Ex10.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "fd09bde9",
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import re\n",
    "from tensorflow.keras.preprocessing.text import Tokenizer \n",
    "from tensorflow.keras.preprocessing.sequence import pad_sequences\n",
    "from tensorflow.keras.utils import to_categorical\n",
    "import numpy as np\n",
    "import os\n",
    "import unicodedata\n",
    "os.environ['KMP_DUPLICATE_LIB_OK']='True'"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "64925c82",
   "metadata": {},
   "source": [
    "## 데이터 불러오기"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "3d7f2d83",
   "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>eng</th>\n",
       "      <th>fra</th>\n",
       "      <th>cc</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Go.</td>\n",
       "      <td>Va !</td>\n",
       "      <td>CC-BY 2.0 (France) Attribution: tatoeba.org #2...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Go.</td>\n",
       "      <td>Marche.</td>\n",
       "      <td>CC-BY 2.0 (France) Attribution: tatoeba.org #2...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Go.</td>\n",
       "      <td>En route !</td>\n",
       "      <td>CC-BY 2.0 (France) Attribution: tatoeba.org #2...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Go.</td>\n",
       "      <td>Bouge !</td>\n",
       "      <td>CC-BY 2.0 (France) Attribution: tatoeba.org #2...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Hi.</td>\n",
       "      <td>Salut !</td>\n",
       "      <td>CC-BY 2.0 (France) Attribution: tatoeba.org #5...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   eng         fra                                                 cc\n",
       "0  Go.        Va !  CC-BY 2.0 (France) Attribution: tatoeba.org #2...\n",
       "1  Go.     Marche.  CC-BY 2.0 (France) Attribution: tatoeba.org #2...\n",
       "2  Go.  En route !  CC-BY 2.0 (France) Attribution: tatoeba.org #2...\n",
       "3  Go.     Bouge !  CC-BY 2.0 (France) Attribution: tatoeba.org #2...\n",
       "4  Hi.     Salut !  CC-BY 2.0 (France) Attribution: tatoeba.org #5..."
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "file_path = 'C:/users/juwon/practice_aiffel/translator_seq2seq/data/fra.txt'\n",
    "lines = pd.read_csv(file_path, names = ['eng', 'fra','cc'], sep='\\t')\n",
    "lines.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "a7088dcf",
   "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>eng</th>\n",
       "      <th>fra</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>3463</th>\n",
       "      <td>I'm relaxed.</td>\n",
       "      <td>Je suis détendue.</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10287</th>\n",
       "      <td>I am in a spot.</td>\n",
       "      <td>Je suis dans un endroit.</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8328</th>\n",
       "      <td>People change.</td>\n",
       "      <td>Les gens changent.</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17395</th>\n",
       "      <td>Are you a golfer?</td>\n",
       "      <td>Es-tu golfeur ?</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>27357</th>\n",
       "      <td>We've got to talk.</td>\n",
       "      <td>Nous devons parler.</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                      eng                       fra\n",
       "3463         I'm relaxed.         Je suis détendue.\n",
       "10287     I am in a spot.  Je suis dans un endroit.\n",
       "8328       People change.        Les gens changent.\n",
       "17395   Are you a golfer?           Es-tu golfeur ?\n",
       "27357  We've got to talk.       Nous devons parler."
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "lines = lines[['eng', 'fra']][:33000]\n",
    "lines.sample(5)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "93a372cc",
   "metadata": {},
   "source": [
    "## 문장 전처리"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "3b1c5db1",
   "metadata": {},
   "outputs": [],
   "source": [
    "def to_ascii(sentence):\n",
    "    return ''.join(c for c in unicodedata.normalize('NFD', sentence) if unicodedata.category(c) != 'Mn')\n",
    "\n",
    "# https://wikidocs.net/86900 참고"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "7071eb80",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'allez y !'"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "to_ascii(lines.fra[551])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "0e229f8c",
   "metadata": {},
   "outputs": [],
   "source": [
    "def preprocessing_sentence(sentence):\n",
    "    to_ascii(sentence)\n",
    "    sentence = sentence.lower()\n",
    "    sentence = re.sub(r'([?.!,¿-])', r\" \\1 \", sentence)\n",
    "    sentence = re.sub(r'[\" \"]+' , \" \", sentence)\n",
    "    sentence = re.sub(r'[^a-zA-Z?.!,¿]+', r\" \", sentence)\n",
    "    sentence = sentence.strip()\n",
    "    return sentence"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "2952f54b",
   "metadata": {},
   "outputs": [],
   "source": [
    "lines.eng = lines.eng.apply(preprocessing_sentence)\n",
    "lines.fra = lines.fra.apply(preprocessing_sentence)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "26eaf1eb",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[['go', '.'], ['go', '.'], ['go', '.']]"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "input_text = [a.split() for a in lines.eng]\n",
    "\n",
    "encoder_input = input_text\n",
    "input_text[:3]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "3e4ed9a6",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[['<sos>', 'va', '!', '<eos>'], ['<sos>', 'marche', '.', '<eos>'], ['<sos>', 'en', 'route', '!', '<eos>']]\n",
      "[['<sos>', 'va', '!'], ['<sos>', 'marche', '.'], ['<sos>', 'en', 'route', '!']]\n",
      "[['va', '!', '<eos>'], ['marche', '.', '<eos>'], ['en', 'route', '!', '<eos>']]\n"
     ]
    }
   ],
   "source": [
    "output_text = [('<sos> '+a+' <eos>').split() for a in lines.fra]\n",
    "decoder_input = [a[:-1] for a in output_text]\n",
    "decoder_target = [a[1:] for a in output_text]\n",
    "print(output_text[:3])\n",
    "print(decoder_input[:3])\n",
    "print(decoder_target[:3])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 125,
   "id": "f94aabbe",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['go', 'ahead', '!']\n",
      "['<sos>', 'allez', 'y', '!', '<eos>']\n"
     ]
    }
   ],
   "source": [
    "print(input_text[551])\n",
    "print(output_text[551])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "0a62d722",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[[28, 1], [28, 1], [28, 1], [28, 1], [762, 1]]"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "eng_tokenizer = Tokenizer()\n",
    "eng_tokenizer.fit_on_texts(encoder_input)\n",
    "encoder_input = eng_tokenizer.texts_to_sequences(encoder_input)\n",
    "encoder_input[:5]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1cc98991",
   "metadata": {},
   "source": [
    "## 문장 토큰화 및 패딩"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "b19cf1b8",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[[1, 76, 8], [1, 366, 3], [1, 28, 513, 8], [1, 717, 8], [1, 751, 8]]"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "fra_tokenizer = Tokenizer(lower=False)\n",
    "fra_tokenizer.fit_on_texts(output_text)\n",
    "decoder_input = fra_tokenizer.texts_to_sequences(decoder_input)\n",
    "decoder_input[:5]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "64c95dad",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[[76, 8, 2], [366, 3, 2], [28, 513, 8, 2], [717, 8, 2], [751, 8, 2]]"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "fra_tokenizer = Tokenizer(lower=False)\n",
    "fra_tokenizer.fit_on_texts(output_text)\n",
    "decoder_target = fra_tokenizer.texts_to_sequences(decoder_target)\n",
    "decoder_target[:5]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "2d67fd6f",
   "metadata": {},
   "outputs": [],
   "source": [
    "eng_voc_size = len(eng_tokenizer.word_index) + 1\n",
    "fra_voc_size = len(fra_tokenizer.word_index) + 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "1b7f6df0",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "영어 단어장 크기: 4672\n",
      "프랑스어 단어장 크기: 7455\n"
     ]
    }
   ],
   "source": [
    "print('영어 단어장 크기:',eng_voc_size)\n",
    "print('프랑스어 단어장 크기:',fra_voc_size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "b9d96f35",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "8\n",
      "16\n"
     ]
    }
   ],
   "source": [
    "maxlen_eng = max([len(a) for a in encoder_input])\n",
    "maxlen_fra = max([len(a) for a in decoder_input])\n",
    "print(maxlen_eng)\n",
    "print(maxlen_fra) # 시작, 종료토큰포함"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "f8aab1aa",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "영어 데이터의 크기(shape) : (33000, 8)\n",
      "프랑스어 입력데이터의 크기(shape) : (33000, 16)\n",
      "프랑스어 출력데이터의 크기(shape) : (33000, 16)\n"
     ]
    }
   ],
   "source": [
    "encoder_input = pad_sequences(encoder_input, padding='post', maxlen = maxlen_eng)\n",
    "decoder_input = pad_sequences(decoder_input, padding='post', maxlen = maxlen_fra)\n",
    "decoder_target = pad_sequences(decoder_target, padding ='post', maxlen = maxlen_fra)\n",
    "\n",
    "print('영어 데이터의 크기(shape) :',np.shape(encoder_input))\n",
    "print('프랑스어 입력데이터의 크기(shape) :',np.shape(decoder_input))\n",
    "print('프랑스어 출력데이터의 크기(shape) :',np.shape(decoder_target))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "567945a0",
   "metadata": {},
   "source": [
    "## 학습, 평가 데이터 분할"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "33991d18",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "영어 훈련 데이터 크기:  (30000, 8)\n",
      "프랑스 훈련 데이터 크기:  (30000, 16)\n",
      "프랑스 테스트 라벨 데이터 크기:  (3000, 16)\n"
     ]
    }
   ],
   "source": [
    "n_of_val = 3000\n",
    "\n",
    "encoder_input_train = encoder_input[:-n_of_val]\n",
    "decoder_input_train = decoder_input[:-n_of_val]\n",
    "decoder_target_train = decoder_target[:-n_of_val]\n",
    "\n",
    "encoder_input_test = encoder_input[-n_of_val:]\n",
    "decoder_input_test = decoder_input[-n_of_val:]\n",
    "decoder_target_test = decoder_target[-n_of_val:]\n",
    "\n",
    "print('영어 훈련 데이터 크기: ',encoder_input_train.shape)\n",
    "print('프랑스 훈련 데이터 크기: ',decoder_input_train.shape)\n",
    "print('프랑스 테스트 라벨 데이터 크기: ',decoder_target_test.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "9d03ef0a",
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.keras.layers import Input, LSTM, Embedding, Dense, Masking\n",
    "from tensorflow.keras.models import Model"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "28e9c082",
   "metadata": {},
   "source": [
    "## 인코더 설계"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 104,
   "id": "67b3a72e",
   "metadata": {},
   "outputs": [],
   "source": [
    "embbeding_dim = 64\n",
    "hidden_units = 128"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 105,
   "id": "7eb4500c",
   "metadata": {},
   "outputs": [],
   "source": [
    "encoder_inputs = Input(shape=(None,))\n",
    "encoder_embbeding = Embedding(eng_voc_size, embbeding_dim)(encoder_inputs)\n",
    "encoder_masking = Masking(mask_value = 0.0)(encoder_embbeding) \n",
    "encoder_lstm = LSTM(hidden_units, return_state = True)\n",
    "encoder_outputs, state_h, state_c =encoder_lstm(encoder_masking)\n",
    "encoder_states = [state_h, state_c]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "35016be3",
   "metadata": {},
   "source": [
    "## 디코더 입력 설계"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 106,
   "id": "3737a94e",
   "metadata": {},
   "outputs": [],
   "source": [
    "decoder_inputs = Input(shape=(None,))\n",
    "decoder_embbeding = Embedding(fra_voc_size, embbeding_dim)(decoder_inputs)\n",
    "decoder_masking = Masking(mask_value=0.0)(decoder_embbeding)\n",
    "decoder_lstm = LSTM(hidden_units, return_sequences = True, return_state = True)\n",
    "decoder_outputs, _, _ = decoder_lstm(decoder_masking, initial_state = encoder_states)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5d50f178",
   "metadata": {},
   "source": [
    "## 디코더 출력"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 107,
   "id": "27a2a4d8",
   "metadata": {},
   "outputs": [],
   "source": [
    "decoder_softmax_layer = Dense(fra_voc_size, activation ='softmax')\n",
    "decoder_outputs = decoder_softmax_layer(decoder_outputs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 108,
   "id": "bc7553ce",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model_15\"\n",
      "__________________________________________________________________________________________________\n",
      " Layer (type)                   Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      " input_21 (InputLayer)          [(None, None)]       0           []                               \n",
      "                                                                                                  \n",
      " input_22 (InputLayer)          [(None, None)]       0           []                               \n",
      "                                                                                                  \n",
      " embedding_10 (Embedding)       (None, None, 64)     299008      ['input_21[0][0]']               \n",
      "                                                                                                  \n",
      " embedding_11 (Embedding)       (None, None, 64)     477120      ['input_22[0][0]']               \n",
      "                                                                                                  \n",
      " masking_10 (Masking)           (None, None, 64)     0           ['embedding_10[0][0]']           \n",
      "                                                                                                  \n",
      " masking_11 (Masking)           (None, None, 64)     0           ['embedding_11[0][0]']           \n",
      "                                                                                                  \n",
      " lstm_10 (LSTM)                 [(None, 128),        98816       ['masking_10[0][0]']             \n",
      "                                 (None, 128),                                                     \n",
      "                                 (None, 128)]                                                     \n",
      "                                                                                                  \n",
      " lstm_11 (LSTM)                 [(None, None, 128),  98816       ['masking_11[0][0]',             \n",
      "                                 (None, 128),                     'lstm_10[0][1]',                \n",
      "                                 (None, 128)]                     'lstm_10[0][2]']                \n",
      "                                                                                                  \n",
      " dense_5 (Dense)                (None, None, 7455)   961695      ['lstm_11[0][0]']                \n",
      "                                                                                                  \n",
      "==================================================================================================\n",
      "Total params: 1,935,455\n",
      "Trainable params: 1,935,455\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model = Model([encoder_inputs, decoder_inputs], decoder_outputs)\n",
    "model.compile(optimizer ='rmsprop',\n",
    "             loss ='sparse_categorical_crossentropy')\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 109,
   "id": "2516faac",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/23\n",
      "235/235 [==============================] - 25s 49ms/step - loss: 2.2563 - val_loss: 1.9682\n",
      "Epoch 2/23\n",
      "235/235 [==============================] - 7s 32ms/step - loss: 1.5714 - val_loss: 1.8154\n",
      "Epoch 3/23\n",
      "235/235 [==============================] - 7s 31ms/step - loss: 1.4108 - val_loss: 1.7094\n",
      "Epoch 4/23\n",
      "235/235 [==============================] - 7s 30ms/step - loss: 1.2900 - val_loss: 1.6161\n",
      "Epoch 5/23\n",
      "235/235 [==============================] - 7s 31ms/step - loss: 1.2045 - val_loss: 1.5244\n",
      "Epoch 6/23\n",
      "235/235 [==============================] - 7s 31ms/step - loss: 1.1396 - val_loss: 1.4787\n",
      "Epoch 7/23\n",
      "235/235 [==============================] - 7s 30ms/step - loss: 1.0859 - val_loss: 1.4208\n",
      "Epoch 8/23\n",
      "235/235 [==============================] - 7s 30ms/step - loss: 1.0378 - val_loss: 1.4031\n",
      "Epoch 9/23\n",
      "235/235 [==============================] - 7s 30ms/step - loss: 0.9969 - val_loss: 1.3733\n",
      "Epoch 10/23\n",
      "235/235 [==============================] - 7s 32ms/step - loss: 0.9594 - val_loss: 1.3372\n",
      "Epoch 11/23\n",
      "235/235 [==============================] - 7s 32ms/step - loss: 0.9266 - val_loss: 1.3185\n",
      "Epoch 12/23\n",
      "235/235 [==============================] - 7s 31ms/step - loss: 0.8971 - val_loss: 1.3009\n",
      "Epoch 13/23\n",
      "235/235 [==============================] - 7s 31ms/step - loss: 0.8704 - val_loss: 1.2848\n",
      "Epoch 14/23\n",
      "235/235 [==============================] - 7s 31ms/step - loss: 0.8465 - val_loss: 1.2753\n",
      "Epoch 15/23\n",
      "235/235 [==============================] - 7s 32ms/step - loss: 0.8241 - val_loss: 1.2698\n",
      "Epoch 16/23\n",
      "235/235 [==============================] - 7s 31ms/step - loss: 0.8027 - val_loss: 1.2537\n",
      "Epoch 17/23\n",
      "235/235 [==============================] - 7s 31ms/step - loss: 0.7828 - val_loss: 1.2451\n",
      "Epoch 18/23\n",
      "235/235 [==============================] - 7s 30ms/step - loss: 0.7631 - val_loss: 1.2297\n",
      "Epoch 19/23\n",
      "235/235 [==============================] - 7s 31ms/step - loss: 0.7455 - val_loss: 1.2262\n",
      "Epoch 20/23\n",
      "235/235 [==============================] - 7s 32ms/step - loss: 0.7279 - val_loss: 1.2244\n",
      "Epoch 21/23\n",
      "235/235 [==============================] - 7s 30ms/step - loss: 0.7122 - val_loss: 1.2186\n",
      "Epoch 22/23\n",
      "235/235 [==============================] - 7s 31ms/step - loss: 0.6965 - val_loss: 1.2185\n",
      "Epoch 23/23\n",
      "235/235 [==============================] - 7s 31ms/step - loss: 0.6807 - val_loss: 1.2127\n"
     ]
    }
   ],
   "source": [
    "epochs = 23\n",
    "batch_size = 128 \n",
    "history = model.fit(x=[encoder_input_train, decoder_input_train], y = decoder_target_train,\n",
    "         validation_data =([encoder_input_test, decoder_input_test], decoder_target_test),\n",
    "          batch_size=batch_size, epochs = epochs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 110,
   "id": "a0510a2e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "train_loss = history.history['loss']\n",
    "val_loss = history.history['val_loss']\n",
    "epochs = range(1,epochs)\n",
    "plt.plot(epochs, train_loss,'b',label = 'Train_loss')\n",
    "plt.plot(epochs, val_loss,'r',label = 'Val_loss')\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "225ec203",
   "metadata": {},
   "source": [
    "## 모델 테스트하기"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 111,
   "id": "3dc17b91",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model_16\"\n",
      "_________________________________________________________________\n",
      " Layer (type)                Output Shape              Param #   \n",
      "=================================================================\n",
      " input_21 (InputLayer)       [(None, None)]            0         \n",
      "                                                                 \n",
      " embedding_10 (Embedding)    (None, None, 64)          299008    \n",
      "                                                                 \n",
      " masking_10 (Masking)        (None, None, 64)          0         \n",
      "                                                                 \n",
      " lstm_10 (LSTM)              [(None, 128),             98816     \n",
      "                              (None, 128),                       \n",
      "                              (None, 128)]                       \n",
      "                                                                 \n",
      "=================================================================\n",
      "Total params: 397,824\n",
      "Trainable params: 397,824\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "encoder_model = Model(inputs= encoder_inputs, outputs= encoder_states)\n",
    "encoder_model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 112,
   "id": "238118ea",
   "metadata": {},
   "outputs": [],
   "source": [
    "decoder_state_input_h = Input(shape=(hidden_units,))\n",
    "decoder_state_input_c = Input(shape=(hidden_units,))\n",
    "decoder_states_inputs = [decoder_state_input_h, decoder_state_input_c]\n",
    "\n",
    "decoder_outputs, state_h, state_c = decoder_lstm(decoder_embbeding, initial_state= decoder_states_inputs)\n",
    "decoder_states = [state_h, state_c]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 113,
   "id": "85eb03e0",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model_17\"\n",
      "__________________________________________________________________________________________________\n",
      " Layer (type)                   Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      " input_22 (InputLayer)          [(None, None)]       0           []                               \n",
      "                                                                                                  \n",
      " embedding_11 (Embedding)       (None, None, 64)     477120      ['input_22[0][0]']               \n",
      "                                                                                                  \n",
      " input_23 (InputLayer)          [(None, 128)]        0           []                               \n",
      "                                                                                                  \n",
      " input_24 (InputLayer)          [(None, 128)]        0           []                               \n",
      "                                                                                                  \n",
      " lstm_11 (LSTM)                 [(None, None, 128),  98816       ['embedding_11[0][0]',           \n",
      "                                 (None, 128),                     'input_23[0][0]',               \n",
      "                                 (None, 128)]                     'input_24[0][0]']               \n",
      "                                                                                                  \n",
      " dense_5 (Dense)                (None, None, 7455)   961695      ['lstm_11[1][0]']                \n",
      "                                                                                                  \n",
      "==================================================================================================\n",
      "Total params: 1,537,631\n",
      "Trainable params: 1,537,631\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "decoder_outputs = decoder_softmax_layer(decoder_outputs)\n",
    "decoder_model = Model(inputs = [decoder_inputs]+ decoder_states_inputs, outputs=[decoder_outputs]+decoder_states)\n",
    "decoder_model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 114,
   "id": "50916969",
   "metadata": {},
   "outputs": [],
   "source": [
    "eng2idx = eng_tokenizer.word_index\n",
    "fra2idx = fra_tokenizer.word_index\n",
    "idx2eng = eng_tokenizer.index_word\n",
    "idx2fra = fra_tokenizer.index_word"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 115,
   "id": "73cb818c",
   "metadata": {},
   "outputs": [],
   "source": [
    "def decode_sequence(input_seq): \n",
    "    state_values = encoder_model.predict(input_seq,verbose=0)\n",
    "    target_seq = np.zeros((1,1))\n",
    "    target_seq[0,0] = fra2idx['<sos>']\n",
    "    \n",
    "    stop_condition = False\n",
    "    decoded_sentence = ''\n",
    "    \n",
    "    while not stop_condition:\n",
    "        output_tokens, h, c = decoder_model.predict([target_seq] + state_values,verbose=0)\n",
    "        \n",
    "        sampled_token_index = np.argmax(output_tokens[0,-1,:])\n",
    "        sampled_seq = idx2fra[sampled_token_index]\n",
    "        \n",
    "        decoded_sentence += sampled_seq+ ' '\n",
    "        \n",
    "        if (sampled_seq=='<eos>' or len(decoded_sentence.split()) > maxlen_fra) : \n",
    "            stop_condition = True\n",
    "            \n",
    "        target_seq = np.zeros((1,1))\n",
    "        target_seq[0, 0] = sampled_token_index\n",
    "        \n",
    "        state_values= [h,c]\n",
    "    return decoded_sentence"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "03790bef",
   "metadata": {},
   "source": [
    "## 비교해보기\n",
    "### 입력된 문장 출력"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 116,
   "id": "dab3bdcb",
   "metadata": {},
   "outputs": [],
   "source": [
    "def seq2input(input_seq):\n",
    "    sentence = ''\n",
    "    for token in input_seq :\n",
    "        if (token != 0) :\n",
    "            sentence += idx2eng[token]+ ' '\n",
    "    return sentence"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "768b37ee",
   "metadata": {},
   "source": [
    "### 정답 문장 출력"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 117,
   "id": "9c044d02",
   "metadata": {},
   "outputs": [],
   "source": [
    "def seq2target(input_seq):\n",
    "    sentence =''\n",
    "    for token in input_seq :\n",
    "        if (token !=0 and token != fra2idx['<sos>'] and token != fra2idx['<eos>']) :\n",
    "            sentence += idx2fra[token] + ' '\n",
    "    return sentence"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 126,
   "id": "17297419",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "입력된 문장:  go . \n",
      "정답 문장 :  en route ! \n",
      "번역한 문장:  va y ! \n",
      "--------------------------------------------------\n",
      "입력된 문장:  i left . \n",
      "정답 문장 :  je suis partie . \n",
      "번역한 문장:  j ai eu de l eau . \n",
      "--------------------------------------------------\n",
      "입력된 문장:  bury it . \n",
      "정답 문장 :  enterrez la . \n",
      "번역한 문장:  ceci . \n",
      "--------------------------------------------------\n",
      "입력된 문장:  go ahead ! \n",
      "정답 문장 :  allez y ! \n",
      "번역한 문장:  allez ! \n",
      "--------------------------------------------------\n",
      "입력된 문장:  here i am . \n",
      "정답 문장 :  me voici . \n",
      "번역한 문장:  je suis en s ! \n",
      "--------------------------------------------------\n"
     ]
    }
   ],
   "source": [
    "seq_index = [2,111,201,551,1000]\n",
    "for index in seq_index:\n",
    "    input_seq= encoder_input_train[index:index+1]\n",
    "    decoded_sequence = decode_sequence(input_seq)\n",
    "    print('입력된 문장: ', seq2input(encoder_input_train[index]))\n",
    "    print('정답 문장 : ',seq2target(decoder_input_train[index]))\n",
    "    print('번역한 문장: ', decoded_sequence[:-6])\n",
    "    print('-' *50)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 127,
   "id": "3572fb58",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "입력된 문장:  i love korean food . \n",
      "정답 문장 :  j adore la cuisine cor enne . \n",
      "번역한 문장:  j aime le chien . \n",
      "--------------------------------------------------\n",
      "입력된 문장:  i often wear a hat . \n",
      "정답 문장 :  je porte souvent un chapeau . \n",
      "번역한 문장:  je mange des enfants . \n",
      "--------------------------------------------------\n",
      "입력된 문장:  i smoke cigarettes . \n",
      "정답 문장 :  je fume des cigarettes . \n",
      "번역한 문장:  je vais la maison . \n",
      "--------------------------------------------------\n",
      "입력된 문장:  i m as tall as you . \n",
      "정답 문장 :  je suis aussi grande que vous . \n",
      "번역한 문장:  je suis si grand gar on . \n",
      "--------------------------------------------------\n",
      "입력된 문장:  it wasn t very fun . \n",
      "정답 문장 :  ce ne fut pas tr s amusant . \n",
      "번역한 문장:  ce n tait pas un gar on . \n",
      "--------------------------------------------------\n"
     ]
    }
   ],
   "source": [
    "for index in seq_index:\n",
    "    input_seq= encoder_input_test[index:index+1]\n",
    "    decoded_sequence = decode_sequence(input_seq)\n",
    "    print('입력된 문장: ', seq2input(encoder_input_test[index]))\n",
    "    print('정답 문장 : ',seq2target(decoder_input_test[index]))\n",
    "    print('번역한 문장: ', decoded_sequence[:-6])\n",
    "    print('-' *50)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3e3a303f",
   "metadata": {},
   "source": [
    "## 한국어로 번역 비교\n",
    "프랑스어를 영어로 바꿔주는 구글 번역기를 참고하여 적었습니다.\n",
    "\n",
    "### 1.Train\n",
    "입력된 문장:  go . 가자!   \n",
    "정답 문장 :  en route !   \n",
    "번역한 문장:  va y !   >> 영어로 번역한 문장 :go!, 가자! 번역 성공\n",
    "--------------------------------------------------\n",
    "입력된 문장:  i left . 나는 떠났다.   \n",
    "정답 문장 :  je suis partie .   \n",
    "번역한 문장:  j ai eu de l eau .   >> I had water,  '나는 물을 가졌다', 주어와 과거형은 맞추었다.\n",
    "--------------------------------------------------\n",
    "입력된 문장:  bury it . 묻어.   \n",
    "정답 문장 :  enterrez la .   \n",
    "번역한 문장:  ceci .   >> this, '이것', 동사를 번역하지 못했다.\n",
    "--------------------------------------------------\n",
    "입력된 문장:  go ahead ! 어서해!     \n",
    "정답 문장 :  allez y !    \n",
    "번역한 문장:  allez !   >> go ahead, '어서 해' , come on도 나오지만 비슷하므로 번역에 성공\n",
    "--------------------------------------------------\n",
    "입력된 문장:  here i am . 나 여기야   \n",
    "정답 문장 :  me voici .   \n",
    "번역한 문장:  je suis en s !   >> I am in s !, '나는 s에 있어' s라는 단어는 악센트가 생략된 것 같다. 하지만 입력된 문장은 나의 위치를 알리는 말이므로 성공했다고 볼 수 있다.\n",
    "--------------------------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "df751947",
   "metadata": {},
   "source": [
    "## 2. Test"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "00690f2c",
   "metadata": {},
   "source": [
    "입력된 문장:  i love korean food . '나는 한국음식을 사랑한다.'    \n",
    "정답 문장 :  j adore la cuisine cor enne .    \n",
    "번역한 문장:  j aime le chien . '나는 개를 사랑한다.' 한국음식을 개로 번역했다.   \n",
    "--------------------------------------------------\n",
    "입력된 문장:  i often wear a hat . '나는 가끔 모자를 쓴다.'   \n",
    "정답 문장 :  je porte souvent un chapeau .    \n",
    "번역한 문장:  je mange des enfants .  '나는 아이들을 먹는다.'     \n",
    "--------------------------------------------------\n",
    "입력된 문장:  i smoke cigarettes . '나는 담배를 핀다.'   \n",
    "정답 문장 :  je fume des cigarettes .    \n",
    "번역한 문장:  je vais la maison . '나는 집에 가는중이다'   \n",
    "--------------------------------------------------\n",
    "입력된 문장:  i m as tall as you . '나는 너만큼 크다'   \n",
    "정답 문장 :  je suis aussi grande que vous .    \n",
    "번역한 문장:  je suis si grand gar on . '나는 꽤 큰 남자다.'   \n",
    "--------------------------------------------------\n",
    "입력된 문장:  it wasn t very fun . '그건 재미 있지 않았어.'   \n",
    "정답 문장 :  ce ne fut pas tr s amusant .    \n",
    "번역한 문장:  ce n tait pas un gar on . '그건 소년이 아니었어.'   \n",
    "--------------------------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e116ca66",
   "metadata": {},
   "source": [
    "### 회고\n",
    "seq2seq의 모델구조를 이해하지 못하여 매우 힘든 과정이었다.   \n",
    "하지만 과정에 비해 결과가 만족스럽지 못했다.   \n",
    "Train같은 경우는 완벽히 맞춘 문장이 있었고, 못맞춘 문장은 문맥상 맞거나 동사 부분을 번역하지 못했다.   \n",
    "Test의 경우 완벽히 맞춘문장은 없었고 4번 문장이 문맥상 비슷했고, 나머지는 다른 의미로 번역을 했다.   \n",
    "Train은 이미 학습했던 문장을 한번 더 번역해보았기 때문에 정확도가 높았고, test는 학습하지 않았던 문장을 넣어줬기 때문에\n",
    "위와 같은 결과가 나왔다고 생각한다.     \n",
    "test문장의 정확도 향상을 위해서 임베딩 차원수와 lstm유닛수를 조절해봐야 할 것 같다."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "771a9a01",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.9.15"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}