, input_length=MAX_SEQUENCE_LENGTH, weights=[embedding_matrix_knowledge_text],trainable=False, mask_zero=True)(kt_input)\n",
    "xencoded_kt = (Bidirectional(LSTM(300,return_sequences=False),merge_mode=\"ave\"))(xembedded_kt)\n",
    "xencoded_kt = Dropout(0.3)(xencoded_kt)\n",
    "xencoded_kt = Dropout(0.2)(xencoded_kt)\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "merged = concatenate([encoded_question,xencoded_kt])\n",
    "output = Dense(4, activation='softmax')(merged)\n",
    "\n",
    "# This is our final model:\n",
    "qa_model_ltsm_2 = Model(inputs=[question_input,kt_input], outputs=output)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_15 (InputLayer)           (None, 50)           0                                            \n",
      "__________________________________________________________________________________________________\n",
      "input_16 (InputLayer)           (None, 50)           0                                            \n",
      "__________________________________________________________________________________________________\n",
      "embedding_22 (Embedding)        (None, 50, 300)      1575000     input_15[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "embedding_23 (Embedding)        (None, 50, 300)      520500      input_16[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "bidirectional_1 (Bidirectional) (None, 300)          1442400     embedding_22[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "bidirectional_2 (Bidirectional) (None, 300)          1442400     embedding_23[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "dropout_1 (Dropout)             (None, 300)          0           bidirectional_1[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "dropout_3 (Dropout)             (None, 300)          0           bidirectional_2[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "dropout_2 (Dropout)             (None, 300)          0           dropout_1[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "dropout_4 (Dropout)             (None, 300)          0           dropout_3[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_8 (Concatenate)     (None, 600)          0           dropout_2[0][0]                  \n",
      "                                                                 dropout_4[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "dense_12 (Dense)                (None, 4)            2404        concatenate_8[0][0]              \n",
      "==================================================================================================\n",
      "Total params: 4,982,704\n",
      "Trainable params: 2,887,204\n",
      "Non-trainable params: 2,095,500\n",
      "__________________________________________________________________________________________________\n",
      "None\n"
     ]
    }
   ],
   "source": [
    "adam = optimizers.Adam(lr=0.0001)\n",
    "qa_model_ltsm_2.compile(loss = 'categorical_crossentropy', optimizer=adam ,metrics = ['accuracy'])\n",
    "print(qa_model_ltsm_2.summary())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 2973 samples, validate on 744 samples\n",
      "Epoch 1/50\n",
      "2973/2973 [==============================] - 50s 17ms/step - loss: 1.3923 - acc: 0.2408 - val_loss: 1.3901 - val_acc: 0.2688\n",
      "Epoch 2/50\n",
      "2973/2973 [==============================] - 45s 15ms/step - loss: 1.3726 - acc: 0.3017 - val_loss: 1.3899 - val_acc: 0.2688\n",
      "Epoch 3/50\n",
      "2973/2973 [==============================] - 44s 15ms/step - loss: 1.3596 - acc: 0.3428 - val_loss: 1.3877 - val_acc: 0.2675\n",
      "Epoch 4/50\n",
      "2973/2973 [==============================] - 44s 15ms/step - loss: 1.3484 - acc: 0.3599 - val_loss: 1.3890 - val_acc: 0.2581\n",
      "Epoch 5/50\n",
      "2973/2973 [==============================] - 44s 15ms/step - loss: 1.3341 - acc: 0.3828 - val_loss: 1.3906 - val_acc: 0.2634\n",
      "Epoch 6/50\n",
      "2973/2973 [==============================] - 45s 15ms/step - loss: 1.3205 - acc: 0.4057 - val_loss: 1.3904 - val_acc: 0.2728\n",
      "Epoch 7/50\n",
      "2973/2973 [==============================] - 45s 15ms/step - loss: 1.3059 - acc: 0.4114 - val_loss: 1.3952 - val_acc: 0.2715\n",
      "Epoch 8/50\n",
      "2973/2973 [==============================] - 44s 15ms/step - loss: 1.2865 - acc: 0.4396 - val_loss: 1.4037 - val_acc: 0.2567\n",
      "Epoch 9/50\n",
      "2973/2973 [==============================] - 44s 15ms/step - loss: 1.2688 - acc: 0.4464 - val_loss: 1.4110 - val_acc: 0.2728\n",
      "Epoch 10/50\n",
      "1824/2973 [=================>............] - ETA: 15s - loss: 1.2446 - acc: 0.4655"
     ]
    }
   ],
   "source": [
    "history3 = qa_model_ltsm_2.fit([X_Question_train,X_Answer_train], Y_train, epochs=epochs_lstm, \n",
    "                    batch_size=batch_size_lstm, validation_split=0.2,callbacks=[EarlyStopping(monitor='val_loss',patience=7, min_delta=0.0001)])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1240/1240 [==============================] - 6s 5ms/step\n",
      "Test set\n",
      "  Loss: 1.445\n",
      "  Accuracy: 0.243\n"
     ]
    }
   ],
   "source": [
    "#loss,acc= qa_model_ltsm_2.evaluate([test_Q,test_C_A],test_A, verbose=0)\n",
    "#print(acc*100,loss)\n",
    "\n",
    "accr =  qa_model_ltsm_2.evaluate([X_Question_test,X_Answer_test],Y_test)\n",
    "print('Test set\\n  Loss: {:0.3f}\\n  Accuracy: {:0.3f}'.format(accr[0],accr[1]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Requirement already satisfied: pydot-ng in /home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages (2.0.0)\n",
      "Requirement already satisfied: pyparsing>=2.0.1 in /home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages (from pydot-ng) (2.2.0)\n",
      "\u001b[33mYou are using pip version 10.0.1, however version 18.1 is available.\n",
      "You should consider upgrading via the 'pip install --upgrade pip' command.\u001b[0m\n",
      "Requirement already satisfied: graphviz in /home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages (0.10.1)\n",
      "\u001b[33mYou are using pip version 10.0.1, however version 18.1 is available.\n",
      "You should consider upgrading via the 'pip install --upgrade pip' command.\u001b[0m\n"
     ]
    }
   ],
   "source": [
    "!pip install pydot-ng\n",
    "!pip install graphviz"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# #QA LSTM 2\n",
    "# question_input = Input(shape=(MAX_SEQUENCE_LENGTH,), dtype='int32')\n",
    "# xanswer_input = Input(shape=(MAX_SEQUENCE_LENGTH,), dtype='int32')\n",
    "# xanswer_input_optionA = Input(shape=(MAX_SEQUENCE_LENGTH,), dtype='int32')\n",
    "# xanswer_input_optionB = Input(shape=(MAX_SEQUENCE_LENGTH,), dtype='int32')\n",
    "# xanswer_input_optionC = Input(shape=(MAX_SEQUENCE_LENGTH,), dtype='int32')\n",
    "# xanswer_input_optionD = Input(shape=(MAX_SEQUENCE_LENGTH,), dtype='int32')\n",
    "\n",
    "# embedded_question = Embedding(input_dim=len(t.word_index)+1, output_dim=300, input_length=MAX_SEQUENCE_LENGTH, weights=[embedding_matrix] ,trainable=False)(question_input)\n",
    "# xembedded_answer = Embedding(input_dim=len(t_candidate_ans.word_index)+1 , output_dim=300, input_length=MAX_SEQUENCE_LENGTH, weights=[embedding_matrix_candidate_ans],trainable=False)(xanswer_input)\n",
    "# xembedded_answer_optionA = Embedding(input_dim=len(t_optionA.word_index)+1 , output_dim=300, input_length=MAX_SEQUENCE_LENGTH, weights=[embedding_matrix_ans_option_A],trainable=False)(xanswer_input_optionA)\n",
    "# xembedded_answer_optionB = Embedding(input_dim=len(t_optionB.word_index)+1 , output_dim=300, input_length=MAX_SEQUENCE_LENGTH, weights=[embedding_matrix_ans_option_B],trainable=False)(xanswer_input_optionB)\n",
    "# xembedded_answer_optionC = Embedding(input_dim=len(t_optionC.word_index)+1 , output_dim=300, input_length=MAX_SEQUENCE_LENGTH, weights=[embedding_matrix_ans_option_C],trainable=False)(xanswer_input_optionC)\n",
    "# xembedded_answer_optionD = Embedding(input_dim=len(t_optionD.word_index)+1 , output_dim=300, input_length=MAX_SEQUENCE_LENGTH, weights=[embedding_matrix_ans_option_D],trainable=False)(xanswer_input_optionD)\n",
    "\n",
    "# merged = concatenate([embedded_question,xembedded_answer,xembedded_answer_optionA,xembedded_answer_optionB,xembedded_answer_optionC,xembedded_answer_optionD])\n",
    "# f_o = Flatten()(merged)\n",
    "# output = Dense(50, activation='softmax')(f_o)\n",
    "\n",
    "# qa_model_2 = Model(inputs=[question_input,xanswer_input,xanswer_input_optionA,xanswer_input_optionB,xanswer_input_optionC,xanswer_input_optionD], outputs=output)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# qa_model_2.compile(loss = 'cosine_proximity', optimizer='sgd' ,metrics = ['accuracy'])\n",
    "# print(qa_model_2.summary())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# qa_model_2.fit([training_Q,training_C_A,train_OptionA,train_OptionB,train_OptionC,train_OptionD],training_A, epochs=2)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# loss,acc= qa_model_2.evaluate([test_Q,test_C_A,test_OptionA,test_OptionB,test_OptionC,test_OptionD],test_A, verbose=0)\n",
    "# print(acc*100,loss)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'qa_model_ltsm_2' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-2-e0c988e66037>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mpred_Actual\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mqa_model_ltsm_2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpredict\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mval_Q\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mval_A\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      2\u001b[0m \u001b[0mpred_OptionA\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mqa_model_ltsm_2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpredict\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mval_Q\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mval_OptionA\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      3\u001b[0m \u001b[0mpred_OptionB\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mqa_model_ltsm_2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpredict\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mval_Q\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mval_OptionB\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      4\u001b[0m \u001b[0mpred_OptionC\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mqa_model_ltsm_2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpredict\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mval_Q\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mval_OptionC\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      5\u001b[0m \u001b[0mpred_OptionD\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mqa_model_ltsm_2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpredict\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mval_Q\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mval_OptionD\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mNameError\u001b[0m: name 'qa_model_ltsm_2' is not defined"
     ]
    }
   ],
   "source": [
    "pred_Actual = qa_model_ltsm_2.predict([val_Q,val_A])\n",
    "pred_OptionA = qa_model_ltsm_2.predict([val_Q,val_OptionA])\n",
    "pred_OptionB = qa_model_ltsm_2.predict([val_Q,val_OptionB])\n",
    "pred_OptionC = qa_model_ltsm_2.predict([val_Q,val_OptionC])\n",
    "pred_OptionD = qa_model_ltsm_2.predict([val_Q,val_OptionD])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 228,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[3.1636861e-01, 2.4699937e-01, 7.7058032e-02, ..., 2.6708059e-07,\n",
       "        2.1902258e-07, 4.5899321e-07],\n",
       "       [6.8173110e-01, 2.8633150e-01, 1.8918674e-02, ..., 1.3709538e-07,\n",
       "        8.9507168e-08, 1.0063054e-07],\n",
       "       [4.1624388e-01, 3.3703947e-01, 2.0543455e-01, ..., 2.5496129e-06,\n",
       "        1.5343737e-06, 1.4949371e-06],\n",
       "       ...,\n",
       "       [6.6174132e-01, 2.2454941e-01, 8.6836174e-02, ..., 3.9186568e-07,\n",
       "        2.6508587e-07, 2.6135220e-07],\n",
       "       [5.8591616e-01, 3.4895763e-01, 4.9634103e-02, ..., 3.0762010e-07,\n",
       "        1.9900428e-07, 1.9822464e-07],\n",
       "       [6.1320698e-01, 2.5521067e-01, 9.9593416e-02, ..., 2.8162424e-07,\n",
       "        1.5990919e-07, 2.5347788e-07]], dtype=float32)"
      ]
     },
     "execution_count": 228,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred_Actual"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 229,
   "metadata": {},
   "outputs": [
    {
     "ename": "ValueError",
     "evalue": "continuous-multioutput is not supported",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mValueError\u001b[0m                                Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-229-fef6afbc76ed>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[0;32mfrom\u001b[0m \u001b[0msklearn\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmetrics\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0mconfusion_matrix\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 2\u001b[0;31m \u001b[0mcm\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mconfusion_matrix\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mpred_OptionA\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mpred_Actual\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m~/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages/sklearn/metrics/classification.py\u001b[0m in \u001b[0;36mconfusion_matrix\u001b[0;34m(y_true, y_pred, labels, sample_weight)\u001b[0m\n\u001b[1;32m    248\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    249\u001b[0m     \"\"\"\n\u001b[0;32m--> 250\u001b[0;31m     \u001b[0my_type\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_true\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_pred\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0m_check_targets\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0my_true\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_pred\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    251\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0my_type\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0;32min\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0;34m\"binary\"\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m\"multiclass\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    252\u001b[0m         \u001b[0;32mraise\u001b[0m \u001b[0mValueError\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"%s is not supported\"\u001b[0m \u001b[0;34m%\u001b[0m \u001b[0my_type\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages/sklearn/metrics/classification.py\u001b[0m in \u001b[0;36m_check_targets\u001b[0;34m(y_true, y_pred)\u001b[0m\n\u001b[1;32m     86\u001b[0m     \u001b[0;31m# No metrics support \"multiclass-multioutput\" format\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     87\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0my_type\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0;32min\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0;34m\"binary\"\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m\"multiclass\"\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m\"multilabel-indicator\"\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 88\u001b[0;31m         \u001b[0;32mraise\u001b[0m \u001b[0mValueError\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"{0} is not supported\"\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mformat\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0my_type\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     89\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     90\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0my_type\u001b[0m \u001b[0;32min\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0;34m\"binary\"\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m\"multiclass\"\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mValueError\u001b[0m: continuous-multioutput is not supported"
     ]
    }
   ],
   "source": [
    "from sklearn.metrics import confusion_matrix\n",
    "cm = confusion_matrix(pred_OptionA, pred_Actual)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 212,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras import backend as K\n",
    "\n",
    "def cosine_distance(vests):\n",
    "    x, y = vests\n",
    "    x = K.l2_normalize(x, axis=-1)\n",
    "    y = K.l2_normalize(y, axis=-1)\n",
    "    return -K.mean(x * y, axis=-1, keepdims=True)\n",
    "\n",
    "def cos_dist_output_shape(shapes):\n",
    "    shape1, shape2 = shapes\n",
    "    return (shape1[0],1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 213,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "\n",
    "data_Actual = tf.convert_to_tensor(pred_Actual, np.float32)\n",
    "data_option_A = tf.convert_to_tensor(pred_OptionA, np.float32)\n",
    "data_option_B = tf.convert_to_tensor(pred_OptionB, np.float32)\n",
    "data_option_C = tf.convert_to_tensor(pred_OptionC, np.float32)\n",
    "data_option_D = tf.convert_to_tensor(pred_OptionD, np.float32)\n",
    "\n",
    "\n",
    "distance_Actual_to_OptionA = Lambda(cosine_distance, output_shape=cos_dist_output_shape)([data_Actual, data_option_A])\n",
    "distance_Actual_to_OptionB = Lambda(cosine_distance, output_shape=cos_dist_output_shape)([data_Actual, data_option_B])\n",
    "distance_Actual_to_OptionC = Lambda(cosine_distance, output_shape=cos_dist_output_shape)([data_Actual, data_option_C])\n",
    "distance_Actual_to_OptionD = Lambda(cosine_distance, output_shape=cos_dist_output_shape)([data_Actual, data_option_D])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 214,
   "metadata": {},
   "outputs": [],
   "source": [
    "predicted_answer=[]\n",
    "with tf.Session() as sess:\n",
    "    predicted_answer = tf.reduce_min(tf.stack([distance_Actual_to_OptionA,distance_Actual_to_OptionB,distance_Actual_to_OptionC,distance_Actual_to_OptionD]),0).eval()\n",
    "    \n",
    "bool_optionA= tf.equal(predicted_answer, distance_Actual_to_OptionA)\n",
    "bool_optionB= tf.equal(predicted_answer, distance_Actual_to_OptionB)\n",
    "bool_optionC= tf.equal(predicted_answer, distance_Actual_to_OptionC)\n",
    "bool_optionD= tf.equal(predicted_answer, distance_Actual_to_OptionD)\n",
    "\n",
    "array_bool_optionA = tf.Session().run(bool_optionA)\n",
    "array_bool_optionB = tf.Session().run(bool_optionB)\n",
    "array_bool_optionC = tf.Session().run(bool_optionC)\n",
    "array_bool_optionD = tf.Session().run(bool_optionD)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 215,
   "metadata": {},
   "outputs": [],
   "source": [
    "df_final = pd.DataFrame(array_bool_optionA)\n",
    "df_final['B'] = pd.DataFrame(array_bool_optionB)\n",
    "df_final['C'] = pd.DataFrame(array_bool_optionC)\n",
    "df_final['D'] = pd.DataFrame(array_bool_optionD)\n",
    "df_final['A'] =df_final[0] \n",
    "df_final = df_final.drop(columns=[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 221,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((510, 10), 518, (4957, 10))"
      ]
     },
     "execution_count": 221,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "\n",
    "Validation_train.shape,len(flat_list_true_col),train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 217,
   "metadata": {},
   "outputs": [],
   "source": [
    "true_columns = []\n",
    "for index, row in df_final.iterrows():\n",
    "    true_col_list = [col for col in df_final.columns if row[col]]\n",
    "    true_columns.append(true_col_list)\n",
    "flat_list_true_col = [item for sublist in true_columns for item in sublist]    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 223,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages/ipykernel/__main__.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/indexing.html#indexing-view-versus-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>ID</th>\n",
       "      <th>Question Stem</th>\n",
       "      <th>Choices</th>\n",
       "      <th>Complete Question</th>\n",
       "      <th>Answer Key</th>\n",
       "      <th>OptionA</th>\n",
       "      <th>OptionB</th>\n",
       "      <th>OptionC</th>\n",
       "      <th>OptionD</th>\n",
       "      <th>Result</th>\n",
       "      <th>predicted_answers</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>4439</th>\n",
       "      <td>14-647</td>\n",
       "      <td>Which is true</td>\n",
       "      <td>(A) an anemometer would measure a tropic front...</td>\n",
       "      <td>Which is true (A) an anemometer would measure ...</td>\n",
       "      <td>A</td>\n",
       "      <td>an anemometer would measure a tropic front</td>\n",
       "      <td>an anemometer would measure thunder</td>\n",
       "      <td>an anemometer would measure a rainbow</td>\n",
       "      <td>an anemometer would measure a lightning bolt</td>\n",
       "      <td>an anemometer would measure a tropic front</td>\n",
       "      <td>A</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4440</th>\n",
       "      <td>14-649</td>\n",
       "      <td>What time are you most likely to see a rainbow</td>\n",
       "      <td>(A) 10:00 A.M. (B) 9:00 P.M. (C) 3 A.M. (D) Mi...</td>\n",
       "      <td>What time are you most likely to see a rainbow...</td>\n",
       "      <td>A</td>\n",
       "      <td>10:00 A.M.</td>\n",
       "      <td>9:00 P.M.</td>\n",
       "      <td>3 A.M.</td>\n",
       "      <td>Midnight</td>\n",
       "      <td>10:00 A.M.</td>\n",
       "      <td>C</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4441</th>\n",
       "      <td>14-651</td>\n",
       "      <td>Why might a polar bear grow white hair?</td>\n",
       "      <td>(A) blend in (B) stand out (C) random (D) look...</td>\n",
       "      <td>Why might a polar bear grow white hair? (A) bl...</td>\n",
       "      <td>A</td>\n",
       "      <td>blend in</td>\n",
       "      <td>stand out</td>\n",
       "      <td>random</td>\n",
       "      <td>look fancy</td>\n",
       "      <td>blend in</td>\n",
       "      <td>D</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4442</th>\n",
       "      <td>14-655</td>\n",
       "      <td>As the earth glides from winter into spring yo...</td>\n",
       "      <td>(A) leaves growing (B) blizzards (C) bears sta...</td>\n",
       "      <td>As the earth glides from winter into spring yo...</td>\n",
       "      <td>A</td>\n",
       "      <td>leaves growing</td>\n",
       "      <td>blizzards</td>\n",
       "      <td>bears starting hibernation</td>\n",
       "      <td>leaves turning orange</td>\n",
       "      <td>leaves growing</td>\n",
       "      <td>D</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4443</th>\n",
       "      <td>14-656</td>\n",
       "      <td>Which of the following actions can make electr...</td>\n",
       "      <td>(A) Clap on! (B) Jumping in place (C) Turning ...</td>\n",
       "      <td>Which of the following actions can make electr...</td>\n",
       "      <td>A</td>\n",
       "      <td>Clap on!</td>\n",
       "      <td>Jumping in place</td>\n",
       "      <td>Turning a doorknob</td>\n",
       "      <td>Rubbing your nose</td>\n",
       "      <td>Clap on!</td>\n",
       "      <td>B</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4444</th>\n",
       "      <td>14-658</td>\n",
       "      <td>Which is an electrical energy conductor?</td>\n",
       "      <td>(A) cobalt (B) plastic (C) concrete (D) mahogany</td>\n",
       "      <td>Which is an electrical energy conductor? (A) c...</td>\n",
       "      <td>A</td>\n",
       "      <td>cobalt</td>\n",
       "      <td>plastic</td>\n",
       "      <td>concrete</td>\n",
       "      <td>mahogany</td>\n",
       "      <td>cobalt</td>\n",
       "      <td>B</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4445</th>\n",
       "      <td>14-659</td>\n",
       "      <td>This object in your pocket is the most likely ...</td>\n",
       "      <td>(A) your phone (B) your credit card (C) your l...</td>\n",
       "      <td>This object in your pocket is the most likely ...</td>\n",
       "      <td>A</td>\n",
       "      <td>your phone</td>\n",
       "      <td>your credit card</td>\n",
       "      <td>your library card</td>\n",
       "      <td>your wallet</td>\n",
       "      <td>your phone</td>\n",
       "      <td>B</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4446</th>\n",
       "      <td>14-661</td>\n",
       "      <td>If I wanted to build endurance I could</td>\n",
       "      <td>(A) cycling to Publix (B) eat more food (C) sl...</td>\n",
       "      <td>If I wanted to build endurance I could (A) cyc...</td>\n",
       "      <td>A</td>\n",
       "      <td>cycling to Publix</td>\n",
       "      <td>eat more food</td>\n",
       "      <td>sleep all day</td>\n",
       "      <td>only eat pizza</td>\n",
       "      <td>cycling to Publix</td>\n",
       "      <td>D</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4447</th>\n",
       "      <td>14-664</td>\n",
       "      <td>An example of combining two substances is</td>\n",
       "      <td>(A) pouring queso into guacamole (B) setting a...</td>\n",
       "      <td>An example of combining two substances is (A) ...</td>\n",
       "      <td>A</td>\n",
       "      <td>pouring queso into guacamole</td>\n",
       "      <td>setting a laptop on a desk</td>\n",
       "      <td>building a roof on a house</td>\n",
       "      <td>putting clothes in a suitcase</td>\n",
       "      <td>pouring queso into guacamole</td>\n",
       "      <td>C</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4448</th>\n",
       "      <td>14-665</td>\n",
       "      <td>A plant needing to photosynthesize will best b...</td>\n",
       "      <td>(A) in a roofless room (B) in a cardboard box ...</td>\n",
       "      <td>A plant needing to photosynthesize will best b...</td>\n",
       "      <td>A</td>\n",
       "      <td>in a roofless room</td>\n",
       "      <td>in a cardboard box</td>\n",
       "      <td>in a windowless room</td>\n",
       "      <td>in a car with tinted windows</td>\n",
       "      <td>in a roofless room</td>\n",
       "      <td>D</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4449</th>\n",
       "      <td>14-667</td>\n",
       "      <td>If a being is moving regularly, then that bein...</td>\n",
       "      <td>(A) hardly healthy (B) probably strong (C) abo...</td>\n",
       "      <td>If a being is moving regularly, then that bein...</td>\n",
       "      <td>B</td>\n",
       "      <td>hardly healthy</td>\n",
       "      <td>probably strong</td>\n",
       "      <td>about to die</td>\n",
       "      <td>making little motion</td>\n",
       "      <td>probably strong</td>\n",
       "      <td>C</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4450</th>\n",
       "      <td>14-668</td>\n",
       "      <td>What kind of impact does exercise have on a hu...</td>\n",
       "      <td>(A) terrible (B) low (C) skyrocketing (D) bad</td>\n",
       "      <td>What kind of impact does exercise have on a hu...</td>\n",
       "      <td>C</td>\n",
       "      <td>terrible</td>\n",
       "      <td>low</td>\n",
       "      <td>skyrocketing</td>\n",
       "      <td>bad</td>\n",
       "      <td>skyrocketing</td>\n",
       "      <td>D</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4451</th>\n",
       "      <td>14-669</td>\n",
       "      <td>Animals help with seed dispersal due to the se...</td>\n",
       "      <td>(A) expelled via excrement (B) smiled (C) expe...</td>\n",
       "      <td>Animals help with seed dispersal due to the se...</td>\n",
       "      <td>A</td>\n",
       "      <td>expelled via excrement</td>\n",
       "      <td>smiled</td>\n",
       "      <td>expelled from school</td>\n",
       "      <td>kicked</td>\n",
       "      <td>expelled via excrement</td>\n",
       "      <td>C</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4452</th>\n",
       "      <td>14-670</td>\n",
       "      <td>f a plant shows up in a new place, what might ...</td>\n",
       "      <td>(A) the plant is able to relocate itself (B) i...</td>\n",
       "      <td>f a plant shows up in a new place, what might ...</td>\n",
       "      <td>C</td>\n",
       "      <td>the plant is able to relocate itself</td>\n",
       "      <td>it spontaneously appeared there</td>\n",
       "      <td>its cased embryo have been dispersed</td>\n",
       "      <td>it walked to the new place</td>\n",
       "      <td>its cased embryo have been dispersed</td>\n",
       "      <td>D</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4453</th>\n",
       "      <td>14-671</td>\n",
       "      <td>A creature that is a biped and which regularly...</td>\n",
       "      <td>(A) kettle shards (B) old feces (C) chia seeds...</td>\n",
       "      <td>A creature that is a biped and which regularly...</td>\n",
       "      <td>C</td>\n",
       "      <td>kettle shards</td>\n",
       "      <td>old feces</td>\n",
       "      <td>chia seeds</td>\n",
       "      <td>hard rocks</td>\n",
       "      <td>chia seeds</td>\n",
       "      <td>D</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4454</th>\n",
       "      <td>14-672</td>\n",
       "      <td>Humans sometimes eat what?</td>\n",
       "      <td>(A) air (B) kernels (C) ocean water (D) rocks</td>\n",
       "      <td>Humans sometimes eat what? (A) air (B) kernels...</td>\n",
       "      <td>B</td>\n",
       "      <td>air</td>\n",
       "      <td>kernels</td>\n",
       "      <td>ocean water</td>\n",
       "      <td>rocks</td>\n",
       "      <td>kernels</td>\n",
       "      <td>A</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4455</th>\n",
       "      <td>14-673</td>\n",
       "      <td>specialized tissues at the ends of plant stems...</td>\n",
       "      <td>(A) boots (B) people (C) animals (D) flora</td>\n",
       "      <td>specialized tissues at the ends of plant stems...</td>\n",
       "      <td>D</td>\n",
       "      <td>boots</td>\n",
       "      <td>people</td>\n",
       "      <td>animals</td>\n",
       "      <td>flora</td>\n",
       "      <td>flora</td>\n",
       "      <td>C</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4456</th>\n",
       "      <td>14-674</td>\n",
       "      <td>Which most likely requires the aid of speciali...</td>\n",
       "      <td>(A) human assisted space travel (B) the quick ...</td>\n",
       "      <td>Which most likely requires the aid of speciali...</td>\n",
       "      <td>B</td>\n",
       "      <td>human assisted space travel</td>\n",
       "      <td>the quick growth of bamboo</td>\n",
       "      <td>cats digging with their claws</td>\n",
       "      <td>the slow growth of mammals</td>\n",
       "      <td>the quick growth of bamboo</td>\n",
       "      <td>A</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4457</th>\n",
       "      <td>14-676</td>\n",
       "      <td>a Punnett square is used to identify the perce...</td>\n",
       "      <td>(A) home (B) younglings (C) animals (D) pets</td>\n",
       "      <td>a Punnett square is used to identify the perce...</td>\n",
       "      <td>B</td>\n",
       "      <td>home</td>\n",
       "      <td>younglings</td>\n",
       "      <td>animals</td>\n",
       "      <td>pets</td>\n",
       "      <td>younglings</td>\n",
       "      <td>C</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4458</th>\n",
       "      <td>14-677</td>\n",
       "      <td>The dust bowl happened because of</td>\n",
       "      <td>(A) two football teams (B) bowling pins (C) sw...</td>\n",
       "      <td>The dust bowl happened because of (A) two foot...</td>\n",
       "      <td>D</td>\n",
       "      <td>two football teams</td>\n",
       "      <td>bowling pins</td>\n",
       "      <td>sweeping reform</td>\n",
       "      <td>unsustainable farming practice</td>\n",
       "      <td>unsustainable farming practice</td>\n",
       "      <td>A</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4459</th>\n",
       "      <td>14-681</td>\n",
       "      <td>Max mixed some orange flavored powder in a gla...</td>\n",
       "      <td>(A) cola (B) dark chocolate milk (C) a citrus ...</td>\n",
       "      <td>Max mixed some orange flavored powder in a gla...</td>\n",
       "      <td>C</td>\n",
       "      <td>cola</td>\n",
       "      <td>dark chocolate milk</td>\n",
       "      <td>a citrus fruit</td>\n",
       "      <td>lemons and apples</td>\n",
       "      <td>a citrus fruit</td>\n",
       "      <td>B</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4460</th>\n",
       "      <td>14-682</td>\n",
       "      <td>A dog can breathe</td>\n",
       "      <td>(A) under an ocean (B) out in space (C) in a c...</td>\n",
       "      <td>A dog can breathe (A) under an ocean (B) out i...</td>\n",
       "      <td>C</td>\n",
       "      <td>under an ocean</td>\n",
       "      <td>out in space</td>\n",
       "      <td>in a church</td>\n",
       "      <td>in a sarcophagus</td>\n",
       "      <td>in a church</td>\n",
       "      <td>C</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4461</th>\n",
       "      <td>14-683</td>\n",
       "      <td>Animals are just like humans in that if they r...</td>\n",
       "      <td>(A) They will perish (B) they will type. (C) t...</td>\n",
       "      <td>Animals are just like humans in that if they r...</td>\n",
       "      <td>A</td>\n",
       "      <td>They will perish</td>\n",
       "      <td>they will type.</td>\n",
       "      <td>they will program</td>\n",
       "      <td>they will Laugh</td>\n",
       "      <td>They will perish</td>\n",
       "      <td>D</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4462</th>\n",
       "      <td>14-684</td>\n",
       "      <td>Which is the likeliest part of a pencil to hav...</td>\n",
       "      <td>(A) the wood (B) the ink (C) the rubber (D) th...</td>\n",
       "      <td>Which is the likeliest part of a pencil to hav...</td>\n",
       "      <td>D</td>\n",
       "      <td>the wood</td>\n",
       "      <td>the ink</td>\n",
       "      <td>the rubber</td>\n",
       "      <td>the lead</td>\n",
       "      <td>the lead</td>\n",
       "      <td>B</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4463</th>\n",
       "      <td>14-685</td>\n",
       "      <td>When writing with an instrument one sharpens, ...</td>\n",
       "      <td>(A) a squid (B) glowing (C) a mineral (D) brig...</td>\n",
       "      <td>When writing with an instrument one sharpens, ...</td>\n",
       "      <td>C</td>\n",
       "      <td>a squid</td>\n",
       "      <td>glowing</td>\n",
       "      <td>a mineral</td>\n",
       "      <td>bright white</td>\n",
       "      <td>a mineral</td>\n",
       "      <td>C</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4464</th>\n",
       "      <td>14-686</td>\n",
       "      <td>A seismograph is a kind of tool for measuring ...</td>\n",
       "      <td>(A) rivers (B) home planet tremors (C) volcani...</td>\n",
       "      <td>A seismograph is a kind of tool for measuring ...</td>\n",
       "      <td>B</td>\n",
       "      <td>rivers</td>\n",
       "      <td>home planet tremors</td>\n",
       "      <td>volcanic activity</td>\n",
       "      <td>lakes</td>\n",
       "      <td>home planet tremors</td>\n",
       "      <td>B</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4465</th>\n",
       "      <td>14-687</td>\n",
       "      <td>What property of waves matter when the ground ...</td>\n",
       "      <td>(A) amplitude (B) fish scales (C) oscillation ...</td>\n",
       "      <td>What property of waves matter when the ground ...</td>\n",
       "      <td>A</td>\n",
       "      <td>amplitude</td>\n",
       "      <td>fish scales</td>\n",
       "      <td>oscillation</td>\n",
       "      <td>ground water</td>\n",
       "      <td>amplitude</td>\n",
       "      <td>A</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4466</th>\n",
       "      <td>14-689</td>\n",
       "      <td>Who stalks dragonflies south?</td>\n",
       "      <td>(A) clouds (B) electricity (C) kestrels (D) po...</td>\n",
       "      <td>Who stalks dragonflies south? (A) clouds (B) e...</td>\n",
       "      <td>C</td>\n",
       "      <td>clouds</td>\n",
       "      <td>electricity</td>\n",
       "      <td>kestrels</td>\n",
       "      <td>polar bears</td>\n",
       "      <td>kestrels</td>\n",
       "      <td>B</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4467</th>\n",
       "      <td>14-691</td>\n",
       "      <td>What is changed when an object is painted?</td>\n",
       "      <td>(A) chemical makeup (B) light absorption (C) s...</td>\n",
       "      <td>What is changed when an object is painted? (A)...</td>\n",
       "      <td>B</td>\n",
       "      <td>chemical makeup</td>\n",
       "      <td>light absorption</td>\n",
       "      <td>sunshine</td>\n",
       "      <td>rust</td>\n",
       "      <td>light absorption</td>\n",
       "      <td>C</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4468</th>\n",
       "      <td>14-692</td>\n",
       "      <td>Marble is unable to form while lava</td>\n",
       "      <td>(A) is still hot (B) hardens (C) cools down (D...</td>\n",
       "      <td>Marble is unable to form while lava (A) is sti...</td>\n",
       "      <td>A</td>\n",
       "      <td>is still hot</td>\n",
       "      <td>hardens</td>\n",
       "      <td>cools down</td>\n",
       "      <td>offends the volcano</td>\n",
       "      <td>is still hot</td>\n",
       "      <td>D</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4469</th>\n",
       "      <td>14-693</td>\n",
       "      <td>You are hiking up the side of an inactive volc...</td>\n",
       "      <td>(A) Mercury (B) Diamonds (C) Minerals (D) Gold</td>\n",
       "      <td>You are hiking up the side of an inactive volc...</td>\n",
       "      <td>C</td>\n",
       "      <td>Mercury</td>\n",
       "      <td>Diamonds</td>\n",
       "      <td>Minerals</td>\n",
       "      <td>Gold</td>\n",
       "      <td>Minerals</td>\n",
       "      <td>A</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4470</th>\n",
       "      <td>14-694</td>\n",
       "      <td>Which substance is likely present for the birt...</td>\n",
       "      <td>(A) venom (B) peat (C) magma (D) sunlight</td>\n",
       "      <td>Which substance is likely present for the birt...</td>\n",
       "      <td>C</td>\n",
       "      <td>venom</td>\n",
       "      <td>peat</td>\n",
       "      <td>magma</td>\n",
       "      <td>sunlight</td>\n",
       "      <td>magma</td>\n",
       "      <td>B</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4471</th>\n",
       "      <td>14-695</td>\n",
       "      <td>Volcanoes, have a</td>\n",
       "      <td>(A) Esophagus (B) Camels (C) Emotional experie...</td>\n",
       "      <td>Volcanoes, have a (A) Esophagus (B) Camels (C)...</td>\n",
       "      <td>A</td>\n",
       "      <td>Esophagus</td>\n",
       "      <td>Camels</td>\n",
       "      <td>Emotional experience</td>\n",
       "      <td>Cigarettes</td>\n",
       "      <td>Esophagus</td>\n",
       "      <td>D</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4472</th>\n",
       "      <td>14-699</td>\n",
       "      <td>Above your head, the daylight sky is gloomy an...</td>\n",
       "      <td>(A) a bikini (B) a picnic lunch (C) garden she...</td>\n",
       "      <td>Above your head, the daylight sky is gloomy an...</td>\n",
       "      <td>D</td>\n",
       "      <td>a bikini</td>\n",
       "      <td>a picnic lunch</td>\n",
       "      <td>garden shears</td>\n",
       "      <td>an umbrella</td>\n",
       "      <td>an umbrella</td>\n",
       "      <td>A</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4473</th>\n",
       "      <td>14-700</td>\n",
       "      <td>Visibility while driving is decreased by bad w...</td>\n",
       "      <td>(A) local forecasts (B) roads (C) music (D) trees</td>\n",
       "      <td>Visibility while driving is decreased by bad w...</td>\n",
       "      <td>A</td>\n",
       "      <td>local forecasts</td>\n",
       "      <td>roads</td>\n",
       "      <td>music</td>\n",
       "      <td>trees</td>\n",
       "      <td>local forecasts</td>\n",
       "      <td>B</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4474</th>\n",
       "      <td>14-701</td>\n",
       "      <td>Driving during a snow storm or strong thunders...</td>\n",
       "      <td>(A) harder to see (B) Easier to see (C) able t...</td>\n",
       "      <td>Driving during a snow storm or strong thunders...</td>\n",
       "      <td>A</td>\n",
       "      <td>harder to see</td>\n",
       "      <td>Easier to see</td>\n",
       "      <td>able to fly</td>\n",
       "      <td>glasses</td>\n",
       "      <td>harder to see</td>\n",
       "      <td>C</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4475</th>\n",
       "      <td>14-702</td>\n",
       "      <td>_____ tide is a stage in the tide cycle proces...</td>\n",
       "      <td>(A) low (B) mud (C) high (D) earthy</td>\n",
       "      <td>_____ tide is a stage in the tide cycle proces...</td>\n",
       "      <td>C</td>\n",
       "      <td>low</td>\n",
       "      <td>mud</td>\n",
       "      <td>high</td>\n",
       "      <td>earthy</td>\n",
       "      <td>high</td>\n",
       "      <td>C</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4476</th>\n",
       "      <td>14-704</td>\n",
       "      <td>As the hours tick passed noon and the day warm...</td>\n",
       "      <td>(A) match the outdoors (B) chill down (C) be c...</td>\n",
       "      <td>As the hours tick passed noon and the day warm...</td>\n",
       "      <td>A</td>\n",
       "      <td>match the outdoors</td>\n",
       "      <td>chill down</td>\n",
       "      <td>be cool</td>\n",
       "      <td>lower temperature</td>\n",
       "      <td>match the outdoors</td>\n",
       "      <td>D</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4477</th>\n",
       "      <td>14-705</td>\n",
       "      <td>Without this a sword would be made of wood</td>\n",
       "      <td>(A) safety regulations (B) ice (C) ore (D) che...</td>\n",
       "      <td>Without this a sword would be made of wood (A)...</td>\n",
       "      <td>C</td>\n",
       "      <td>safety regulations</td>\n",
       "      <td>ice</td>\n",
       "      <td>ore</td>\n",
       "      <td>cheap labor</td>\n",
       "      <td>ore</td>\n",
       "      <td>D</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4478</th>\n",
       "      <td>14-707</td>\n",
       "      <td>Which makes a sound when you interact with it?</td>\n",
       "      <td>(A) something invented by Pasteur (B) somethin...</td>\n",
       "      <td>Which makes a sound when you interact with it?...</td>\n",
       "      <td>C</td>\n",
       "      <td>something invented by Pasteur</td>\n",
       "      <td>something invented by van Leeuwenhoek</td>\n",
       "      <td>something invented by Sax</td>\n",
       "      <td>something invented by Galileo</td>\n",
       "      <td>something invented by Sax</td>\n",
       "      <td>D</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          ID                                      Question Stem  \\\n",
       "4439  14-647                                      Which is true   \n",
       "4440  14-649     What time are you most likely to see a rainbow   \n",
       "4441  14-651            Why might a polar bear grow white hair?   \n",
       "4442  14-655  As the earth glides from winter into spring yo...   \n",
       "4443  14-656  Which of the following actions can make electr...   \n",
       "4444  14-658           Which is an electrical energy conductor?   \n",
       "4445  14-659  This object in your pocket is the most likely ...   \n",
       "4446  14-661             If I wanted to build endurance I could   \n",
       "4447  14-664          An example of combining two substances is   \n",
       "4448  14-665  A plant needing to photosynthesize will best b...   \n",
       "4449  14-667  If a being is moving regularly, then that bein...   \n",
       "4450  14-668  What kind of impact does exercise have on a hu...   \n",
       "4451  14-669  Animals help with seed dispersal due to the se...   \n",
       "4452  14-670  f a plant shows up in a new place, what might ...   \n",
       "4453  14-671  A creature that is a biped and which regularly...   \n",
       "4454  14-672                         Humans sometimes eat what?   \n",
       "4455  14-673  specialized tissues at the ends of plant stems...   \n",
       "4456  14-674  Which most likely requires the aid of speciali...   \n",
       "4457  14-676  a Punnett square is used to identify the perce...   \n",
       "4458  14-677                  The dust bowl happened because of   \n",
       "4459  14-681  Max mixed some orange flavored powder in a gla...   \n",
       "4460  14-682                                  A dog can breathe   \n",
       "4461  14-683  Animals are just like humans in that if they r...   \n",
       "4462  14-684  Which is the likeliest part of a pencil to hav...   \n",
       "4463  14-685  When writing with an instrument one sharpens, ...   \n",
       "4464  14-686  A seismograph is a kind of tool for measuring ...   \n",
       "4465  14-687  What property of waves matter when the ground ...   \n",
       "4466  14-689                      Who stalks dragonflies south?   \n",
       "4467  14-691         What is changed when an object is painted?   \n",
       "4468  14-692                Marble is unable to form while lava   \n",
       "4469  14-693  You are hiking up the side of an inactive volc...   \n",
       "4470  14-694  Which substance is likely present for the birt...   \n",
       "4471  14-695                                  Volcanoes, have a   \n",
       "4472  14-699  Above your head, the daylight sky is gloomy an...   \n",
       "4473  14-700  Visibility while driving is decreased by bad w...   \n",
       "4474  14-701  Driving during a snow storm or strong thunders...   \n",
       "4475  14-702  _____ tide is a stage in the tide cycle proces...   \n",
       "4476  14-704  As the hours tick passed noon and the day warm...   \n",
       "4477  14-705         Without this a sword would be made of wood   \n",
       "4478  14-707     Which makes a sound when you interact with it?   \n",
       "\n",
       "                                                Choices  \\\n",
       "4439  (A) an anemometer would measure a tropic front...   \n",
       "4440  (A) 10:00 A.M. (B) 9:00 P.M. (C) 3 A.M. (D) Mi...   \n",
       "4441  (A) blend in (B) stand out (C) random (D) look...   \n",
       "4442  (A) leaves growing (B) blizzards (C) bears sta...   \n",
       "4443  (A) Clap on! (B) Jumping in place (C) Turning ...   \n",
       "4444   (A) cobalt (B) plastic (C) concrete (D) mahogany   \n",
       "4445  (A) your phone (B) your credit card (C) your l...   \n",
       "4446  (A) cycling to Publix (B) eat more food (C) sl...   \n",
       "4447  (A) pouring queso into guacamole (B) setting a...   \n",
       "4448  (A) in a roofless room (B) in a cardboard box ...   \n",
       "4449  (A) hardly healthy (B) probably strong (C) abo...   \n",
       "4450      (A) terrible (B) low (C) skyrocketing (D) bad   \n",
       "4451  (A) expelled via excrement (B) smiled (C) expe...   \n",
       "4452  (A) the plant is able to relocate itself (B) i...   \n",
       "4453  (A) kettle shards (B) old feces (C) chia seeds...   \n",
       "4454      (A) air (B) kernels (C) ocean water (D) rocks   \n",
       "4455         (A) boots (B) people (C) animals (D) flora   \n",
       "4456  (A) human assisted space travel (B) the quick ...   \n",
       "4457       (A) home (B) younglings (C) animals (D) pets   \n",
       "4458  (A) two football teams (B) bowling pins (C) sw...   \n",
       "4459  (A) cola (B) dark chocolate milk (C) a citrus ...   \n",
       "4460  (A) under an ocean (B) out in space (C) in a c...   \n",
       "4461  (A) They will perish (B) they will type. (C) t...   \n",
       "4462  (A) the wood (B) the ink (C) the rubber (D) th...   \n",
       "4463  (A) a squid (B) glowing (C) a mineral (D) brig...   \n",
       "4464  (A) rivers (B) home planet tremors (C) volcani...   \n",
       "4465  (A) amplitude (B) fish scales (C) oscillation ...   \n",
       "4466  (A) clouds (B) electricity (C) kestrels (D) po...   \n",
       "4467  (A) chemical makeup (B) light absorption (C) s...   \n",
       "4468  (A) is still hot (B) hardens (C) cools down (D...   \n",
       "4469     (A) Mercury (B) Diamonds (C) Minerals (D) Gold   \n",
       "4470          (A) venom (B) peat (C) magma (D) sunlight   \n",
       "4471  (A) Esophagus (B) Camels (C) Emotional experie...   \n",
       "4472  (A) a bikini (B) a picnic lunch (C) garden she...   \n",
       "4473  (A) local forecasts (B) roads (C) music (D) trees   \n",
       "4474  (A) harder to see (B) Easier to see (C) able t...   \n",
       "4475                (A) low (B) mud (C) high (D) earthy   \n",
       "4476  (A) match the outdoors (B) chill down (C) be c...   \n",
       "4477  (A) safety regulations (B) ice (C) ore (D) che...   \n",
       "4478  (A) something invented by Pasteur (B) somethin...   \n",
       "\n",
       "                                      Complete Question Answer Key  \\\n",
       "4439  Which is true (A) an anemometer would measure ...          A   \n",
       "4440  What time are you most likely to see a rainbow...          A   \n",
       "4441  Why might a polar bear grow white hair? (A) bl...          A   \n",
       "4442  As the earth glides from winter into spring yo...          A   \n",
       "4443  Which of the following actions can make electr...          A   \n",
       "4444  Which is an electrical energy conductor? (A) c...          A   \n",
       "4445  This object in your pocket is the most likely ...          A   \n",
       "4446  If I wanted to build endurance I could (A) cyc...          A   \n",
       "4447  An example of combining two substances is (A) ...          A   \n",
       "4448  A plant needing to photosynthesize will best b...          A   \n",
       "4449  If a being is moving regularly, then that bein...          B   \n",
       "4450  What kind of impact does exercise have on a hu...          C   \n",
       "4451  Animals help with seed dispersal due to the se...          A   \n",
       "4452  f a plant shows up in a new place, what might ...          C   \n",
       "4453  A creature that is a biped and which regularly...          C   \n",
       "4454  Humans sometimes eat what? (A) air (B) kernels...          B   \n",
       "4455  specialized tissues at the ends of plant stems...          D   \n",
       "4456  Which most likely requires the aid of speciali...          B   \n",
       "4457  a Punnett square is used to identify the perce...          B   \n",
       "4458  The dust bowl happened because of (A) two foot...          D   \n",
       "4459  Max mixed some orange flavored powder in a gla...          C   \n",
       "4460  A dog can breathe (A) under an ocean (B) out i...          C   \n",
       "4461  Animals are just like humans in that if they r...          A   \n",
       "4462  Which is the likeliest part of a pencil to hav...          D   \n",
       "4463  When writing with an instrument one sharpens, ...          C   \n",
       "4464  A seismograph is a kind of tool for measuring ...          B   \n",
       "4465  What property of waves matter when the ground ...          A   \n",
       "4466  Who stalks dragonflies south? (A) clouds (B) e...          C   \n",
       "4467  What is changed when an object is painted? (A)...          B   \n",
       "4468  Marble is unable to form while lava (A) is sti...          A   \n",
       "4469  You are hiking up the side of an inactive volc...          C   \n",
       "4470  Which substance is likely present for the birt...          C   \n",
       "4471  Volcanoes, have a (A) Esophagus (B) Camels (C)...          A   \n",
       "4472  Above your head, the daylight sky is gloomy an...          D   \n",
       "4473  Visibility while driving is decreased by bad w...          A   \n",
       "4474  Driving during a snow storm or strong thunders...          A   \n",
       "4475  _____ tide is a stage in the tide cycle proces...          C   \n",
       "4476  As the hours tick passed noon and the day warm...          A   \n",
       "4477  Without this a sword would be made of wood (A)...          C   \n",
       "4478  Which makes a sound when you interact with it?...          C   \n",
       "\n",
       "                                           OptionA  \\\n",
       "4439   an anemometer would measure a tropic front    \n",
       "4440                                   10:00 A.M.    \n",
       "4441                                     blend in    \n",
       "4442                               leaves growing    \n",
       "4443                                     Clap on!    \n",
       "4444                                       cobalt    \n",
       "4445                                   your phone    \n",
       "4446                            cycling to Publix    \n",
       "4447                 pouring queso into guacamole    \n",
       "4448                           in a roofless room    \n",
       "4449                               hardly healthy    \n",
       "4450                                     terrible    \n",
       "4451                       expelled via excrement    \n",
       "4452         the plant is able to relocate itself    \n",
       "4453                                kettle shards    \n",
       "4454                                          air    \n",
       "4455                                        boots    \n",
       "4456                  human assisted space travel    \n",
       "4457                                         home    \n",
       "4458                           two football teams    \n",
       "4459                                         cola    \n",
       "4460                               under an ocean    \n",
       "4461                             They will perish    \n",
       "4462                                     the wood    \n",
       "4463                                      a squid    \n",
       "4464                                       rivers    \n",
       "4465                                    amplitude    \n",
       "4466                                       clouds    \n",
       "4467                              chemical makeup    \n",
       "4468                                 is still hot    \n",
       "4469                                      Mercury    \n",
       "4470                                        venom    \n",
       "4471                                    Esophagus    \n",
       "4472                                     a bikini    \n",
       "4473                              local forecasts    \n",
       "4474                                harder to see    \n",
       "4475                                          low    \n",
       "4476                           match the outdoors    \n",
       "4477                           safety regulations    \n",
       "4478                something invented by Pasteur    \n",
       "\n",
       "                                      OptionB  \\\n",
       "4439     an anemometer would measure thunder    \n",
       "4440                               9:00 P.M.    \n",
       "4441                               stand out    \n",
       "4442                               blizzards    \n",
       "4443                        Jumping in place    \n",
       "4444                                 plastic    \n",
       "4445                        your credit card    \n",
       "4446                           eat more food    \n",
       "4447              setting a laptop on a desk    \n",
       "4448                      in a cardboard box    \n",
       "4449                         probably strong    \n",
       "4450                                     low    \n",
       "4451                                  smiled    \n",
       "4452         it spontaneously appeared there    \n",
       "4453                               old feces    \n",
       "4454                                 kernels    \n",
       "4455                                  people    \n",
       "4456              the quick growth of bamboo    \n",
       "4457                              younglings    \n",
       "4458                            bowling pins    \n",
       "4459                     dark chocolate milk    \n",
       "4460                            out in space    \n",
       "4461                         they will type.    \n",
       "4462                                 the ink    \n",
       "4463                                 glowing    \n",
       "4464                     home planet tremors    \n",
       "4465                             fish scales    \n",
       "4466                             electricity    \n",
       "4467                        light absorption    \n",
       "4468                                 hardens    \n",
       "4469                                Diamonds    \n",
       "4470                                    peat    \n",
       "4471                                  Camels    \n",
       "4472                          a picnic lunch    \n",
       "4473                                   roads    \n",
       "4474                           Easier to see    \n",
       "4475                                     mud    \n",
       "4476                              chill down    \n",
       "4477                                     ice    \n",
       "4478   something invented by van Leeuwenhoek    \n",
       "\n",
       "                                      OptionC  \\\n",
       "4439   an anemometer would measure a rainbow    \n",
       "4440                                  3 A.M.    \n",
       "4441                                  random    \n",
       "4442              bears starting hibernation    \n",
       "4443                      Turning a doorknob    \n",
       "4444                                concrete    \n",
       "4445                       your library card    \n",
       "4446                           sleep all day    \n",
       "4447              building a roof on a house    \n",
       "4448                    in a windowless room    \n",
       "4449                            about to die    \n",
       "4450                            skyrocketing    \n",
       "4451                    expelled from school    \n",
       "4452    its cased embryo have been dispersed    \n",
       "4453                              chia seeds    \n",
       "4454                             ocean water    \n",
       "4455                                 animals    \n",
       "4456           cats digging with their claws    \n",
       "4457                                 animals    \n",
       "4458                         sweeping reform    \n",
       "4459                          a citrus fruit    \n",
       "4460                             in a church    \n",
       "4461                       they will program    \n",
       "4462                              the rubber    \n",
       "4463                               a mineral    \n",
       "4464                       volcanic activity    \n",
       "4465                             oscillation    \n",
       "4466                                kestrels    \n",
       "4467                                sunshine    \n",
       "4468                              cools down    \n",
       "4469                                Minerals    \n",
       "4470                                   magma    \n",
       "4471                    Emotional experience    \n",
       "4472                           garden shears    \n",
       "4473                                   music    \n",
       "4474                             able to fly    \n",
       "4475                                    high    \n",
       "4476                                 be cool    \n",
       "4477                                     ore    \n",
       "4478               something invented by Sax    \n",
       "\n",
       "                                            OptionD  \\\n",
       "4439   an anemometer would measure a lightning bolt   \n",
       "4440                                       Midnight   \n",
       "4441                                     look fancy   \n",
       "4442                          leaves turning orange   \n",
       "4443                              Rubbing your nose   \n",
       "4444                                       mahogany   \n",
       "4445                                    your wallet   \n",
       "4446                                 only eat pizza   \n",
       "4447                  putting clothes in a suitcase   \n",
       "4448                   in a car with tinted windows   \n",
       "4449                           making little motion   \n",
       "4450                                            bad   \n",
       "4451                                         kicked   \n",
       "4452                     it walked to the new place   \n",
       "4453                                     hard rocks   \n",
       "4454                                          rocks   \n",
       "4455                                          flora   \n",
       "4456                     the slow growth of mammals   \n",
       "4457                                           pets   \n",
       "4458                 unsustainable farming practice   \n",
       "4459                              lemons and apples   \n",
       "4460                               in a sarcophagus   \n",
       "4461                                they will Laugh   \n",
       "4462                                       the lead   \n",
       "4463                                   bright white   \n",
       "4464                                          lakes   \n",
       "4465                                   ground water   \n",
       "4466                                    polar bears   \n",
       "4467                                           rust   \n",
       "4468                            offends the volcano   \n",
       "4469                                           Gold   \n",
       "4470                                       sunlight   \n",
       "4471                                     Cigarettes   \n",
       "4472                                    an umbrella   \n",
       "4473                                          trees   \n",
       "4474                                        glasses   \n",
       "4475                                         earthy   \n",
       "4476                              lower temperature   \n",
       "4477                                    cheap labor   \n",
       "4478                  something invented by Galileo   \n",
       "\n",
       "                                            Result predicted_answers  \n",
       "4439   an anemometer would measure a tropic front                  A  \n",
       "4440                                   10:00 A.M.                  C  \n",
       "4441                                     blend in                  D  \n",
       "4442                               leaves growing                  D  \n",
       "4443                                     Clap on!                  B  \n",
       "4444                                       cobalt                  B  \n",
       "4445                                   your phone                  B  \n",
       "4446                            cycling to Publix                  D  \n",
       "4447                 pouring queso into guacamole                  C  \n",
       "4448                           in a roofless room                  D  \n",
       "4449                              probably strong                  C  \n",
       "4450                                 skyrocketing                  D  \n",
       "4451                       expelled via excrement                  C  \n",
       "4452         its cased embryo have been dispersed                  D  \n",
       "4453                                   chia seeds                  D  \n",
       "4454                                      kernels                  A  \n",
       "4455                                         flora                 C  \n",
       "4456                   the quick growth of bamboo                  A  \n",
       "4457                                   younglings                  C  \n",
       "4458                unsustainable farming practice                 A  \n",
       "4459                               a citrus fruit                  B  \n",
       "4460                                  in a church                  C  \n",
       "4461                             They will perish                  D  \n",
       "4462                                      the lead                 B  \n",
       "4463                                    a mineral                  C  \n",
       "4464                          home planet tremors                  B  \n",
       "4465                                    amplitude                  A  \n",
       "4466                                     kestrels                  B  \n",
       "4467                             light absorption                  C  \n",
       "4468                                 is still hot                  D  \n",
       "4469                                     Minerals                  A  \n",
       "4470                                        magma                  B  \n",
       "4471                                    Esophagus                  D  \n",
       "4472                                   an umbrella                 A  \n",
       "4473                              local forecasts                  B  \n",
       "4474                                harder to see                  C  \n",
       "4475                                         high                  C  \n",
       "4476                           match the outdoors                  D  \n",
       "4477                                          ore                  D  \n",
       "4478                    something invented by Sax                  D  "
      ]
     },
     "execution_count": 223,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Validation_train = train[4439:]\n",
    "Validation_train['predicted_answers'] = flat_list_true_col\n",
    "Validation_train.head(40)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 224,
   "metadata": {},
   "outputs": [],
   "source": [
    "def percentage(part, whole):\n",
    "    return 100 * float(part)/float(whole)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 225,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model1 accuracy rate is 23 percent \n"
     ]
    }
   ],
   "source": [
    "predict_final = Validation_train['Answer Key'] == Validation_train['predicted_answers']\n",
    "cntr = 0    \n",
    "for i in predict_final:\n",
    "    if i == True:\n",
    "        cntr = cntr+1\n",
    "\n",
    "\n",
    "print('Model1 accuracy rate is %d percent ' % (percentage(cntr,len(Validation_train))))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "conda_tensorflow_p36",
   "language": "python",
   "name": "conda_tensorflow_p36"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}