Project08.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "710e87d1",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.3.3\n",
      "0.5.2\n",
      "4.1.2\n"
     ]
    }
   ],
   "source": [
    "import pandas\n",
    "import konlpy\n",
    "import gensim\n",
    "\n",
    "print(pandas.__version__)\n",
    "print(konlpy.__version__)\n",
    "print(gensim.__version__)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "1b8c9500",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>id</th>\n",
       "      <th>document</th>\n",
       "      <th>label</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>9976970</td>\n",
       "      <td>아 더빙.. 진짜 짜증나네요 목소리</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>3819312</td>\n",
       "      <td>흠...포스터보고 초딩영화줄....오버연기조차 가볍지 않구나</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>10265843</td>\n",
       "      <td>너무재밓었다그래서보는것을추천한다</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>9045019</td>\n",
       "      <td>교도소 이야기구먼 ..솔직히 재미는 없다..평점 조정</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>6483659</td>\n",
       "      <td>사이몬페그의 익살스런 연기가 돋보였던 영화!스파이더맨에서 늙어보이기만 했던 커스틴 ...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         id                                           document  label\n",
       "0   9976970                                아 더빙.. 진짜 짜증나네요 목소리      0\n",
       "1   3819312                  흠...포스터보고 초딩영화줄....오버연기조차 가볍지 않구나      1\n",
       "2  10265843                                  너무재밓었다그래서보는것을추천한다      0\n",
       "3   9045019                      교도소 이야기구먼 ..솔직히 재미는 없다..평점 조정      0\n",
       "4   6483659  사이몬페그의 익살스런 연기가 돋보였던 영화!스파이더맨에서 늙어보이기만 했던 커스틴 ...      1"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import pandas as pd\n",
    "\n",
    "# 데이터. \n",
    "train_data = pd.read_table('~/aiffel/sentiment_classification/data/ratings_train.txt')\n",
    "test_data = pd.read_table('~/aiffel/sentiment_classification/data/ratings_test.txt')\n",
    "\n",
    "train_data.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "cb45daca",
   "metadata": {},
   "outputs": [],
   "source": [
    "from konlpy.tag import Mecab\n",
    "import numpy as np\n",
    "from tensorflow import keras\n",
    "from collections import Counter\n",
    "import matplotlib.pyplot as plt\n",
    "import urllib.request\n",
    "%matplotlib inline\n",
    "import os\n",
    "import re\n",
    "from gensim.models import KeyedVectors\n",
    "from gensim.models import Word2Vec\n",
    "from gensim.models.keyedvectors import Word2VecKeyedVectors\n",
    "from tensorflow.keras.initializers import Constant\n",
    "\n",
    "tokenizer = Mecab()\n",
    "stopwords = ['의','가','이','은','들','는','좀','잘','걍','과','도','를','으로','자','에','와','한','하다']\n",
    "\n",
    "def load_data(train_data, test_data, num_words=10000):\n",
    "    train_data.drop_duplicates(subset=['document'], inplace=True)\n",
    "    train_data = train_data.dropna(how = 'any') \n",
    "    test_data.drop_duplicates(subset=['document'], inplace=True)\n",
    "    test_data = test_data.dropna(how = 'any') \n",
    "\n",
    "    X_train = []\n",
    "    for sentence in train_data['document']:\n",
    "        temp_X = tokenizer.morphs(sentence) # 토큰화\n",
    "        temp_X = [word for word in temp_X if not word in stopwords] # 불용어 제거\n",
    "        X_train.append(temp_X)\n",
    "\n",
    "    X_test = []\n",
    "    for sentence in test_data['document']:\n",
    "        temp_X = tokenizer.morphs(sentence) # 토큰화\n",
    "        temp_X = [word for word in temp_X if not word in stopwords] # 불용어 제거\n",
    "        X_test.append(temp_X)\n",
    "    \n",
    "    words = np.concatenate(X_train).tolist()\n",
    "    counter = Counter(words)\n",
    "    counter = counter.most_common(10000-4)\n",
    "    vocab = ['<PAD>', '<BOS>', '<UNK>', '<UNUSED>'] + [key for key, _ in counter]\n",
    "    word_to_index = {word:index for index, word in enumerate(vocab)}\n",
    "\n",
    "    def wordlist_to_indexlist(wordlist):\n",
    "        return [word_to_index[word] if word in word_to_index else word_to_index['<UNK>'] for word in wordlist]\n",
    "\n",
    "    X_train = list(map(wordlist_to_indexlist, X_train))\n",
    "    X_test = list(map(wordlist_to_indexlist, X_test))\n",
    "\n",
    "    return X_train, np.array(list(train_data['label'])), X_test, np.array(list(test_data['label'])), word_to_index\n",
    "\n",
    "X_train, y_train, X_test, y_test, word_to_index = load_data(train_data, test_data)\n",
    "index_to_word = {index:word for word, index in word_to_index.items()}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "7a4b72b9",
   "metadata": {},
   "outputs": [],
   "source": [
    "index_to_word = {index:word for word, index in word_to_index.items()}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "193d4d6c",
   "metadata": {},
   "outputs": [],
   "source": [
    "# - \n",
    "# -- \n",
    "def get_encoded_sentence(sentence, word_to_index):\n",
    "    return [word_to_index['<BOS>']]+[word_to_index[word] if word in word_to_index else word_to_index['<UNK>'] for word in sentence.split()]\n",
    "\n",
    "# --\n",
    "def get_encoded_sentences(sentences, word_to_index):\n",
    "    return [get_encoded_sentence(sentence, word_to_index) for sentence in sentences]\n",
    "\n",
    "# -- \n",
    "def get_decoded_sentence(encoded_sentence, index_to_word):\n",
    "    return ' '.join(index_to_word[index] if index in index_to_word else '<UNK>' for index in encoded_sentence[1:])  #[1:]를 통해 <BOS>를 제외\n",
    "\n",
    "# -- \n",
    "def get_decoded_sentences(encoded_sentences, index_to_word):\n",
    "    return [get_decoded_sentence(encoded_sentence, index_to_word) for encoded_sentence in encoded_sentences]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "5072a4d8",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "문장길이 평균 :  15.96940191154864\n",
      "문장길이 최대 :  116\n",
      "문장길이 표준편차 :  12.843571191092\n",
      "pad_sequences maxlen :  41\n",
      "전체의 0.9342988343341575%가 maxlen값안에 포함됨. \n"
     ]
    }
   ],
   "source": [
    "#3) 모델 구성을 위한 데이터 분석 및 가공 \n",
    "\n",
    "total_data_text = list(X_train) + list(X_test)\n",
    "\n",
    "num_tokens = [len(tokens) for tokens in total_data_text]\n",
    "num_tokens = np.array(num_tokens)\n",
    "\n",
    "print('문장길이 평균 : ', np.mean(num_tokens))\n",
    "print('문장길이 최대 : ', np.max(num_tokens))\n",
    "print('문장길이 표준편차 : ', np.std(num_tokens))\n",
    "\n",
    "max_tokens = np.mean(num_tokens) + 2 * np.std(num_tokens)\n",
    "maxlen = int(max_tokens) # 최대 길이\n",
    "print('pad_sequences maxlen : ', maxlen)\n",
    "print('전체 문장의 {}%가 maxlen 설정값 이내에 포함됨. '.format(np.sum(num_tokens < max_tokens) / len(num_tokens)))\n",
    "\n",
    "# 패딩 추가\n",
    "# -\n",
    "X_train = keras.preprocessing.sequence.pad_sequences(X_train,\n",
    "                                                        value=word_to_index[\"<PAD>\"],\n",
    "                                                        padding='pre',\n",
    "                                                        maxlen=maxlen)\n",
    "\n",
    "X_test = keras.preprocessing.sequence.pad_sequences(X_test,\n",
    "                                                       value=word_to_index[\"<PAD>\"],\n",
    "                                                       padding='pre',\n",
    "                                                       maxlen=maxlen)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "66912366",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(126182, 41)\n",
      "(126182,)\n",
      "Model: \"sequential\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding (Embedding)        (None, None, 200)         2000000   \n",
      "_________________________________________________________________\n",
      "lstm (LSTM)                  (None, 8)                 6688      \n",
      "_________________________________________________________________\n",
      "dense (Dense)                (None, 8)                 72        \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 1)                 9         \n",
      "=================================================================\n",
      "Total params: 2,006,769\n",
      "Trainable params: 2,006,769\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "Model: \"sequential_1\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_1 (Embedding)      (None, None, 200)         2000000   \n",
      "_________________________________________________________________\n",
      "conv1d (Conv1D)              (None, None, 16)          22416     \n",
      "_________________________________________________________________\n",
      "max_pooling1d (MaxPooling1D) (None, None, 16)          0         \n",
      "_________________________________________________________________\n",
      "conv1d_1 (Conv1D)            (None, None, 16)          1808      \n",
      "_________________________________________________________________\n",
      "global_max_pooling1d (Global (None, 16)                0         \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 16)                272       \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 1)                 17        \n",
      "=================================================================\n",
      "Total params: 2,024,513\n",
      "Trainable params: 2,024,513\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "Model: \"sequential_2\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_2 (Embedding)      (None, None, 200)         2000000   \n",
      "_________________________________________________________________\n",
      "global_average_pooling1d (Gl (None, 200)               0         \n",
      "_________________________________________________________________\n",
      "dense_4 (Dense)              (None, 8)                 1608      \n",
      "_________________________________________________________________\n",
      "dense_5 (Dense)              (None, 1)                 9         \n",
      "=================================================================\n",
      "Total params: 2,001,617\n",
      "Trainable params: 2,001,617\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "#4) 모델 구성 및 validation set 구성\n",
    "# 학습 데이터 중 20000개를 validation set로 use!!!\n",
    "X_val = X_train[:20000]   \n",
    "y_val = y_train[:20000]\n",
    "\n",
    "partial_X_train = X_train[20000:]  \n",
    "partial_y_train = y_train[20000:]\n",
    "\n",
    "print(partial_X_train.shape)\n",
    "print(partial_y_train.shape)\n",
    "\n",
    "#모델 구성\n",
    "vocab_size = 10000\n",
    "word_vector_dim = 200\n",
    "\n",
    "# LSTM\n",
    "lstm = keras.Sequential()\n",
    "lstm.add(keras.layers.Embedding(vocab_size, word_vector_dim, input_shape=(None,)))\n",
    "lstm.add(keras.layers.LSTM(8))\n",
    "lstm.add(keras.layers.Dense(8, activation='relu'))\n",
    "lstm.add(keras.layers.Dense(1, activation='sigmoid'))\n",
    "\n",
    "lstm.summary()\n",
    "\n",
    "# 1-D CNN\n",
    "cnn = keras.Sequential()\n",
    "cnn.add(keras.layers.Embedding(vocab_size, word_vector_dim, input_shape=(None,))) \n",
    "cnn.add(keras.layers.Conv1D(16, 7, activation='relu'))\n",
    "cnn.add(keras.layers.MaxPooling1D(5))\n",
    "cnn.add(keras.layers.Conv1D(16, 7, activation='relu'))\n",
    "cnn.add(keras.layers.GlobalMaxPooling1D())\n",
    "cnn.add(keras.layers.Dense(16, activation='relu'))\n",
    "cnn.add(keras.layers.Dense(1, activation='sigmoid'))\n",
    "\n",
    "cnn.summary()\n",
    "\n",
    "# one GlobalAveragePooling layer\n",
    "simple = keras.Sequential()\n",
    "simple.add(keras.layers.Embedding(vocab_size, word_vector_dim, input_shape=(None,)))\n",
    "simple.add(keras.layers.GlobalAveragePooling1D())\n",
    "simple.add(keras.layers.Dense(8, activation='relu'))\n",
    "simple.add(keras.layers.Dense(1, activation='sigmoid'))\n",
    "\n",
    "simple.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "b0bd2382",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/5\n",
      "247/247 [==============================] - 5s 9ms/step - loss: 0.4489 - accuracy: 0.7997 - val_loss: 0.3466 - val_accuracy: 0.8483\n",
      "Epoch 2/5\n",
      "247/247 [==============================] - 2s 7ms/step - loss: 0.3221 - accuracy: 0.8624 - val_loss: 0.3396 - val_accuracy: 0.8536\n",
      "Epoch 3/5\n",
      "247/247 [==============================] - 2s 7ms/step - loss: 0.2934 - accuracy: 0.8757 - val_loss: 0.3434 - val_accuracy: 0.8546\n",
      "Epoch 4/5\n",
      "247/247 [==============================] - 2s 7ms/step - loss: 0.2730 - accuracy: 0.8848 - val_loss: 0.3476 - val_accuracy: 0.8530\n",
      "Epoch 5/5\n",
      "247/247 [==============================] - 2s 7ms/step - loss: 0.2533 - accuracy: 0.8944 - val_loss: 0.3612 - val_accuracy: 0.8514\n",
      "Epoch 1/3\n",
      "247/247 [==============================] - 4s 8ms/step - loss: 0.4466 - accuracy: 0.7897 - val_loss: 0.3420 - val_accuracy: 0.8521\n",
      "Epoch 2/3\n",
      "247/247 [==============================] - 2s 7ms/step - loss: 0.3066 - accuracy: 0.8715 - val_loss: 0.3276 - val_accuracy: 0.8594\n",
      "Epoch 3/3\n",
      "247/247 [==============================] - 2s 6ms/step - loss: 0.2552 - accuracy: 0.8977 - val_loss: 0.3322 - val_accuracy: 0.8585\n",
      "Epoch 1/3\n",
      "247/247 [==============================] - 2s 5ms/step - loss: 0.5468 - accuracy: 0.7637 - val_loss: 0.3993 - val_accuracy: 0.8331\n",
      "Epoch 2/3\n",
      "247/247 [==============================] - 1s 4ms/step - loss: 0.3683 - accuracy: 0.8446 - val_loss: 0.3570 - val_accuracy: 0.8457\n",
      "Epoch 3/3\n",
      "247/247 [==============================] - 1s 4ms/step - loss: 0.3381 - accuracy: 0.8569 - val_loss: 0.3495 - val_accuracy: 0.8485\n",
      "LSTM\n",
      "1537/1537 - 3s - loss: 0.3701 - accuracy: 0.8470\n",
      "[0.3701038956642151, 0.8470004200935364]\n",
      "-------------------------------------------------------------------------\n",
      "1-D CNN\n",
      "1537/1537 - 3s - loss: 0.3426 - accuracy: 0.8539\n",
      "[0.34256982803344727, 0.8538763523101807]\n",
      "-------------------------------------------------------------------------\n"
     ]
    }
   ],
   "source": [
    "#5)모델 훈련 \n",
    "epochs = 5\n",
    "\n",
    "lstm.compile(optimizer='adam',\n",
    "              loss='binary_crossentropy',\n",
    "              metrics=['accuracy'])\n",
    "\n",
    "history_lstm = lstm.fit(partial_X_train,\n",
    "                    partial_y_train,\n",
    "                    epochs=epochs,\n",
    "                    batch_size=512,\n",
    "                    validation_data=(X_val, y_val),\n",
    "                    verbose=1)\n",
    "              \n",
    "epochs = 3\n",
    "\n",
    "cnn.compile(optimizer='adam',\n",
    "              loss='binary_crossentropy',\n",
    "              metrics=['accuracy'])\n",
    "\n",
    "\n",
    "history_cnn = cnn.fit(partial_X_train,\n",
    "                    partial_y_train,\n",
    "                    epochs=epochs,\n",
    "                    batch_size=512,\n",
    "                    validation_data=(X_val, y_val),\n",
    "                    verbose=1)\n",
    "\n",
    "# 1-D CNN과 동일하게 3 epochs로 진행\n",
    "simple.compile(optimizer='adam',\n",
    "              loss='binary_crossentropy',\n",
    "              metrics=['accuracy'])\n",
    "\n",
    "history_simple = simple.fit(partial_X_train,\n",
    "                    partial_y_train,\n",
    "                    epochs=epochs,\n",
    "                    batch_size=512,\n",
    "                    validation_data=(X_val, y_val),\n",
    "                    verbose=1)\n",
    "\n",
    "#모델 평가\n",
    "# LSTM\n",
    "print(\"LSTM\")\n",
    "results_lstm = lstm.evaluate(X_test, y_test, verbose=2)\n",
    "print(results_lstm)\n",
    "print(\"-------------------------------------------------------------------------\")\n",
    "\n",
    "# 1-D CNN\n",
    "print(\"1-D CNN\")\n",
    "results_cnn = cnn.evaluate(X_test, y_test, verbose=2)\n",
    "print(results_cnn)\n",
    "print(\"-------------------------------------------------------------------------\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "55137bd1",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "GlobalAveragePooling1D layer\n",
      "1537/1537 - 2s - loss: 0.3618 - accuracy: 0.8456\n",
      "[0.3618220388889313, 0.8456171154975891]\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"
    },
    {
     "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": [
    "#6-1) Loss, Accuracy 그래프 시각화\n",
    "print(\"GlobalAveragePooling1D layer\")\n",
    "results_simple = simple.evaluate(X_test, y_test, verbose=2)\n",
    "print(results_simple)\n",
    "\n",
    "#6-2) Loss, Accuracy 그래프 시각화\n",
    "history_dict = history_lstm.history\n",
    "\n",
    "acc = history_dict['accuracy']\n",
    "val_acc = history_dict['val_accuracy']\n",
    "loss = history_dict['loss']\n",
    "val_loss = history_dict['val_loss']\n",
    "\n",
    "epochs = range(1, len(acc) + 1)\n",
    "\n",
    "plt.plot(epochs, loss, 'bo', label='Training loss')\n",
    "plt.plot(epochs, val_loss, 'b', label='Validation loss')\n",
    "\n",
    "plt.title('Training and validation loss')\n",
    "plt.xlabel('Epochs')\n",
    "plt.ylabel('Loss')\n",
    "plt.legend()\n",
    "\n",
    "plt.show()\n",
    "\n",
    "# 그림을 초기화\n",
    "plt.clf()\n",
    "\n",
    "plt.plot(epochs, acc, 'bo', label='Training acc')\n",
    "plt.plot(epochs, val_acc, 'b', label='Validation acc')\n",
    "plt.title('Training and validation accuracy')\n",
    "plt.xlabel('Epochs')\n",
    "plt.ylabel('Accuracy')\n",
    "plt.legend()\n",
    "\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "763c9445",
   "metadata": {},
   "outputs": [],
   "source": [
    "#7) Embedding레이어분석\n",
    "\n",
    "embedding_lstm = lstm.layers[0]\n",
    "weights_lstm = embedding_lstm.get_weights()[0]\n",
    "\n",
    "embedding_cnn = cnn.layers[0]\n",
    "weights_cnn = embedding_cnn.get_weights()[0]\n",
    "\n",
    "embedding_simple = simple.layers[0]\n",
    "weights_simple = embedding_simple.get_weights()[0]\n",
    "\n",
    "word2vec_file_path_lstm = os.getenv('HOME')+'/aiffel/sentiment_classification/word2vec_lstm.txt'\n",
    "f = open(word2vec_file_path_lstm, 'w')\n",
    "f.write('{} {}\\n'.format(vocab_size-4, word_vector_dim))\n",
    "\n",
    "vectors = lstm.get_weights()[0]\n",
    "for i in range(4,vocab_size):\n",
    "    f.write('{} {}\\n'.format(index_to_word[i], ' '.join(map(str, list(vectors[i, :])))))\n",
    "f.close()\n",
    "\n",
    "word2vec_file_path_cnn = os.getenv('HOME')+'/aiffel/sentiment_classification/word2vec_cnn.txt'\n",
    "f = open(word2vec_file_path_cnn, 'w')\n",
    "f.write('{} {}\\n'.format(vocab_size-4, word_vector_dim))\n",
    "\n",
    "vectors = cnn.get_weights()[0]\n",
    "for i in range(4,vocab_size):\n",
    "    f.write('{} {}\\n'.format(index_to_word[i], ' '.join(map(str, list(vectors[i, :])))))\n",
    "f.close()\n",
    "\n",
    "word2vec_file_path_simple = os.getenv('HOME')+'/aiffel/sentiment_classification/word2vec_simple.txt'\n",
    "f = open(word2vec_file_path_simple, 'w')\n",
    "f.write('{} {}\\n'.format(vocab_size-4, word_vector_dim))\n",
    "\n",
    "vectors = simple.get_weights()[0]\n",
    "for i in range(4,vocab_size):\n",
    "    f.write('{} {}\\n'.format(index_to_word[i], ' '.join(map(str, list(vectors[i, :])))))\n",
    "f.close()\n",
    "\n",
    "word_vectors_lstm = Word2VecKeyedVectors.load_word2vec_format(word2vec_file_path_lstm, binary=False)\n",
    "\n",
    "word_vectors_cnn = Word2VecKeyedVectors.load_word2vec_format(word2vec_file_path_cnn, binary=False)\n",
    "\n",
    "word_vectors_simple = Word2VecKeyedVectors.load_word2vec_format(word2vec_file_path_simple, binary=False)\n"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.9.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}