tf_siamese-wtwt.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "b1ade922-da4f-4107-9467-95a1a972a694",
   "metadata": {},
   "source": [
    "## https://www.kaggle.com/sauravjoshi23/text-classification-using-siamesenet-glove\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9427d2ba-1815-4f9d-8f9a-1a95b468332e",
   "metadata": {},
   "source": [
    "# Check every comment in this notebook and understand it. You will have to uncomment according to need."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 142,
   "id": "1a426512",
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import random\n",
    "import warnings\n",
    "import time\n",
    "import datetime\n",
    "import re\n",
    "import string\n",
    "import itertools\n",
    "import pickle\n",
    "import joblib\n",
    "import nltk\n",
    "import csv\n",
    "from nltk.corpus import stopwords, wordnet\n",
    "stop = set(stopwords.words('english'))\n",
    "from wordcloud import WordCloud, STOPWORDS\n",
    "from nltk.stem import WordNetLemmatizer\n",
    "from nltk.tokenize import word_tokenize\n",
    "from nltk.probability import FreqDist\n",
    "from collections import Counter, defaultdict\n",
    "from keras.utils import np_utils\n",
    "\n",
    "import keras.backend as K\n",
    "import tensorflow as tf\n",
    "from tensorflow import keras\n",
    "from sklearn.metrics import classification_report\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import roc_auc_score\n",
    "\n",
    "from tensorflow.keras.preprocessing.sequence import pad_sequences\n",
    "from tensorflow.keras.preprocessing.text import Tokenizer\n",
    "from tensorflow.keras.regularizers import l2\n",
    "from tensorflow.keras.models import Sequential\n",
    "from tensorflow.keras.optimizers import Adam\n",
    "from tensorflow.keras.layers import Conv2D, ZeroPadding2D, Activation, Input, concatenate,Embedding, LSTM\n",
    "from tensorflow.keras.models import Model\n",
    "from tensorflow.keras.layers import BatchNormalization\n",
    "from tensorflow.keras.layers import MaxPooling2D\n",
    "from tensorflow.keras.layers import Concatenate\n",
    "from tensorflow.keras.layers import Lambda, Flatten, Dense\n",
    "from tensorflow.keras.initializers import glorot_uniform\n",
    "from tensorflow.keras.layers import Input, Dense, Flatten, GlobalMaxPool2D, GlobalAvgPool2D, Concatenate, Multiply, Dropout, Subtract, Add, Conv2D\n",
    "\n",
    "from sklearn.manifold import TSNE\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "%matplotlib inline\n",
    "\n",
    "import warnings\n",
    "warnings.filterwarnings('ignore')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 143,
   "id": "260304ff",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_data_file_m = '/data/parush/stance_mohammed/new_train.txt'\n",
    "test_data_file_m = '/data/parush/stance_mohammed/new_test.txt'\n",
    "TARGETS_m = [ 'Atheism','Climate Change is a Real Concern', 'Feminist Movement','Hillary Clinton', 'Legalization of Abortion', 'Donald Trump']\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "fafa254a",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 144,
   "id": "585c2fb8",
   "metadata": {},
   "outputs": [],
   "source": [
    "def df_gen_mohammed(train_data_file_m,pos_target_label,neg_target_label ):\n",
    "    \n",
    "    df = pd.DataFrame(columns=['text', 'target'])\n",
    "    with open(train_data_file_m, 'r') as trainfile:\n",
    "        for line in trainfile:\n",
    "            line = line.replace('#SemST', '').strip()\n",
    "            line = line.split('\\t')\n",
    "            if line[0].strip() != 'ID' and \"Donald Trump\" in line[1].strip():\n",
    "                tweet = line[2]\n",
    "                df.loc[len(df)] = [tweet,pos_target_label]\n",
    "            elif line[0].strip() != 'ID' and 'Hillary Clinton' in line[1].strip():\n",
    "                tweet = line[2]\n",
    "                df.loc[len(df)] = [tweet,neg_target_label]\n",
    "    df = df.sample(frac = 1)\n",
    "    return df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "538dfb8b-f9c6-4f51-b87e-95b42146bb76",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 145,
   "id": "8a3afee1-9ac4-476e-8696-54552a7f5ea9",
   "metadata": {},
   "outputs": [],
   "source": [
    "df_politics = df_gen_mohammed(train_data_file_m,0,0 )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 146,
   "id": "c974c139-281e-4b14-8412-5ecd77412268",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1219"
      ]
     },
     "execution_count": 146,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(df_politics)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 147,
   "id": "c729a273-06df-40cc-95eb-969ec3c6b6de",
   "metadata": {},
   "outputs": [],
   "source": [
    "df_hlt_train = pd.read_csv(\"/data/parush/wtwt/healthcare_train.txt\", sep='\\t')\n",
    "df_hlt_test = pd.read_csv(\"/data/parush/wtwt/healthcare_test.txt\", sep='\\t')\n",
    "df_ent_train = pd.read_csv(\"/data/parush/wtwt/entertainment_train.txt\", sep='\\t')\n",
    "df_ent_test = pd.read_csv(\"/data/parush/wtwt/entertainment_test.txt\", sep='\\t')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 148,
   "id": "921abe80-1158-41b2-b683-50ff56470105",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array(['AET_HUM', 'ANTM_CI', 'CI_ESRX', 'CVS_AET'], dtype=object)"
      ]
     },
     "execution_count": 148,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.unique(df_hlt_train['target'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 149,
   "id": "b2a4d500",
   "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>text</th>\n",
       "      <th>target</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Essential reading from @jasondashbailey on all...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Foxtel and Fox Sports Australia have plans to ...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>@sabya2km @AnimalDesatado @Keda_92 @ManaByte T...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>All the Disney classics should be on Netflix b...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>@LostMittenCP @CPRewritten @rainbertt Better t...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1195</th>\n",
       "      <td>ALL DEM BET CYPHERS WAS PHONY DEY LET WEIRD NI...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1196</th>\n",
       "      <td>@JoshuaYehl Will the Kitty Pryde Solo Film Be ...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1197</th>\n",
       "      <td>FOX AND DISNEY (and thereby Marvel) IS REAL! M...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1198</th>\n",
       "      <td>It’s everyday bro with that Disney/Fox merger ...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1199</th>\n",
       "      <td>I liked a @YouTube video https://t.co/e0bCUybo...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>1200 rows × 2 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                   text  target\n",
       "0     Essential reading from @jasondashbailey on all...       1\n",
       "1     Foxtel and Fox Sports Australia have plans to ...       1\n",
       "2     @sabya2km @AnimalDesatado @Keda_92 @ManaByte T...       1\n",
       "3     All the Disney classics should be on Netflix b...       1\n",
       "4     @LostMittenCP @CPRewritten @rainbertt Better t...       1\n",
       "...                                                 ...     ...\n",
       "1195  ALL DEM BET CYPHERS WAS PHONY DEY LET WEIRD NI...       1\n",
       "1196  @JoshuaYehl Will the Kitty Pryde Solo Film Be ...       1\n",
       "1197  FOX AND DISNEY (and thereby Marvel) IS REAL! M...       1\n",
       "1198  It’s everyday bro with that Disney/Fox merger ...       1\n",
       "1199  I liked a @YouTube video https://t.co/e0bCUybo...       1\n",
       "\n",
       "[1200 rows x 2 columns]"
      ]
     },
     "execution_count": 149,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_enter = pd.DataFrame()\n",
    "df_a = df_ent_train[df_ent_train['target'] == 'FOXA_DIS'][:1200]\n",
    "df_enter = df_enter.append(df_a, ignore_index=True)\n",
    "\n",
    "df_enter_new = pd.DataFrame(columns=['text', 'target'])\n",
    "df_enter_new['text'] = df_enter['tweet']\n",
    "df_enter_new['target'] = 1\n",
    "df_enter_new"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 150,
   "id": "e2e136cd-d28c-49ec-ac34-d64ae8877acc",
   "metadata": {},
   "outputs": [],
   "source": [
    "# df_health = pd.DataFrame()\n",
    "# df_a = df_hlt_train[df_hlt_train['target'] == 'AET_HUM'][:300]\n",
    "# df_b = df_hlt_train[df_hlt_train['target'] == 'ANTM_CI'][:300]\n",
    "# df_c = df_hlt_train[df_hlt_train['target'] == 'CI_ESRX'][:300]\n",
    "# df_d = df_hlt_train[df_hlt_train['target'] == 'CVS_AET'][:300]\n",
    "# df_health = df_health.append(df_a, ignore_index=True)\n",
    "# df_health = df_health.append(df_b, ignore_index=True)\n",
    "# df_health = df_health.append(df_c, ignore_index=True)\n",
    "# df_health = df_health.append(df_d, ignore_index=True)\n",
    "# df_health_new = pd.DataFrame(columns=['text', 'target'])\n",
    "# df_health_new['text'] = df_health['tweet']\n",
    "# df_health_new['target'] = 1\n",
    "# df_health_new"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 151,
   "id": "8425e495-2ee0-4073-8d85-3d8014dabece",
   "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>text</th>\n",
       "      <th>target</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>It's what's best for business and presidential...</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Another #Hillary supporter committed to caucus...</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>@politico   YAWNzzzzzzzzzzzzzz could keep the ...</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Isis celebrates Islam with 3 terror attacks ov...</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Was asked by @HillaryClinton to help with her ...</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2414</th>\n",
       "      <td>ALL DEM BET CYPHERS WAS PHONY DEY LET WEIRD NI...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2415</th>\n",
       "      <td>@JoshuaYehl Will the Kitty Pryde Solo Film Be ...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2416</th>\n",
       "      <td>FOX AND DISNEY (and thereby Marvel) IS REAL! M...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2417</th>\n",
       "      <td>It’s everyday bro with that Disney/Fox merger ...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2418</th>\n",
       "      <td>I liked a @YouTube video https://t.co/e0bCUybo...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>2419 rows × 2 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                   text target\n",
       "0     It's what's best for business and presidential...      0\n",
       "1     Another #Hillary supporter committed to caucus...      0\n",
       "2     @politico   YAWNzzzzzzzzzzzzzz could keep the ...      0\n",
       "3     Isis celebrates Islam with 3 terror attacks ov...      0\n",
       "4     Was asked by @HillaryClinton to help with her ...      0\n",
       "...                                                 ...    ...\n",
       "2414  ALL DEM BET CYPHERS WAS PHONY DEY LET WEIRD NI...      1\n",
       "2415  @JoshuaYehl Will the Kitty Pryde Solo Film Be ...      1\n",
       "2416  FOX AND DISNEY (and thereby Marvel) IS REAL! M...      1\n",
       "2417  It’s everyday bro with that Disney/Fox merger ...      1\n",
       "2418  I liked a @YouTube video https://t.co/e0bCUybo...      1\n",
       "\n",
       "[2419 rows x 2 columns]"
      ]
     },
     "execution_count": 151,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df = df_politics\n",
    "#df = df.append(df_health_new, ignore_index=True)\n",
    "df = df.append(df_enter_new, ignore_index=True)\n",
    "df"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "38dfe75d",
   "metadata": {},
   "source": [
    "# Data Cleaning"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 152,
   "id": "fe37df96",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Cleaning Data\n",
      "Cleaning Done\n"
     ]
    }
   ],
   "source": [
    "def remove_url(text):\n",
    "    url = re.compile(r'https?://\\S+|www\\.\\S+')\n",
    "    return url.sub(r'', text)\n",
    "\n",
    "def remove_emoji(text):\n",
    "    emoji_pattern = re.compile(\n",
    "        '['\n",
    "        u'\\U0001F600-\\U0001F64F'  # emoticons\n",
    "        u'\\U0001F300-\\U0001F5FF'  # symbols & pictographs\n",
    "        u'\\U0001F680-\\U0001F6FF'  # transport & map symbols\n",
    "        u'\\U0001F1E0-\\U0001F1FF'  # flags (iOS)\n",
    "        u'\\U00002702-\\U000027B0'\n",
    "        u'\\U000024C2-\\U0001F251'\n",
    "        ']+',\n",
    "        flags=re.UNICODE)\n",
    "    return emoji_pattern.sub(r'', text)\n",
    "\n",
    "def remove_html(text):\n",
    "    html = re.compile(r'<.*?>|&([a-z0-9]+|#[0-9]{1,6}|#x[0-9a-f]{1,6});')\n",
    "    return re.sub(html, '', text)\n",
    "\n",
    "def remove_punct(text):\n",
    "    table = str.maketrans('', '', string.punctuation)\n",
    "    return text.translate(table)\n",
    "\n",
    "df['clean_text'] = df['text'].apply(lambda x: remove_url(str(x)))\n",
    "df['clean_text'] = df['clean_text'].apply(lambda x: remove_emoji(str(x)))\n",
    "df['clean_text'] = df['clean_text'].apply(lambda x: remove_html(str(x)))\n",
    "df['clean_text'] = df['clean_text'].apply(lambda x: remove_punct(str(x)))\n",
    "df['clean_text'] = df['clean_text'].apply(word_tokenize)\n",
    "df['clean_text'] = df['clean_text'].apply(lambda x: [word.lower() for word in x])\n",
    "df['clean_text'] = df['clean_text'].apply(lambda x: [word for word in x if word not in stop])\n",
    "df['clean_text'] = df['clean_text'].apply(nltk.tag.pos_tag)\n",
    "\n",
    "def get_wordnet_pos(tag):\n",
    "    if tag.startswith('J'):\n",
    "        return wordnet.ADJ\n",
    "    elif tag.startswith('V'):\n",
    "        return wordnet.VERB\n",
    "    elif tag.startswith('N'):\n",
    "        return wordnet.NOUN\n",
    "    elif tag.startswith('R'):\n",
    "        return wordnet.ADV\n",
    "    else:\n",
    "        return wordnet.NOUN\n",
    "print(\"Cleaning Data\")    \n",
    "df['clean_text'] = df['clean_text'].apply(\n",
    "    lambda x: [(word, get_wordnet_pos(pos_tag)) for (word, pos_tag) in x])\n",
    "wnl = WordNetLemmatizer()\n",
    "df['clean_text'] = df['clean_text'].apply(\n",
    "    lambda x: [wnl.lemmatize(word, tag) for word, tag in x])\n",
    "df['clean_text'] = df['clean_text'].apply(\n",
    "    lambda x: [word for word in x if word not in stop])\n",
    "df['clean_text'] = [' '.join(map(str, l)) for l in df['clean_text']]\n",
    "print(\"Cleaning Done\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e90a6858",
   "metadata": {},
   "source": [
    "# Triple Loss"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 153,
   "id": "4ef15558",
   "metadata": {},
   "outputs": [],
   "source": [
    "positive_initial = list(df[df['target'] == 1]['clean_text'])\n",
    "negative_initial = list(df[df['target'] == 0]['clean_text'])\n",
    "\n",
    "# Taking 1000 samples from the entire data\n",
    "pos = positive_initial[:]\n",
    "neg = negative_initial[:]\n",
    "\n",
    "# Creating pairs of data for siamese training => labels for identity loss and class for disaster or non disaster\n",
    "df2 = pd.DataFrame(columns=['Anchor', 'Positive', 'Negative', 'label', 'class'])\n",
    "\n",
    "for data in pos:\n",
    "    a = data\n",
    "    p = random.choice(pos)\n",
    "    n = random.choice(neg)\n",
    "    df2.loc[len(df2)] = [a, p, n, 1, 1]\n",
    "\n",
    "\n",
    "for data in neg:\n",
    "    a = data\n",
    "    p = random.choice(neg)\n",
    "    n = random.choice(pos)\n",
    "    df2.loc[len(df2)] = [a, p, n, 1, 0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 154,
   "id": "b8acea3a",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The new DF with anchor, positive, negative, label, class is created. The shape is :  (2419, 5)\n"
     ]
    }
   ],
   "source": [
    "print(\"The new DF with anchor, positive, negative, label, class is created. The shape is : \", df2.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "5006ce55",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 155,
   "id": "ba4f373c",
   "metadata": {},
   "outputs": [],
   "source": [
    "X, X_test, y, y_test = train_test_split(df2[['Anchor', 'Positive', 'Negative']], df2[['label', 'class']], test_size=0.2, random_state=0)\n",
    "X_train, X_val, y_train, y_val = train_test_split(X[['Anchor', 'Positive', 'Negative']], y[['label', 'class']], test_size=0.2, random_state=0)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 156,
   "id": "24410b4d",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The X_train = (1548, 3), X_val = (387, 3), X_test = (484, 3), y_train = (1548, 2), y_val = (387, 2), y_test = (484, 2)\n"
     ]
    }
   ],
   "source": [
    "print(\"The X_train = {}, X_val = {}, X_test = {}, y_train = {}, y_val = {}, y_test = {}\".format(X_train.shape, X_val.shape, X_test.shape, y_train.shape, y_val.shape, y_test.shape))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "0497d9ac",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 157,
   "id": "b0b8fe05",
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train['text'] = X_train[['Anchor', 'Positive', 'Negative']].apply(lambda x: str(x[0])+\" \"+str(x[1])+\" \"+str(x[2]), axis=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2b3eadfd",
   "metadata": {},
   "source": [
    "# Tokenizer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 158,
   "id": "4f1f8ee3",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Starting to Tokenize and Padding\n"
     ]
    }
   ],
   "source": [
    "print(\"Starting to Tokenize and Padding\")\n",
    "t = Tokenizer()\n",
    "t.fit_on_texts(X_train['text'].values)\n",
    "\n",
    "X_train['Anchor'] = X_train['Anchor'].astype(str)\n",
    "X_train['Positive'] = X_train['Positive'].astype(str)\n",
    "X_train['Negative'] = X_train['Negative'].astype(str)\n",
    "X_val['Anchor'] = X_val['Anchor'].astype(str)\n",
    "X_val['Positive'] = X_val['Positive'].astype(str)\n",
    "X_val['Negative'] = X_val['Negative'].astype(str)\n",
    "X_test['Anchor'] = X_test['Anchor'].astype(str)\n",
    "X_test['Positive'] = X_test['Positive'].astype(str)\n",
    "X_test['Negative'] = X_test['Negative'].astype(str)\n",
    "\n",
    "train_q1_seq = t.texts_to_sequences(X_train['Anchor'].values)\n",
    "train_q2_seq = t.texts_to_sequences(X_train['Positive'].values)\n",
    "train_q3_seq = t.texts_to_sequences(X_train['Negative'].values)\n",
    "val_q1_seq = t.texts_to_sequences(X_val['Anchor'].values)\n",
    "val_q2_seq = t.texts_to_sequences(X_val['Positive'].values)\n",
    "val_q3_seq = t.texts_to_sequences(X_val['Negative'].values)\n",
    "test_q1_seq = t.texts_to_sequences(X_test['Anchor'].values)\n",
    "test_q2_seq = t.texts_to_sequences(X_test['Positive'].values)\n",
    "test_q3_seq = t.texts_to_sequences(X_test['Negative'].values)\n",
    "\n",
    "max_len = 200\n",
    "train_q1_seq = pad_sequences(train_q1_seq, maxlen=max_len, padding='post')\n",
    "train_q2_seq = pad_sequences(train_q2_seq, maxlen=max_len, padding='post')\n",
    "train_q3_seq = pad_sequences(train_q3_seq, maxlen=max_len, padding='post')\n",
    "val_q1_seq = pad_sequences(val_q1_seq, maxlen=max_len, padding='post')\n",
    "val_q2_seq = pad_sequences(val_q2_seq, maxlen=max_len, padding='post')\n",
    "val_q3_seq = pad_sequences(val_q3_seq, maxlen=max_len, padding='post')\n",
    "test_q1_seq = pad_sequences(test_q1_seq, maxlen=max_len, padding='post')\n",
    "test_q2_seq = pad_sequences(test_q2_seq, maxlen=max_len, padding='post')\n",
    "test_q3_seq = pad_sequences(test_q3_seq, maxlen=max_len, padding='post')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 159,
   "id": "f72b7097",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loading Embedding and making embedding_matrix\n",
      "Found 400000 word vectors.\n"
     ]
    }
   ],
   "source": [
    "print(\"Loading Embedding and making embedding_matrix\")\n",
    "embeddings_index = {}\n",
    "f = open('/data/parush/embeddings/wikipedia/glove.6B.300d.txt')\n",
    "for line in f:\n",
    "    values = line.split()\n",
    "    word = values[0]\n",
    "    coefs = np.asarray(values[1:], dtype='float32')\n",
    "    embeddings_index[word] = coefs\n",
    "f.close()\n",
    "\n",
    "print('Found %s word vectors.' % len(embeddings_index))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "713e99c5-5d28-43f7-9a44-fa3460c3bb73",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 160,
   "id": "d3263e8d",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loaded 400000 word vectors.\n"
     ]
    }
   ],
   "source": [
    "not_present_list = []\n",
    "vocab_size = len(t.word_index) + 1\n",
    "print('Loaded %s word vectors.' % len(embeddings_index))\n",
    "embedding_matrix = np.zeros((vocab_size, len(embeddings_index['no'])))\n",
    "for word, i in t.word_index.items():\n",
    "    if word in embeddings_index.keys():\n",
    "        \n",
    "        embedding_vector = embeddings_index.get(word)\n",
    "    else:\n",
    "        embedding_vector = None\n",
    "        not_present_list.append(word)\n",
    "    if embedding_vector is not None:\n",
    "        embedding_matrix[i] = embedding_vector\n",
    "    else:\n",
    "        embedding_matrix[i] = np.zeros(300)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 161,
   "id": "3a0555a9",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Embedding matrix shape (6344, 300)\n"
     ]
    }
   ],
   "source": [
    "print(\"Embedding matrix shape\",embedding_matrix.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f9d8355f",
   "metadata": {},
   "source": [
    "# Siamese Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 162,
   "id": "d248dcca",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Creating Model and Loss Function\n"
     ]
    }
   ],
   "source": [
    "print(\"Creating Model and Loss Function\")\n",
    "def identity_loss(y_true, y_pred):\n",
    "    return K.mean(y_pred)\n",
    "\n",
    "def triplet_loss(x, alpha = 0.2):\n",
    "    # Triplet Loss function.\n",
    "    anchor,positive,negative = x\n",
    "    # distance between the anchor and the positive\n",
    "    pos_dist = K.sum(K.square(anchor-positive),axis=1)\n",
    "    # distance between the anchor and the negative\n",
    "    neg_dist = K.sum(K.square(anchor-negative),axis=1)\n",
    "    # compute loss\n",
    "    basic_loss = pos_dist-neg_dist+alpha\n",
    "    loss = K.maximum(basic_loss,0.0)\n",
    "    return loss"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 163,
   "id": "293fe586",
   "metadata": {},
   "outputs": [],
   "source": [
    "def embedding_model():\n",
    "    network = Sequential()\n",
    "    network.add(Embedding(name=\"synopsis_embedd\",input_dim =len(t.word_index)+1, \n",
    "                       output_dim=len(embeddings_index['no']),weights=[embedding_matrix], \n",
    "                       input_length=train_q1_seq.shape[1],trainable=False))\n",
    "    network.add(LSTM(64,return_sequences=True, activation=\"relu\",dropout=0.3))\n",
    "    network.add(Flatten())\n",
    "    network.add(Dense(128, activation='relu',\n",
    "                  kernel_regularizer=l2(1e-3),\n",
    "                  kernel_initializer='he_uniform'))\n",
    "    network.add(Dropout(0.5))\n",
    "    network.add(Dense(2, activation=None,\n",
    "                  kernel_regularizer=l2(1e-3),\n",
    "                  kernel_initializer='he_uniform'))\n",
    "    #Force the encoding to live on the d-dimentional hypershpere\n",
    "    # network.add(Lambda(lambda x: K.l2_normalize(x,axis=-1)))\n",
    "    return network\n",
    "\n",
    "\n",
    "def build_network(base_model):\n",
    "    input_1 = Input(shape=(train_q1_seq.shape[1],))\n",
    "    input_2 = Input(shape=(train_q2_seq.shape[1],))\n",
    "    input_3 = Input(shape=(train_q3_seq.shape[1],))\n",
    "    \n",
    "    A = base_model(input_1)\n",
    "    P = base_model(input_2)\n",
    "    N = base_model(input_3)\n",
    "\n",
    "    loss = Lambda(triplet_loss)([A, P, N])\n",
    "    model = Model(inputs = [input_1, input_2, input_3], outputs = loss)\n",
    "    model.compile(loss = identity_loss, optimizer = Adam(0.001))\n",
    "    return model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 164,
   "id": "c56b2f01",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:Layer lstm_2 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.\n",
      "Model: \"model_4\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_9 (InputLayer)            [(None, 200)]        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "input_10 (InputLayer)           [(None, 200)]        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "input_11 (InputLayer)           [(None, 200)]        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "sequential_2 (Sequential)       (None, 2)            3635426     input_9[0][0]                    \n",
      "                                                                 input_10[0][0]                   \n",
      "                                                                 input_11[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "lambda_2 (Lambda)               (None,)              0           sequential_2[0][0]               \n",
      "                                                                 sequential_2[1][0]               \n",
      "                                                                 sequential_2[2][0]               \n",
      "==================================================================================================\n",
      "Total params: 3,635,426\n",
      "Trainable params: 1,732,226\n",
      "Non-trainable params: 1,903,200\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "base_model = embedding_model()\n",
    "model = build_network(base_model)\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 165,
   "id": "9988ba7f",
   "metadata": {},
   "outputs": [],
   "source": [
    "y_train_label = np.asarray(y_train['label']).astype('float32')\n",
    "y_val_label = np.asarray(y_val['label']).astype('float32')\n",
    "y_test_label = np.asarray(y_test['label']).astype('float32')\n",
    "\n",
    "y_train_class = np.asarray(y_train['class']).astype('float32')\n",
    "y_val_class = np.asarray(y_val['class']).astype('float32')\n",
    "y_test_class = np.asarray(y_test['class']).astype('float32')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 166,
   "id": "ffe45e64",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Learning from training data and validating from validation data\n",
      "Epoch 1/50\n",
      "25/25 [==============================] - 17s 588ms/step - loss: 0.2315 - val_loss: 0.0889\n",
      "Epoch 2/50\n",
      "25/25 [==============================] - 15s 593ms/step - loss: 0.0798 - val_loss: 0.0486\n",
      "Epoch 3/50\n",
      "25/25 [==============================] - 14s 569ms/step - loss: 0.0524 - val_loss: 0.0355\n",
      "Epoch 4/50\n",
      "25/25 [==============================] - 14s 555ms/step - loss: 0.0414 - val_loss: 0.0303\n",
      "Epoch 5/50\n",
      "25/25 [==============================] - 14s 578ms/step - loss: 0.0305 - val_loss: 0.0262\n",
      "Epoch 6/50\n",
      "25/25 [==============================] - 15s 582ms/step - loss: 0.0326 - val_loss: 0.0236\n",
      "Epoch 7/50\n",
      "25/25 [==============================] - 14s 569ms/step - loss: 0.0297 - val_loss: 0.0240\n",
      "Epoch 8/50\n",
      "25/25 [==============================] - 14s 578ms/step - loss: 0.0239 - val_loss: 0.0250\n",
      "Epoch 9/50\n",
      "25/25 [==============================] - 15s 587ms/step - loss: 0.0238 - val_loss: 0.0247\n",
      "Epoch 10/50\n",
      "25/25 [==============================] - 14s 571ms/step - loss: 0.0242 - val_loss: 0.0220\n",
      "Epoch 11/50\n",
      "25/25 [==============================] - 14s 580ms/step - loss: 0.0219 - val_loss: 0.0207\n",
      "Epoch 12/50\n",
      "25/25 [==============================] - 15s 602ms/step - loss: 0.0208 - val_loss: 0.0194\n",
      "Epoch 13/50\n",
      "25/25 [==============================] - 14s 573ms/step - loss: 0.0174 - val_loss: 0.0194\n",
      "Epoch 14/50\n",
      "25/25 [==============================] - 14s 575ms/step - loss: 0.0168 - val_loss: 0.0175\n",
      "Epoch 15/50\n",
      "25/25 [==============================] - 14s 555ms/step - loss: 0.0152 - val_loss: 0.0158\n",
      "Epoch 16/50\n",
      "25/25 [==============================] - 15s 585ms/step - loss: 0.0143 - val_loss: 0.0140\n",
      "Epoch 17/50\n",
      "25/25 [==============================] - 14s 566ms/step - loss: 0.0171 - val_loss: 0.0143\n",
      "Epoch 18/50\n",
      "25/25 [==============================] - 15s 580ms/step - loss: 0.0154 - val_loss: 0.0153\n",
      "Epoch 19/50\n",
      "25/25 [==============================] - 14s 578ms/step - loss: 0.0136 - val_loss: 0.0113\n",
      "Epoch 20/50\n",
      "25/25 [==============================] - 15s 583ms/step - loss: 0.0128 - val_loss: 0.0110\n",
      "Epoch 21/50\n",
      "25/25 [==============================] - 14s 574ms/step - loss: 0.0140 - val_loss: 0.0103\n",
      "Epoch 22/50\n",
      "25/25 [==============================] - 15s 592ms/step - loss: 0.0116 - val_loss: 0.0085\n",
      "Epoch 23/50\n",
      "25/25 [==============================] - 14s 568ms/step - loss: 0.0098 - val_loss: 0.0087\n",
      "Epoch 24/50\n",
      "25/25 [==============================] - 15s 588ms/step - loss: 0.0100 - val_loss: 0.0096\n",
      "Epoch 25/50\n",
      "25/25 [==============================] - 14s 577ms/step - loss: 0.0092 - val_loss: 0.0100\n",
      "Epoch 26/50\n",
      "25/25 [==============================] - 14s 573ms/step - loss: 0.0113 - val_loss: 0.0128\n",
      "Epoch 27/50\n",
      "25/25 [==============================] - 14s 569ms/step - loss: 0.0113 - val_loss: 0.0115\n",
      "Epoch 28/50\n",
      "25/25 [==============================] - 14s 567ms/step - loss: 0.0129 - val_loss: 0.0093\n",
      "Epoch 29/50\n",
      "25/25 [==============================] - 14s 558ms/step - loss: 0.0110 - val_loss: 0.0117\n",
      "Epoch 30/50\n",
      "25/25 [==============================] - 14s 560ms/step - loss: 0.0122 - val_loss: 0.0089\n",
      "Epoch 31/50\n",
      "25/25 [==============================] - 14s 556ms/step - loss: 0.0103 - val_loss: 0.0075\n",
      "Epoch 32/50\n",
      "25/25 [==============================] - 14s 553ms/step - loss: 0.0101 - val_loss: 0.0069\n",
      "Epoch 33/50\n",
      "25/25 [==============================] - 15s 579ms/step - loss: 0.0090 - val_loss: 0.0075\n",
      "Epoch 34/50\n",
      "25/25 [==============================] - 14s 561ms/step - loss: 0.0119 - val_loss: 0.0083\n",
      "Epoch 35/50\n",
      "25/25 [==============================] - 15s 585ms/step - loss: 0.0124 - val_loss: 0.0097\n",
      "Epoch 36/50\n",
      "25/25 [==============================] - 14s 572ms/step - loss: 0.0097 - val_loss: 0.0082\n",
      "Epoch 37/50\n",
      "25/25 [==============================] - 14s 561ms/step - loss: 0.0096 - val_loss: 0.0076\n",
      "Epoch 38/50\n",
      "25/25 [==============================] - 15s 588ms/step - loss: 0.0072 - val_loss: 0.0058\n",
      "Epoch 39/50\n",
      "25/25 [==============================] - 15s 584ms/step - loss: 0.0067 - val_loss: 0.0059\n",
      "Epoch 40/50\n",
      "25/25 [==============================] - 14s 559ms/step - loss: 0.0079 - val_loss: 0.0067\n",
      "Epoch 41/50\n",
      "25/25 [==============================] - 14s 572ms/step - loss: 0.0082 - val_loss: 0.0071\n",
      "Epoch 42/50\n",
      "25/25 [==============================] - 14s 554ms/step - loss: 0.0076 - val_loss: 0.0095\n",
      "Epoch 43/50\n",
      "25/25 [==============================] - 14s 568ms/step - loss: 0.0076 - val_loss: 0.0077\n",
      "Epoch 44/50\n",
      "25/25 [==============================] - 14s 567ms/step - loss: 0.0068 - val_loss: 0.0070\n",
      "Epoch 45/50\n",
      "25/25 [==============================] - 14s 579ms/step - loss: 0.0079 - val_loss: 0.0070\n",
      "Epoch 46/50\n",
      "25/25 [==============================] - 14s 555ms/step - loss: 0.0066 - val_loss: 0.0066\n",
      "Epoch 47/50\n",
      "25/25 [==============================] - 14s 569ms/step - loss: 0.0072 - val_loss: 0.0072\n",
      "Epoch 48/50\n",
      "25/25 [==============================] - 14s 557ms/step - loss: 0.0064 - val_loss: 0.0057\n",
      "Epoch 49/50\n",
      "25/25 [==============================] - 15s 596ms/step - loss: 0.0077 - val_loss: 0.0058\n",
      "Epoch 50/50\n",
      "25/25 [==============================] - 14s 578ms/step - loss: 0.0078 - val_loss: 0.0060\n"
     ]
    }
   ],
   "source": [
    "print('Learning from training data and validating from validation data')\n",
    "history = model.fit([train_q1_seq,train_q2_seq, train_q3_seq],y_train_label.reshape(-1,1), epochs = 50, \n",
    "          batch_size=64,validation_data=([val_q1_seq, val_q2_seq, val_q3_seq],y_val_label.reshape(-1,1)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 167,
   "id": "8f587744",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "\n",
    "plt.plot(history.history['loss'])\n",
    "plt.plot(history.history['val_loss'])\n",
    "plt.title('Training and Validation Losses',size = 12)\n",
    "plt.ylabel('loss')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'val'], loc='upper right')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 168,
   "id": "9926a334",
   "metadata": {},
   "outputs": [],
   "source": [
    "# model_json = model.to_json()\n",
    "# with open(\"siamesemodel-triplet-loss.json\", \"w\") as json_file:\n",
    "#     json_file.write(model_json)\n",
    "# #serialize weights to HDF5\n",
    "# model.save_weights(\"siamesemodel-triplet-loss.h5\")\n",
    "# print(\"Saved model to disk\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 169,
   "id": "50ed38fe",
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "X_train_eval = base_model.predict(train_q1_seq)\n",
    "X_test_eval = base_model.predict(test_q1_seq)\n",
    "# The above embedding outputs will be used to predict. \n",
    "\n",
    "# TSNE - dimensionality reduction for data visualization\n",
    "tsne = TSNE()\n",
    "train_tsne_embeds = tsne.fit_transform(X_train_eval)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 170,
   "id": "754a9577",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "def scatter(x, labels, subtitle=None):\n",
    "    # Create a scatter plot of all the \n",
    "    # the embeddings of the model.\n",
    "    # We choose a color palette with seaborn.\n",
    "    palette = np.array(sns.color_palette(\"hls\", 2))\n",
    "    # We create a scatter plot.\n",
    "    f = plt.figure(figsize=(8, 8))\n",
    "    ax = plt.subplot(aspect='equal')\n",
    "    sc = ax.scatter(x[:,0], x[:,1], lw=0,alpha = 0.5, s=40,\n",
    "                    c=palette[labels.astype(np.int)] )\n",
    "    plt.xlim(-25, 25)\n",
    "    plt.ylim(-25, 25)\n",
    "    ax.axis('off')\n",
    "    ax.axis('tight')\n",
    "\n",
    "scatter(train_tsne_embeds, y_train_class)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "5bdd7564-88f1-44b6-8aec-b2f7c2c8bf2f",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 171,
   "id": "39efcd6c",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/100\n",
      "49/49 [==============================] - 0s 4ms/step - loss: 0.6565 - accuracy: 0.9987 - val_loss: 0.6362 - val_accuracy: 0.9959\n",
      "Epoch 2/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.6130 - accuracy: 0.9987 - val_loss: 0.5967 - val_accuracy: 0.9959\n",
      "Epoch 3/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.5729 - accuracy: 0.9987 - val_loss: 0.5598 - val_accuracy: 0.9959\n",
      "Epoch 4/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.5359 - accuracy: 0.9987 - val_loss: 0.5258 - val_accuracy: 0.9959\n",
      "Epoch 5/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.5017 - accuracy: 0.9987 - val_loss: 0.4944 - val_accuracy: 0.9959\n",
      "Epoch 6/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.4703 - accuracy: 0.9987 - val_loss: 0.4654 - val_accuracy: 0.9959\n",
      "Epoch 7/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.4413 - accuracy: 0.9987 - val_loss: 0.4384 - val_accuracy: 0.9959\n",
      "Epoch 8/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.4147 - accuracy: 0.9987 - val_loss: 0.4137 - val_accuracy: 0.9959\n",
      "Epoch 9/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.3900 - accuracy: 0.9987 - val_loss: 0.3909 - val_accuracy: 0.9959\n",
      "Epoch 10/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.3673 - accuracy: 0.9987 - val_loss: 0.3695 - val_accuracy: 0.9959\n",
      "Epoch 11/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.3463 - accuracy: 0.9987 - val_loss: 0.3499 - val_accuracy: 0.9959\n",
      "Epoch 12/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.3270 - accuracy: 0.9987 - val_loss: 0.3318 - val_accuracy: 0.9959\n",
      "Epoch 13/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.3092 - accuracy: 0.9987 - val_loss: 0.3148 - val_accuracy: 0.9959\n",
      "Epoch 14/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.2925 - accuracy: 0.9987 - val_loss: 0.2995 - val_accuracy: 0.9959\n",
      "Epoch 15/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.2772 - accuracy: 0.9987 - val_loss: 0.2848 - val_accuracy: 0.9959\n",
      "Epoch 16/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.2629 - accuracy: 0.9987 - val_loss: 0.2713 - val_accuracy: 0.9959\n",
      "Epoch 17/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.2497 - accuracy: 0.9987 - val_loss: 0.2587 - val_accuracy: 0.9959\n",
      "Epoch 18/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.2374 - accuracy: 0.9987 - val_loss: 0.2472 - val_accuracy: 0.9959\n",
      "Epoch 19/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.2260 - accuracy: 0.9987 - val_loss: 0.2362 - val_accuracy: 0.9959\n",
      "Epoch 20/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.2153 - accuracy: 0.9987 - val_loss: 0.2260 - val_accuracy: 0.9959\n",
      "Epoch 21/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.2053 - accuracy: 0.9987 - val_loss: 0.2165 - val_accuracy: 0.9959\n",
      "Epoch 22/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.1960 - accuracy: 0.9987 - val_loss: 0.2076 - val_accuracy: 0.9959\n",
      "Epoch 23/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.1873 - accuracy: 0.9987 - val_loss: 0.1992 - val_accuracy: 0.9959\n",
      "Epoch 24/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.1792 - accuracy: 0.9987 - val_loss: 0.1914 - val_accuracy: 0.9959\n",
      "Epoch 25/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.1715 - accuracy: 0.9987 - val_loss: 0.1839 - val_accuracy: 0.9959\n",
      "Epoch 26/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.1643 - accuracy: 0.9987 - val_loss: 0.1770 - val_accuracy: 0.9959\n",
      "Epoch 27/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.1576 - accuracy: 0.9987 - val_loss: 0.1705 - val_accuracy: 0.9959\n",
      "Epoch 28/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.1512 - accuracy: 0.9987 - val_loss: 0.1643 - val_accuracy: 0.9959\n",
      "Epoch 29/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.1452 - accuracy: 0.9987 - val_loss: 0.1585 - val_accuracy: 0.9959\n",
      "Epoch 30/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.1396 - accuracy: 0.9987 - val_loss: 0.1530 - val_accuracy: 0.9959\n",
      "Epoch 31/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.1342 - accuracy: 0.9987 - val_loss: 0.1478 - val_accuracy: 0.9959\n",
      "Epoch 32/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.1292 - accuracy: 0.9987 - val_loss: 0.1428 - val_accuracy: 0.9959\n",
      "Epoch 33/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.1244 - accuracy: 0.9987 - val_loss: 0.1381 - val_accuracy: 0.9959\n",
      "Epoch 34/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.1199 - accuracy: 0.9987 - val_loss: 0.1337 - val_accuracy: 0.9959\n",
      "Epoch 35/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.1156 - accuracy: 0.9987 - val_loss: 0.1294 - val_accuracy: 0.9959\n",
      "Epoch 36/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.1116 - accuracy: 0.9987 - val_loss: 0.1255 - val_accuracy: 0.9959\n",
      "Epoch 37/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.1077 - accuracy: 0.9987 - val_loss: 0.1217 - val_accuracy: 0.9959\n",
      "Epoch 38/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.1040 - accuracy: 0.9987 - val_loss: 0.1180 - val_accuracy: 0.9959\n",
      "Epoch 39/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.1005 - accuracy: 0.9987 - val_loss: 0.1145 - val_accuracy: 0.9959\n",
      "Epoch 40/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0972 - accuracy: 0.9987 - val_loss: 0.1112 - val_accuracy: 0.9959\n",
      "Epoch 41/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0941 - accuracy: 0.9987 - val_loss: 0.1080 - val_accuracy: 0.9959\n",
      "Epoch 42/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0910 - accuracy: 0.9987 - val_loss: 0.1051 - val_accuracy: 0.9959\n",
      "Epoch 43/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0882 - accuracy: 0.9987 - val_loss: 0.1022 - val_accuracy: 0.9959\n",
      "Epoch 44/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0854 - accuracy: 0.9987 - val_loss: 0.0995 - val_accuracy: 0.9959\n",
      "Epoch 45/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0828 - accuracy: 0.9987 - val_loss: 0.0967 - val_accuracy: 0.9959\n",
      "Epoch 46/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0803 - accuracy: 0.9987 - val_loss: 0.0942 - val_accuracy: 0.9959\n",
      "Epoch 47/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0779 - accuracy: 0.9987 - val_loss: 0.0918 - val_accuracy: 0.9959\n",
      "Epoch 48/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0756 - accuracy: 0.9987 - val_loss: 0.0895 - val_accuracy: 0.9959\n",
      "Epoch 49/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0734 - accuracy: 0.9987 - val_loss: 0.0873 - val_accuracy: 0.9959\n",
      "Epoch 50/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0713 - accuracy: 0.9987 - val_loss: 0.0851 - val_accuracy: 0.9959\n",
      "Epoch 51/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0693 - accuracy: 0.9987 - val_loss: 0.0831 - val_accuracy: 0.9959\n",
      "Epoch 52/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0673 - accuracy: 0.9987 - val_loss: 0.0811 - val_accuracy: 0.9959\n",
      "Epoch 53/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0655 - accuracy: 0.9987 - val_loss: 0.0792 - val_accuracy: 0.9959\n",
      "Epoch 54/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0637 - accuracy: 0.9987 - val_loss: 0.0774 - val_accuracy: 0.9959\n",
      "Epoch 55/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0620 - accuracy: 0.9987 - val_loss: 0.0756 - val_accuracy: 0.9959\n",
      "Epoch 56/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0603 - accuracy: 0.9987 - val_loss: 0.0739 - val_accuracy: 0.9959\n",
      "Epoch 57/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0587 - accuracy: 0.9987 - val_loss: 0.0722 - val_accuracy: 0.9959\n",
      "Epoch 58/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0572 - accuracy: 0.9987 - val_loss: 0.0707 - val_accuracy: 0.9959\n",
      "Epoch 59/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0557 - accuracy: 0.9987 - val_loss: 0.0692 - val_accuracy: 0.9959\n",
      "Epoch 60/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0543 - accuracy: 0.9987 - val_loss: 0.0677 - val_accuracy: 0.9959\n",
      "Epoch 61/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0529 - accuracy: 0.9987 - val_loss: 0.0663 - val_accuracy: 0.9959\n",
      "Epoch 62/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0516 - accuracy: 0.9987 - val_loss: 0.0649 - val_accuracy: 0.9959\n",
      "Epoch 63/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0503 - accuracy: 0.9987 - val_loss: 0.0636 - val_accuracy: 0.9959\n",
      "Epoch 64/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0491 - accuracy: 0.9987 - val_loss: 0.0623 - val_accuracy: 0.9959\n",
      "Epoch 65/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0479 - accuracy: 0.9987 - val_loss: 0.0611 - val_accuracy: 0.9959\n",
      "Epoch 66/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0468 - accuracy: 0.9987 - val_loss: 0.0599 - val_accuracy: 0.9959\n",
      "Epoch 67/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0457 - accuracy: 0.9987 - val_loss: 0.0587 - val_accuracy: 0.9959\n",
      "Epoch 68/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0446 - accuracy: 0.9987 - val_loss: 0.0576 - val_accuracy: 0.9959\n",
      "Epoch 69/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0436 - accuracy: 0.9987 - val_loss: 0.0565 - val_accuracy: 0.9959\n",
      "Epoch 70/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0426 - accuracy: 0.9987 - val_loss: 0.0555 - val_accuracy: 0.9959\n",
      "Epoch 71/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0416 - accuracy: 0.9987 - val_loss: 0.0545 - val_accuracy: 0.9959\n",
      "Epoch 72/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0407 - accuracy: 0.9987 - val_loss: 0.0535 - val_accuracy: 0.9959\n",
      "Epoch 73/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0398 - accuracy: 0.9987 - val_loss: 0.0525 - val_accuracy: 0.9959\n",
      "Epoch 74/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0389 - accuracy: 0.9987 - val_loss: 0.0516 - val_accuracy: 0.9959\n",
      "Epoch 75/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0381 - accuracy: 0.9987 - val_loss: 0.0507 - val_accuracy: 0.9959\n",
      "Epoch 76/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0372 - accuracy: 0.9987 - val_loss: 0.0498 - val_accuracy: 0.9959\n",
      "Epoch 77/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0364 - accuracy: 0.9987 - val_loss: 0.0489 - val_accuracy: 0.9959\n",
      "Epoch 78/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0357 - accuracy: 0.9987 - val_loss: 0.0481 - val_accuracy: 0.9959\n",
      "Epoch 79/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0349 - accuracy: 0.9987 - val_loss: 0.0473 - val_accuracy: 0.9959\n",
      "Epoch 80/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0342 - accuracy: 0.9987 - val_loss: 0.0465 - val_accuracy: 0.9959\n",
      "Epoch 81/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0335 - accuracy: 0.9987 - val_loss: 0.0458 - val_accuracy: 0.9959\n",
      "Epoch 82/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0328 - accuracy: 0.9987 - val_loss: 0.0450 - val_accuracy: 0.9959\n",
      "Epoch 83/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0322 - accuracy: 0.9987 - val_loss: 0.0443 - val_accuracy: 0.9959\n",
      "Epoch 84/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0315 - accuracy: 0.9987 - val_loss: 0.0436 - val_accuracy: 0.9959\n",
      "Epoch 85/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0309 - accuracy: 0.9987 - val_loss: 0.0430 - val_accuracy: 0.9959\n",
      "Epoch 86/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0303 - accuracy: 0.9987 - val_loss: 0.0423 - val_accuracy: 0.9959\n",
      "Epoch 87/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0297 - accuracy: 0.9987 - val_loss: 0.0417 - val_accuracy: 0.9959\n",
      "Epoch 88/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0292 - accuracy: 0.9987 - val_loss: 0.0410 - val_accuracy: 0.9959\n",
      "Epoch 89/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0286 - accuracy: 0.9987 - val_loss: 0.0404 - val_accuracy: 0.9959\n",
      "Epoch 90/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0281 - accuracy: 0.9987 - val_loss: 0.0399 - val_accuracy: 0.9959\n",
      "Epoch 91/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0276 - accuracy: 0.9987 - val_loss: 0.0393 - val_accuracy: 0.9959\n",
      "Epoch 92/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0271 - accuracy: 0.9987 - val_loss: 0.0387 - val_accuracy: 0.9959\n",
      "Epoch 93/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0266 - accuracy: 0.9987 - val_loss: 0.0382 - val_accuracy: 0.9959\n",
      "Epoch 94/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0261 - accuracy: 0.9987 - val_loss: 0.0376 - val_accuracy: 0.9959\n",
      "Epoch 95/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0256 - accuracy: 0.9987 - val_loss: 0.0371 - val_accuracy: 0.9959\n",
      "Epoch 96/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0252 - accuracy: 0.9987 - val_loss: 0.0366 - val_accuracy: 0.9959\n",
      "Epoch 97/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0248 - accuracy: 0.9987 - val_loss: 0.0361 - val_accuracy: 0.9959\n",
      "Epoch 98/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0243 - accuracy: 0.9987 - val_loss: 0.0356 - val_accuracy: 0.9959\n",
      "Epoch 99/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0239 - accuracy: 0.9987 - val_loss: 0.0351 - val_accuracy: 0.9959\n",
      "Epoch 100/100\n",
      "49/49 [==============================] - 0s 2ms/step - loss: 0.0235 - accuracy: 0.9987 - val_loss: 0.0347 - val_accuracy: 0.9959\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x7f6a86320c40>"
      ]
     },
     "execution_count": 171,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# classifier that computes the class of a specific embedding for prediction purposes\n",
    "classifier_input = Input(shape=(2,))\n",
    "classifier_output = Dense(2, activation='softmax')(classifier_input)\n",
    "classifier_model = Model(classifier_input, classifier_output)\n",
    "\n",
    "# onehot encoding vectors to 2 classes\n",
    "Y_train_onehot = np_utils.to_categorical(y_train_class, 2)\n",
    "Y_test_onehot = np_utils.to_categorical(y_test_class, 2)\n",
    "\n",
    "classifier_model.compile(optimizer='adam',loss='categorical_crossentropy',metrics=['accuracy'])\n",
    "classifier_model.fit(X_train_eval,Y_train_onehot, validation_data=(X_test_eval, Y_test_onehot),epochs=100)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 172,
   "id": "aa8dafaf",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Classification report on validation data \n",
      "               precision    recall  f1-score   support\n",
      "\n",
      "           0     0.9913    1.0000    0.9957       229\n",
      "           1     1.0000    0.9922    0.9961       255\n",
      "\n",
      "    accuracy                         0.9959       484\n",
      "   macro avg     0.9957    0.9961    0.9959       484\n",
      "weighted avg     0.9959    0.9959    0.9959       484\n",
      "\n"
     ]
    }
   ],
   "source": [
    "\n",
    "input = X_test_eval\n",
    "y_pred = classifier_model.predict(input)\n",
    "y_pred=np.argmax(y_pred, axis=1)\n",
    "y_test = np.argmax(Y_test_onehot,axis = 1)\n",
    "print(\"Classification report on validation data \\n\",classification_report(y_test, y_pred, digits=4))\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 173,
   "id": "0910b1d9-8a2c-4559-a36f-cdbebcf8fa42",
   "metadata": {},
   "outputs": [],
   "source": [
    "df_test = pd.DataFrame(columns=['text', 'label', 'stance'])\n",
    "\n",
    "for target in ['Donald Trump','Hillary Clinton']:\n",
    "    \n",
    "    with open(test_data_file_m, 'r') as testfile:\n",
    "        for line in testfile:\n",
    "            line = line.replace('#SemST', '').strip()\n",
    "            line = line.split('\\t')\n",
    "            if line[0].strip() != 'ID' and target in line[1].strip():\n",
    "                tweet = line[2]\n",
    "                df_test.loc[len(df_test)] = [str(tweet),0, line[3]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 174,
   "id": "38edcd0b-bf09-457b-a493-d4f9da213a28",
   "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>text</th>\n",
       "      <th>label</th>\n",
       "      <th>stance</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>@chriswallace There may be 2000 nutjobs of tha...</td>\n",
       "      <td>0</td>\n",
       "      <td>NONE</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>@realDonaldTrump u talk a big bigoted talk &amp; y...</td>\n",
       "      <td>0</td>\n",
       "      <td>AGAINST</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>@ac360 @ananavarro @andersoncooper #AC360 HECK...</td>\n",
       "      <td>0</td>\n",
       "      <td>FAVOR</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Watching what Donald Trump said about Mexicans...</td>\n",
       "      <td>0</td>\n",
       "      <td>AGAINST</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>\"Are people deaf to the word illegal? I'm all ...</td>\n",
       "      <td>0</td>\n",
       "      <td>NONE</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>467</th>\n",
       "      <td>.@HillaryClinton Looking 4ward 2 hearing your ...</td>\n",
       "      <td>0</td>\n",
       "      <td>FAVOR</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>468</th>\n",
       "      <td>@mataharikrishna I'm loving it too! Draw that ...</td>\n",
       "      <td>0</td>\n",
       "      <td>NONE</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>469</th>\n",
       "      <td>@finneyk Can't stand @msnbc anymore, but hope ...</td>\n",
       "      <td>0</td>\n",
       "      <td>FAVOR</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>470</th>\n",
       "      <td>Hillary can't create jobs! Last time she had a...</td>\n",
       "      <td>0</td>\n",
       "      <td>AGAINST</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>471</th>\n",
       "      <td>It's amazing to me how if you want a secure bo...</td>\n",
       "      <td>0</td>\n",
       "      <td>NONE</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>472 rows × 3 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                  text label   stance\n",
       "0    @chriswallace There may be 2000 nutjobs of tha...     0     NONE\n",
       "1    @realDonaldTrump u talk a big bigoted talk & y...     0  AGAINST\n",
       "2    @ac360 @ananavarro @andersoncooper #AC360 HECK...     0    FAVOR\n",
       "3    Watching what Donald Trump said about Mexicans...     0  AGAINST\n",
       "4    \"Are people deaf to the word illegal? I'm all ...     0     NONE\n",
       "..                                                 ...   ...      ...\n",
       "467  .@HillaryClinton Looking 4ward 2 hearing your ...     0    FAVOR\n",
       "468  @mataharikrishna I'm loving it too! Draw that ...     0     NONE\n",
       "469  @finneyk Can't stand @msnbc anymore, but hope ...     0    FAVOR\n",
       "470  Hillary can't create jobs! Last time she had a...     0  AGAINST\n",
       "471  It's amazing to me how if you want a secure bo...     0     NONE\n",
       "\n",
       "[472 rows x 3 columns]"
      ]
     },
     "execution_count": 174,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 175,
   "id": "da4fa5d7-a7f1-4cd2-aee5-f2ed61013b3e",
   "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>text</th>\n",
       "      <th>label</th>\n",
       "      <th>stance</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>@chriswallace There may be 2000 nutjobs of tha...</td>\n",
       "      <td>0</td>\n",
       "      <td>NONE</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>@realDonaldTrump u talk a big bigoted talk &amp; y...</td>\n",
       "      <td>0</td>\n",
       "      <td>AGAINST</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>@ac360 @ananavarro @andersoncooper #AC360 HECK...</td>\n",
       "      <td>0</td>\n",
       "      <td>FAVOR</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Watching what Donald Trump said about Mexicans...</td>\n",
       "      <td>0</td>\n",
       "      <td>AGAINST</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>\"Are people deaf to the word illegal? I'm all ...</td>\n",
       "      <td>0</td>\n",
       "      <td>NONE</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1595</th>\n",
       "      <td>Disney CEO Bob Iger To Reportedly Stay On To H...</td>\n",
       "      <td>1</td>\n",
       "      <td>comment</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1596</th>\n",
       "      <td>If the Disney/Fox deal goes through, #Collider...</td>\n",
       "      <td>1</td>\n",
       "      <td>comment</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1597</th>\n",
       "      <td>Check out CHILDRENS , VHS TAPES  LOT OF 26, In...</td>\n",
       "      <td>1</td>\n",
       "      <td>unrelated</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1598</th>\n",
       "      <td>@SpencerMemes If Disney buys some of  Fox that...</td>\n",
       "      <td>1</td>\n",
       "      <td>comment</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1599</th>\n",
       "      <td>Disney &amp;amp; Fox Deal Discussions Rumored to B...</td>\n",
       "      <td>1</td>\n",
       "      <td>support</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>872 rows × 3 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                   text label     stance\n",
       "0     @chriswallace There may be 2000 nutjobs of tha...     0       NONE\n",
       "1     @realDonaldTrump u talk a big bigoted talk & y...     0    AGAINST\n",
       "2     @ac360 @ananavarro @andersoncooper #AC360 HECK...     0      FAVOR\n",
       "3     Watching what Donald Trump said about Mexicans...     0    AGAINST\n",
       "4     \"Are people deaf to the word illegal? I'm all ...     0       NONE\n",
       "...                                                 ...   ...        ...\n",
       "1595  Disney CEO Bob Iger To Reportedly Stay On To H...     1    comment\n",
       "1596  If the Disney/Fox deal goes through, #Collider...     1    comment\n",
       "1597  Check out CHILDRENS , VHS TAPES  LOT OF 26, In...     1  unrelated\n",
       "1598  @SpencerMemes If Disney buys some of  Fox that...     1    comment\n",
       "1599  Disney &amp; Fox Deal Discussions Rumored to B...     1    support\n",
       "\n",
       "[872 rows x 3 columns]"
      ]
     },
     "execution_count": 175,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_test_2 = pd.DataFrame()\n",
    "df_a = df_ent_train[df_ent_train['target'] == 'FOXA_DIS'][1200:1600]\n",
    "\n",
    "df_test_new = pd.DataFrame(columns=['text', 'label', 'stance'])\n",
    "df_test_new['text'] = df_a['tweet']\n",
    "df_test_new['label'] = 1\n",
    "df_test_new['stance'] = df_a['stance']\n",
    "\n",
    "df_test = df_test.append(df_test_new)\n",
    "df_test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 122,
   "id": "a01918d3-873d-465d-b6a1-499dd9190f01",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array(['AET_HUM', 'ANTM_CI', 'CI_ESRX', 'CVS_AET'], dtype=object)"
      ]
     },
     "execution_count": 122,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# np.unique(df_hlt_train['target'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 176,
   "id": "85442232-8d04-4f9e-97b8-f1adbc4e443e",
   "metadata": {},
   "outputs": [],
   "source": [
    "# df_test_2 = pd.DataFrame()\n",
    "# df_a = df_hlt_train[df_hlt_train['target'] == 'AET_HUM'][300:400]\n",
    "# df_b = df_hlt_train[df_hlt_train['target'] == 'ANTM_CI'][300:400]\n",
    "# df_c = df_hlt_train[df_hlt_train['target'] == 'CI_ESRX'][300:400]\n",
    "# df_d = df_hlt_train[df_hlt_train['target'] == 'CVS_AET'][300:400]\n",
    "\n",
    "# df_test_2 = df_test_2.append(df_a, ignore_index=True)\n",
    "# df_test_2 = df_test_2.append(df_b, ignore_index=True)\n",
    "# df_test_2 = df_test_2.append(df_c, ignore_index=True)\n",
    "# df_test_2 = df_test_2.append(df_d, ignore_index=True)\n",
    "# df_test_new = pd.DataFrame(columns=['text', 'label', 'stance'])\n",
    "# df_test_new['text'] = df_test_2['tweet']\n",
    "# df_test_new['label'] = 1\n",
    "# df_test_new['stance'] = df_test_2['stance']\n",
    "\n",
    "# df_test = df_test.append(df_test_new)\n",
    "# df_test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 177,
   "id": "31756d70-6dce-4d6e-a533-574a28b586a8",
   "metadata": {},
   "outputs": [
    {
     "ename": "SyntaxError",
     "evalue": "invalid syntax (3375635915.py, line 1)",
     "output_type": "error",
     "traceback": [
      "\u001b[0;36m  File \u001b[0;32m\"/tmp/ipykernel_77682/3375635915.py\"\u001b[0;36m, line \u001b[0;32m1\u001b[0m\n\u001b[0;31m    here---\u001b[0m\n\u001b[0m           ^\u001b[0m\n\u001b[0;31mSyntaxError\u001b[0m\u001b[0;31m:\u001b[0m invalid syntax\n"
     ]
    }
   ],
   "source": [
    "here---"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 178,
   "id": "2156e31f",
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "\n",
    "    \n",
    "df_test['clean_text'] = df_test['text'].apply(lambda x: remove_url(str(x)))\n",
    "df_test['clean_text'] = df_test['clean_text'].apply(lambda x: remove_emoji(str(x)))\n",
    "df_test['clean_text'] = df_test['clean_text'].apply(lambda x: remove_html(str(x)))\n",
    "df_test['clean_text'] = df_test['clean_text'].apply(lambda x: remove_punct(str(x)))\n",
    "df_test['clean_text'] = df_test['clean_text'].apply(word_tokenize)\n",
    "df_test['clean_text'] = df_test['clean_text'].apply(lambda x: [word.lower() for word in x])\n",
    "df_test['clean_text'] = df_test['clean_text'].apply(lambda x: [word for word in x if word not in stop])\n",
    "df_test['clean_text'] = df_test['clean_text'].apply(nltk.tag.pos_tag)\n",
    "\n",
    "def get_wordnet_pos(tag):\n",
    "    if tag.startswith('J'):\n",
    "        return wordnet.ADJ\n",
    "    elif tag.startswith('V'):\n",
    "        return wordnet.VERB\n",
    "    elif tag.startswith('N'):\n",
    "        return wordnet.NOUN\n",
    "    elif tag.startswith('R'):\n",
    "        return wordnet.ADV\n",
    "    else:\n",
    "        return wordnet.NOUN\n",
    "    \n",
    "df_test['clean_text'] = df_test['clean_text'].apply(\n",
    "    lambda x: [(word, get_wordnet_pos(pos_tag)) for (word, pos_tag) in x])\n",
    "wnl = WordNetLemmatizer()\n",
    "df_test['clean_text'] = df_test['clean_text'].apply(\n",
    "    lambda x: [wnl.lemmatize(word, tag) for word, tag in x])\n",
    "df_test['clean_text'] = df_test['clean_text'].apply(\n",
    "    lambda x: [word for word in x if word not in stop])\n",
    "df_test['clean_text'] = [' '.join(map(str, l)) for l in df_test['clean_text']]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 179,
   "id": "2614ac42",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Classification report on Test data \n",
      "               precision    recall  f1-score   support\n",
      "\n",
      "           0     0.9812    0.9958    0.9884       472\n",
      "           1     0.9949    0.9775    0.9861       400\n",
      "\n",
      "    accuracy                         0.9874       872\n",
      "   macro avg     0.9881    0.9866    0.9873       872\n",
      "weighted avg     0.9875    0.9874    0.9874       872\n",
      "\n"
     ]
    }
   ],
   "source": [
    "df_test['clean_text'] = df_test['clean_text'].astype(str)\n",
    "y_q1_seq = t.texts_to_sequences(df_test['clean_text'].values)\n",
    "\n",
    "\n",
    "y_q1_seq = pad_sequences(y_q1_seq, maxlen=max_len, padding='post')\n",
    "y_test_eval = base_model.predict(y_q1_seq)\n",
    "\n",
    "input_2 = y_test_eval\n",
    "y_pred_2 = classifier_model.predict(input_2)\n",
    "\n",
    "y2_test_class = np.asarray(df_test['label']).astype('float32')\n",
    "y2_test_onehot = np_utils.to_categorical(y2_test_class, 2)\n",
    "\n",
    "y_pred_2=np.argmax(y_pred_2, axis=1)\n",
    "y_test_2 = np.argmax(y2_test_onehot,axis = 1)\n",
    "print(\"\")\n",
    "print(\"Classification report on Test data \\n\",classification_report(y_test_2, y_pred_2, digits=4))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f352a8f0",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "b2d63e44",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "1b2e8c43",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f288c20d-1cac-4160-aa4e-fbb7f8b42c1f",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "6bac4379",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "d9d31daf-26df-46ea-8bb8-ce70182afbd8",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "d02a5278",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "ad824d7d",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "b13766df",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "id": "68204cb4",
   "metadata": {},
   "source": [
    "# predicting above model on other target and getting that data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 180,
   "id": "09df8135-d10d-47ea-8dc7-a8bdedcce653",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "22101\n"
     ]
    }
   ],
   "source": [
    "print(len(df_hlt_train))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 181,
   "id": "970a9ec7",
   "metadata": {},
   "outputs": [],
   "source": [
    "# df_oth = pd.DataFrame(columns=['text', 'label', 'stance'])\n",
    "\n",
    "df_oth = pd.DataFrame()\n",
    "#df_oth['text'] = df_ent_train['tweet']\n",
    "df_oth['text'] = df_hlt_train['tweet']\n",
    "df_oth['label'] = \"NA\"\n",
    "#df_oth['stance'] = df_ent_train['stance']\n",
    "df_oth['stance'] = df_hlt_train['stance']\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 182,
   "id": "25522757",
   "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>text</th>\n",
       "      <th>label</th>\n",
       "      <th>stance</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Deals , deals and more ...CVS buying Aetna for...</td>\n",
       "      <td>NA</td>\n",
       "      <td>support</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Do investigating all the Chinese hacks: OPM wh...</td>\n",
       "      <td>NA</td>\n",
       "      <td>unrelated</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>INVESTMENT NEWS Zacks Industry Outlook Highlig...</td>\n",
       "      <td>NA</td>\n",
       "      <td>unrelated</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>#TSXVenture #stocks: #Merger #Acquisition: Ant...</td>\n",
       "      <td>NA</td>\n",
       "      <td>unrelated</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>#NTT #NipponTelegraph &amp;amp; #Telephone Corpora...</td>\n",
       "      <td>NA</td>\n",
       "      <td>unrelated</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>22096</th>\n",
       "      <td>Recent industry milestones like #Amazon, #Berk...</td>\n",
       "      <td>NA</td>\n",
       "      <td>comment</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>22097</th>\n",
       "      <td>How the CVS, Aetna Deal Will Overhaul Healthca...</td>\n",
       "      <td>NA</td>\n",
       "      <td>comment</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>22098</th>\n",
       "      <td>@berthacoombs Obama and DOJ must decide if the...</td>\n",
       "      <td>NA</td>\n",
       "      <td>comment</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>22099</th>\n",
       "      <td>Part ANIMA HUMANA: What is A listing Contract?...</td>\n",
       "      <td>NA</td>\n",
       "      <td>unrelated</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>22100</th>\n",
       "      <td>Anthem-Cigna's market doubters have a point ht...</td>\n",
       "      <td>NA</td>\n",
       "      <td>comment</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>22101 rows × 3 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                    text label     stance\n",
       "0      Deals , deals and more ...CVS buying Aetna for...    NA    support\n",
       "1      Do investigating all the Chinese hacks: OPM wh...    NA  unrelated\n",
       "2      INVESTMENT NEWS Zacks Industry Outlook Highlig...    NA  unrelated\n",
       "3      #TSXVenture #stocks: #Merger #Acquisition: Ant...    NA  unrelated\n",
       "4      #NTT #NipponTelegraph &amp; #Telephone Corpora...    NA  unrelated\n",
       "...                                                  ...   ...        ...\n",
       "22096  Recent industry milestones like #Amazon, #Berk...    NA    comment\n",
       "22097  How the CVS, Aetna Deal Will Overhaul Healthca...    NA    comment\n",
       "22098  @berthacoombs Obama and DOJ must decide if the...    NA    comment\n",
       "22099  Part ANIMA HUMANA: What is A listing Contract?...    NA  unrelated\n",
       "22100  Anthem-Cigna's market doubters have a point ht...    NA    comment\n",
       "\n",
       "[22101 rows x 3 columns]"
      ]
     },
     "execution_count": 182,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_oth"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 183,
   "id": "4ee49ba6",
   "metadata": {},
   "outputs": [],
   "source": [
    "#  with open(train_data_file_m, 'r') as testfile:\n",
    "#         for line in testfile:\n",
    "            \n",
    "#             line = line.replace('#SemST', '').strip()\n",
    "#             line = line.split('\\t')\n",
    "            \n",
    "#             if line[0].strip() != 'ID' and test_on in line[1].strip():\n",
    "#                 tweet = line[2]\n",
    "#                 #print(tweet)\n",
    "#                 df_oth.loc[len(df_oth)] = [str(tweet),'NA',line[3]]\n",
    "                "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "4a65a0e0",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 184,
   "id": "bef43359",
   "metadata": {},
   "outputs": [],
   "source": [
    "df_oth['clean_text'] = df_oth['text'].apply(lambda x: remove_url(str(x)))\n",
    "df_oth['clean_text'] = df_oth['clean_text'].apply(lambda x: remove_emoji(str(x)))\n",
    "df_oth['clean_text'] = df_oth['clean_text'].apply(lambda x: remove_html(str(x)))\n",
    "df_oth['clean_text'] = df_oth['clean_text'].apply(lambda x: remove_punct(str(x)))\n",
    "df_oth['clean_text'] = df_oth['clean_text'].apply(word_tokenize)\n",
    "df_oth['clean_text'] = df_oth['clean_text'].apply(lambda x: [word.lower() for word in x])\n",
    "df_oth['clean_text'] = df_oth['clean_text'].apply(lambda x: [word for word in x if word not in stop])\n",
    "df_oth['clean_text'] = df_oth['clean_text'].apply(nltk.tag.pos_tag)\n",
    "\n",
    "def get_wordnet_pos(tag):\n",
    "    if tag.startswith('J'):\n",
    "        return wordnet.ADJ\n",
    "    elif tag.startswith('V'):\n",
    "        return wordnet.VERB\n",
    "    elif tag.startswith('N'):\n",
    "        return wordnet.NOUN\n",
    "    elif tag.startswith('R'):\n",
    "        return wordnet.ADV\n",
    "    else:\n",
    "        return wordnet.NOUN\n",
    "    \n",
    "df_oth['clean_text'] = df_oth['clean_text'].apply(\n",
    "    lambda x: [(word, get_wordnet_pos(pos_tag)) for (word, pos_tag) in x])\n",
    "wnl = WordNetLemmatizer()\n",
    "df_oth['clean_text'] = df_oth['clean_text'].apply(\n",
    "    lambda x: [wnl.lemmatize(word, tag) for word, tag in x])\n",
    "df_oth['clean_text'] = df_oth['clean_text'].apply(\n",
    "    lambda x: [word for word in x if word not in stop])\n",
    "df_oth['clean_text'] = [' '.join(map(str, l)) for l in df_oth['clean_text']]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 185,
   "id": "a37f8ca7",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'comment'"
      ]
     },
     "execution_count": 185,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_oth['stance'][5]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 186,
   "id": "d23d2b5a",
   "metadata": {},
   "outputs": [],
   "source": [
    "df_oth['clean_text'] = df_oth['clean_text'].astype(str)\n",
    "oth_q1_seq = t.texts_to_sequences(df_oth['clean_text'].values)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 187,
   "id": "52b17e7d",
   "metadata": {},
   "outputs": [],
   "source": [
    "oth_q1_seq = pad_sequences(oth_q1_seq, maxlen=max_len, padding='post')\n",
    "oth_test_eval = base_model.predict(oth_q1_seq)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 188,
   "id": "8513bbe8",
   "metadata": {},
   "outputs": [],
   "source": [
    "input_3 = oth_test_eval\n",
    "oth_pred_3 = classifier_model.predict(input_3)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 189,
   "id": "91c15ac2-5048-433b-8c88-ff2114bc37b9",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0.01769373, 0.9823063 ],\n",
       "       [0.97473735, 0.02526271],\n",
       "       [0.9808865 , 0.01911353],\n",
       "       ...,\n",
       "       [0.12097012, 0.8790299 ],\n",
       "       [0.01264185, 0.9873582 ],\n",
       "       [0.9434776 , 0.05652239]], dtype=float32)"
      ]
     },
     "execution_count": 189,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "oth_pred_3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "077bf2b9-4bf9-45fb-8acc-3716642e2130",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "302bb6a5-9230-41bb-8f38-d59f1ac15aaf",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 190,
   "id": "934c5b48",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([1, 0, 0, ..., 1, 1, 0])"
      ]
     },
     "execution_count": 190,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "oth_pred_4=np.argmax(oth_pred_3, axis=1)\n",
    "oth_pred_4"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 191,
   "id": "249959d0-cb8a-4010-b92a-aa8f56d93096",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([0.01769373, 0.9823063 ], dtype=float32)"
      ]
     },
     "execution_count": 191,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "oth_pred_3[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 192,
   "id": "a76e5fe5",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{0: 5645, 1: 16456}"
      ]
     },
     "execution_count": 192,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "unique,count = np.unique(oth_pred_4,return_counts=True)\n",
    "dict(zip(unique, count))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b7913f01-aed4-4c9e-8f34-9092a4580d4d",
   "metadata": {
    "tags": []
   },
   "source": [
    "# Change the file name "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 140,
   "id": "94ed4734-8aaa-4225-b6b6-b6126ae52fb0",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0.9939083 , 0.00609171],\n",
       "       [0.02480351, 0.9751965 ],\n",
       "       [0.06026034, 0.9397397 ],\n",
       "       ...,\n",
       "       [0.99692506, 0.0030749 ],\n",
       "       [0.06324334, 0.9367567 ],\n",
       "       [0.02469326, 0.9753067 ]], dtype=float32)"
      ]
     },
     "execution_count": 140,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "oth_pred_3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 193,
   "id": "f86ce53f",
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "threshold = [0.50, 0.60, 0.70, 0.80, 0.90]\n",
    "for th in threshold:\n",
    "    data = []\n",
    "    \n",
    "    for i in range(len(oth_pred_3)):\n",
    "        if oth_pred_3[i][1] >= th:\n",
    "            values = {}\n",
    "            values['tweet'] = df_oth['text'][i]\n",
    "            values[\"stance\"] = df_oth['stance'][i]\n",
    "            data.append(values)\n",
    "    jsonString = json.dumps(data)\n",
    "    with open(\"rich_data/ent_pol_hlt/ent_pol_hlt\" + str(th) + \".json\", \"w\") as jsonFile:\n",
    "        jsonFile.write(jsonString)\n",
    "        jsonFile.close()\n",
    "    \n",
    "# file_name = 'rich_data/hlt_pol_ent/hlt_pol_ent' + str(th) +'.txt'\n",
    "# with open(file_name,'w') as f:\n",
    "#     for i in range(len(oth_pred_3)):\n",
    "#         if oth_pred_4[i] == 1:\n",
    "#             if oth_pred_3[i][1] >= th:\n",
    "#                 c+= 1\n",
    "#                 f.write(str(df_oth['text'][i]).strip() + \"\\t\" + str(df_oth['stance'][i]).strip())\n",
    "#                 f.write('\\n')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "id": "5d92cc73-afac-4e6d-89b8-dc33830cda78",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "3458"
      ]
     },
     "execution_count": 71,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "c"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "id": "9577c8d9",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "7022"
      ]
     },
     "execution_count": 83,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "c = 0\n",
    "with open(file_name,'r') as f:\n",
    "    for i in f:\n",
    "        c += 1\n",
    "c"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "id": "10e20262-9020-47a1-8f15-5209bf7522de",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "rich_data/hlt_pol_ent/hlt_pol_ent0.5.txt\n"
     ]
    }
   ],
   "source": [
    "print(file_name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "id": "58a212d2",
   "metadata": {},
   "outputs": [],
   "source": [
    "# with open('rich_data.txt','w') as f:\n",
    "#     for tweet in rich_data:\n",
    "#         f.write(tweet)\n",
    "#         f.write('\\n')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "id": "522621f5",
   "metadata": {},
   "outputs": [],
   "source": [
    "df_rich_data.to_csv('rich_data.csv',sep='\\t')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "23dc36b7",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "nlpKernel",
   "language": "python",
   "name": "nlpkernel"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}