2.0-Word2Vec.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "_cell_guid": "b1076dfc-b9ad-4769-8c92-a6c4dae69d19",
    "_uuid": "8f2839f25d086af736a60e9eeb907d3b93b6e0e5"
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import re\n",
    "import gensim\n",
    "\n",
    "import warnings\n",
    "warnings.filterwarnings('ignore')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "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>LABEL</th>\n",
       "      <th>RATING</th>\n",
       "      <th>VERIFIED_PURCHASE</th>\n",
       "      <th>PRODUCT_CATEGORY</th>\n",
       "      <th>REVIEW_TEXT</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "      <td>9</td>\n",
       "      <td>['useful', 'think', 'product', 'save', 'day', ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1</td>\n",
       "      <td>4</td>\n",
       "      <td>1</td>\n",
       "      <td>22</td>\n",
       "      <td>['new', 'era', 'battery', 'lithium', 'battery'...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>1</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>18</td>\n",
       "      <td>['doesn', 'swing', 'purchase', 'swing', 'baby'...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>1</td>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "      <td>15</td>\n",
       "      <td>['great', 'computing', 'look', 'inexpensive', ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>1</td>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "      <td>28</td>\n",
       "      <td>['use', 'twice', 'week', 'use', 'twice', 'week...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   LABEL  RATING  VERIFIED_PURCHASE  PRODUCT_CATEGORY  \\\n",
       "0      1       4                  0                 9   \n",
       "1      1       4                  1                22   \n",
       "2      1       3                  0                18   \n",
       "3      1       4                  0                15   \n",
       "4      1       4                  0                28   \n",
       "\n",
       "                                         REVIEW_TEXT  \n",
       "0  ['useful', 'think', 'product', 'save', 'day', ...  \n",
       "1  ['new', 'era', 'battery', 'lithium', 'battery'...  \n",
       "2  ['doesn', 'swing', 'purchase', 'swing', 'baby'...  \n",
       "3  ['great', 'computing', 'look', 'inexpensive', ...  \n",
       "4  ['use', 'twice', 'week', 'use', 'twice', 'week...  "
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data = pd.read_csv(\"/kaggle/input/1-preprocessing/data_df.csv\",sep=\"\\t\")\n",
    "data.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "data['REVIEW_TEXT'] = data['REVIEW_TEXT'].apply(lambda x:\" \".join(eval(x)))\n",
    "data.drop('PRODUCT_CATEGORY',axis=1,inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "data['RATING'] = data['RATING'].apply(lambda x: int(x>3))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "data[\"LABEL\"] = data[\"LABEL\"].apply(lambda x: int(x==1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "_cell_guid": "79c7e3d0-c299-4dcb-8224-4455121ee9b0",
    "_uuid": "d629ff2d2480ee46fbb7e2d37f6b5fab8052498a"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 60.7 ms, sys: 27.3 ms, total: 88 ms\n",
      "Wall time: 87.7 ms\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "documents = []\n",
    "for _text in data.REVIEW_TEXT:    \n",
    "    documents.append((_text.split(\" \")))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "W2V_SIZE = 48\n",
    "W2V_WINDOW = 7\n",
    "W2V_EPOCH = 64\n",
    "W2V_MIN_COUNT = 5\n",
    "w2v_model = gensim.models.word2vec.Word2Vec(size=W2V_SIZE, \n",
    "                                            window=W2V_WINDOW, \n",
    "                                            min_count=W2V_MIN_COUNT, \n",
    "                                            workers=8)\n",
    "\n",
    "\n",
    "w2v_model.build_vocab(documents)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 1min 52s, sys: 623 ms, total: 1min 53s\n",
      "Wall time: 1min 2s\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(36888183, 43163392)"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "%%time\n",
    "w2v_model.train(documents, total_examples=len(documents), epochs=W2V_EPOCH)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[('terrible', 0.6950481534004211),\n",
       " ('horrible', 0.5735267996788025),\n",
       " ('poor', 0.569575309753418),\n",
       " ('awful', 0.5676127076148987),\n",
       " ('obviously', 0.5643770694732666),\n",
       " ('guess', 0.5598137974739075),\n",
       " ('waste', 0.517963171005249),\n",
       " ('skip', 0.5136400461196899),\n",
       " ('junk', 0.5127378106117249),\n",
       " ('worst', 0.5090329647064209)]"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "w2v_model.most_similar(\"bad\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "202.9610476190476"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.mean(data.REVIEW_TEXT.map(len))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "from keras.preprocessing.text import Tokenizer\n",
    "from keras.preprocessing.sequence import pad_sequences\n",
    "from keras.models import Model,Sequential\n",
    "from keras.layers import Input,concatenate,Activation, Dense, Dropout, Embedding, Flatten, Conv1D, MaxPooling1D,Bidirectional, LSTM\n",
    "from keras import utils\n",
    "from keras.callbacks import ReduceLROnPlateau, EarlyStopping\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Vocab Size is  27031\n"
     ]
    }
   ],
   "source": [
    "tokenizer = Tokenizer()\n",
    "tokenizer.fit_on_texts(data.REVIEW_TEXT)\n",
    "vocab_size = len(tokenizer.word_index)+1\n",
    "print('Vocab Size is ',vocab_size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "SEQUENCE_LENGTH = 180"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(27031, 48)\n"
     ]
    }
   ],
   "source": [
    "embedding_matrix = np.zeros((vocab_size , W2V_SIZE))\n",
    "for word , i in tokenizer.word_index.items():\n",
    "    if word in w2v_model.wv:\n",
    "        embedding_matrix[i] = w2v_model.wv[word]\n",
    "print(embedding_matrix.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1    10500\n",
       "0    10500\n",
       "Name: LABEL, dtype: int64"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data.LABEL.value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "ones = data[data[\"LABEL\"]==1]\n",
    "twos = data[data[\"LABEL\"]==0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(15000, 4)\n",
      "(1000, 4)\n"
     ]
    }
   ],
   "source": [
    "train_data = ones[:7500]\n",
    "train_data = train_data.append(twos[:7500])\n",
    "val_data = ones[7500:8000]\n",
    "val_data = val_data.append(twos[7500:8000])\n",
    "print(train_data.shape)\n",
    "print(val_data.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(5000, 4)"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_data = ones[8000:]\n",
    "test_data = test_data.append(twos[8000:])\n",
    "test_data.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "data_nlp = train_data[\"REVIEW_TEXT\"]\n",
    "data_meta_input = train_data.drop([\"REVIEW_TEXT\",\"LABEL\"],axis=1)\n",
    "data_op = train_data[\"LABEL\"]\n",
    "\n",
    "val_nlp = val_data[\"REVIEW_TEXT\"]\n",
    "val_meta_input = val_data.drop([\"REVIEW_TEXT\",\"LABEL\"],axis=1)\n",
    "val_op = val_data[\"LABEL\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_data_nlp = test_data[\"REVIEW_TEXT\"]\n",
    "test_data_meta_input = test_data.drop([\"REVIEW_TEXT\",\"LABEL\"],axis=1)\n",
    "test_data_op = test_data[\"LABEL\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 676 ms, sys: 8 ms, total: 684 ms\n",
      "Wall time: 684 ms\n"
     ]
    }
   ],
   "source": [
    "%%time \n",
    "x_data = pad_sequences(tokenizer.texts_to_sequences(data_nlp) , maxlen = SEQUENCE_LENGTH)\n",
    "y_data = data_op\n",
    "y_data = y_data.values.reshape(-1,1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 42.8 ms, sys: 10 µs, total: 42.8 ms\n",
      "Wall time: 42.3 ms\n"
     ]
    }
   ],
   "source": [
    "%%time \n",
    "vx_data = pad_sequences(tokenizer.texts_to_sequences(val_nlp) , maxlen = SEQUENCE_LENGTH)\n",
    "vy_data = val_op\n",
    "vy_data = vy_data.values.reshape(-1,1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 207 ms, sys: 4 ms, total: 211 ms\n",
      "Wall time: 209 ms\n"
     ]
    }
   ],
   "source": [
    "%%time \n",
    "testx = pad_sequences(tokenizer.texts_to_sequences(test_data_nlp) , maxlen = SEQUENCE_LENGTH)\n",
    "testy = test_data_op\n",
    "testy = testy.values.reshape(-1,1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "nlp_input = Input(shape=(SEQUENCE_LENGTH,)) \n",
    "                   \n",
    "emb=Embedding(vocab_size,W2V_SIZE,weights=[embedding_matrix],input_length=SEQUENCE_LENGTH)(nlp_input)\n",
    "nlp_out = LSTM(64)(emb)                         \n",
    "classifier1 = Dense(128, activation='relu')(nlp_out) \n",
    "dropout = Dropout(0.2)(classifier1) \n",
    "classifier2 = Dense(32, activation='relu')(dropout) \n",
    "output = Dense(1, activation='sigmoid')(classifier2) \n",
    "\n",
    "model = Model(inputs=[nlp_input], outputs=[output])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model_1\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_1 (InputLayer)         (None, 180)               0         \n",
      "_________________________________________________________________\n",
      "embedding_1 (Embedding)      (None, 180, 48)           1297488   \n",
      "_________________________________________________________________\n",
      "lstm_1 (LSTM)                (None, 64)                28928     \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 128)               8320      \n",
      "_________________________________________________________________\n",
      "dropout_1 (Dropout)          (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 32)                4128      \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 1)                 33        \n",
      "=================================================================\n",
      "Total params: 1,338,897\n",
      "Trainable params: 1,338,897\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 15000 samples, validate on 1000 samples\n",
      "Epoch 1/3\n",
      "15000/15000 [==============================] - 82s 5ms/step - loss: 0.6712 - accuracy: 0.5786 - val_loss: 0.6725 - val_accuracy: 0.5940\n",
      "Epoch 2/3\n",
      "15000/15000 [==============================] - 80s 5ms/step - loss: 0.6199 - accuracy: 0.6559 - val_loss: 0.7018 - val_accuracy: 0.5720\n",
      "Epoch 3/3\n",
      "15000/15000 [==============================] - 79s 5ms/step - loss: 0.5648 - accuracy: 0.7063 - val_loss: 0.7518 - val_accuracy: 0.5740\n"
     ]
    }
   ],
   "source": [
    "history = model.fit(x_data,y_data,batch_size=32,epochs=3,\n",
    "                    validation_data=(vx_data,vy_data))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 15000 samples, validate on 1000 samples\n",
      "Epoch 4/5\n",
      "15000/15000 [==============================] - 79s 5ms/step - loss: 0.5010 - accuracy: 0.7547 - val_loss: 0.8087 - val_accuracy: 0.5810\n",
      "Epoch 5/5\n",
      "15000/15000 [==============================] - 79s 5ms/step - loss: 0.4198 - accuracy: 0.8066 - val_loss: 0.8032 - val_accuracy: 0.5760\n"
     ]
    }
   ],
   "source": [
    "history2 = model.fit(x_data,y_data,batch_size=32,epochs=5,initial_epoch=3,\n",
    "                    validation_data=(vx_data,vy_data))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.save(\"Word2Vec.hdf5\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.metrics import confusion_matrix"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "pred = model.predict(testx)\n",
    "# np.unique(pred,return_counts=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "pred = (pred>0.5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([0, 1]), array([2314, 2686]))"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred = pred.reshape(5000).astype(int)\n",
    "np.unique(pred , return_counts=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[1344,  970],\n",
       "       [1156, 1530]])"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "confusion_matrix(pred,testy)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AUC score is:  0.5748000000000001\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.metrics import roc_curve\n",
    "fpr, tpr, thresholds = roc_curve(testy, pred)\n",
    "\n",
    "# find threshold closest to zero:\n",
    "close_zero = np.argmin(np.abs(thresholds))\n",
    "\n",
    "plt.plot(fpr[close_zero], tpr[close_zero], 'o', \n",
    "         markersize=10, label=\"threshold zero(default)\", fillstyle=\"none\", c='k', mew=2)\n",
    "plt.plot([0,1], linestyle='-', lw=2, color='r', label='random', alpha=0.8)\n",
    "plt.legend(loc=4)\n",
    "plt.plot(fpr, tpr, label=\"ROC Curve\")\n",
    "plt.xlabel(\"False Positive Rate\")\n",
    "plt.ylabel(\"True Positive Rate (recall)\")\n",
    "plt.title(\"roc_curve\");\n",
    "\n",
    "from sklearn.metrics import auc\n",
    "print(\"AUC score is: \", auc(fpr, tpr));"
   ]
  },
  {
   "cell_type": "markdown",
   "execution_count": null,
   "metadata": {},
   "source": [
    "**CONCATENATED MODEL**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "nlp_input = Input(shape=(SEQUENCE_LENGTH,)) \n",
    "meta_input = Input(shape=(2,))\n",
    "                   \n",
    "emb=Embedding(vocab_size,W2V_SIZE,weights=[embedding_matrix],input_length=SEQUENCE_LENGTH,trainable = False)(nlp_input)\n",
    "nlp_out = LSTM(128)(emb) \n",
    "\n",
    "concat = concatenate([nlp_out, meta_input]) \n",
    "                        \n",
    "classifier = Dense(128, activation='relu')(concat) \n",
    "\n",
    "output = Dense(1, activation='sigmoid')(classifier) \n",
    "\n",
    "model2 = Model(inputs=[nlp_input , meta_input], outputs=[output])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "model2.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model_2\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_2 (InputLayer)            (None, 180)          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "embedding_2 (Embedding)         (None, 180, 48)      1297488     input_2[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "lstm_2 (LSTM)                   (None, 128)          90624       embedding_2[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "input_3 (InputLayer)            (None, 2)            0                                            \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_1 (Concatenate)     (None, 130)          0           lstm_2[0][0]                     \n",
      "                                                                 input_3[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dense_4 (Dense)                 (None, 128)          16768       concatenate_1[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "dense_5 (Dense)                 (None, 1)            129         dense_4[0][0]                    \n",
      "==================================================================================================\n",
      "Total params: 1,405,009\n",
      "Trainable params: 107,521\n",
      "Non-trainable params: 1,297,488\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model2.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 15000 samples, validate on 1000 samples\n",
      "Epoch 1/3\n",
      "15000/15000 [==============================] - 76s 5ms/step - loss: 0.5229 - accuracy: 0.7641 - val_loss: 0.5392 - val_accuracy: 0.7590\n",
      "Epoch 2/3\n",
      "15000/15000 [==============================] - 77s 5ms/step - loss: 0.4404 - accuracy: 0.8115 - val_loss: 0.5191 - val_accuracy: 0.7590\n",
      "Epoch 3/3\n",
      "15000/15000 [==============================] - 77s 5ms/step - loss: 0.3899 - accuracy: 0.8360 - val_loss: 0.5453 - val_accuracy: 0.7620\n"
     ]
    }
   ],
   "source": [
    "historyx = model2.fit([x_data,data_meta_input],y_data,batch_size=32,epochs=3,\n",
    "                    validation_data=([vx_data,val_meta_input],vy_data))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 15000 samples, validate on 1000 samples\n",
      "Epoch 4/5\n",
      "15000/15000 [==============================] - 76s 5ms/step - loss: 0.3377 - accuracy: 0.8579 - val_loss: 0.5798 - val_accuracy: 0.7510\n",
      "Epoch 5/5\n",
      "15000/15000 [==============================] - 76s 5ms/step - loss: 0.2741 - accuracy: 0.8876 - val_loss: 0.6413 - val_accuracy: 0.7510\n"
     ]
    }
   ],
   "source": [
    "historyx = model2.fit([x_data,data_meta_input],y_data,batch_size=32,epochs=5,initial_epoch=3,\n",
    "                    validation_data=([vx_data,val_meta_input],vy_data))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "model2.save(\"Word2Vec_V2.hdf5\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "pred = model2.predict([testx,test_data_meta_input])\n",
    "pred = (pred>0.5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([0, 1]), array([2713, 2287]))"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred = pred.reshape(5000).astype(int)\n",
    "np.unique(pred , return_counts=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[2002,  711],\n",
       "       [ 498, 1789]])"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "confusion_matrix(pred,testy)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AUC score is:  0.7582\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.metrics import roc_curve\n",
    "fpr, tpr, thresholds = roc_curve(testy, pred)\n",
    "\n",
    "# find threshold closest to zero:\n",
    "close_zero = np.argmin(np.abs(thresholds))\n",
    "\n",
    "plt.plot(fpr[close_zero], tpr[close_zero], 'o', \n",
    "         markersize=10, label=\"threshold zero(default)\", fillstyle=\"none\", c='k', mew=2)\n",
    "plt.plot([0,1], linestyle='-', lw=2, color='r', label='random', alpha=0.8)\n",
    "plt.legend(loc=4)\n",
    "plt.plot(fpr, tpr, label=\"ROC Curve\")\n",
    "plt.xlabel(\"False Positive Rate\")\n",
    "plt.ylabel(\"True Positive Rate (recall)\")\n",
    "plt.title(\"roc_curve\");\n",
    "\n",
    "from sklearn.metrics import auc\n",
    "print(\"AUC score is: \", auc(fpr, tpr));"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}