seq2word_seq2seq.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'2.0.0'"
      ]
     },
     "execution_count": 64,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import math\n",
    "import sklearn\n",
    "import sklearn.preprocessing\n",
    "import os\n",
    "import h5py\n",
    "import matplotlib.pyplot as plt\n",
    "import tensorflow as tf\n",
    "import utils\n",
    "\n",
    "from tensorflow.keras.models import Model, Sequential\n",
    "from tensorflow.keras.optimizers import Adam\n",
    "from tensorflow.keras.layers import LSTM, GRU, Bidirectional, Dense, Dropout, Input, TimeDistributed, RepeatVector, Reshape, Lambda, concatenate\n",
    "from tensorflow.keras.layers import Flatten, LeakyReLU, BatchNormalization\n",
    "from tensorflow.keras.callbacks import ModelCheckpoint,TensorBoard\n",
    "# from tensorflow.keras.callbacks.tensorboard_v1 import Tensorboard\n",
    "from tensorflow.keras import backend as K\n",
    "\n",
    "tf.__version__\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['DJI_fe.csv',\n",
       " 'DJI_STO_20period_sma10.csv',\n",
       " 'N225.csv',\n",
       " 'DJI_modified_features - DJI.csv',\n",
       " 'N225fillVOL.csv',\n",
       " 'DJI_modified_sma.csv',\n",
       " 'downjone5y.csv',\n",
       " 'N225withoutNULL.csv',\n",
       " 'dowjone1985_2019.csv',\n",
       " '.ipynb_checkpoints',\n",
       " 'DJI_ema.csv',\n",
       " 'DJI_denoise.csv']"
      ]
     },
     "execution_count": 65,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "os.listdir('./data')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [],
   "source": [
    "# data stats parameter\n",
    "val_set_size_percentage = 10\n",
    "test_set_size_percentage = 10\n",
    "seq_len = 20\n",
    "\n",
    "# hyperparameter\n",
    "batch_size = 1024\n",
    "epochs = 100\n",
    "lr = 1e-3\n",
    "decay = 1e-2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [],
   "source": [
    "# callback setup\n",
    "log_name = 'seq2seq_1'\n",
    "\n",
    "tensorboard_callback = TensorBoard(log_dir=\"logs/\" + log_name)\n",
    "modelcheckpoint_callback = ModelCheckpoint(\"save_model/\" + log_name + \"_{epoch:02d}_weight.hdf5\", save_best_only=True)\n",
    "\n",
    "callbacks_list = [\n",
    "    tensorboard_callback,\n",
    "    modelcheckpoint_callback\n",
    "]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "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>Date</th>\n",
       "      <th>Open</th>\n",
       "      <th>High</th>\n",
       "      <th>Low</th>\n",
       "      <th>Close</th>\n",
       "      <th>Adj Close</th>\n",
       "      <th>Volume</th>\n",
       "      <th>sma_5</th>\n",
       "      <th>sma_10</th>\n",
       "      <th>sma_25</th>\n",
       "      <th>...</th>\n",
       "      <th>sma_125</th>\n",
       "      <th>sma_200</th>\n",
       "      <th>%K</th>\n",
       "      <th>%D</th>\n",
       "      <th>ema_5</th>\n",
       "      <th>ema_10</th>\n",
       "      <th>ema_25</th>\n",
       "      <th>ema_75</th>\n",
       "      <th>ema_125</th>\n",
       "      <th>ema_200</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1985-01-29</td>\n",
       "      <td>1277.719971</td>\n",
       "      <td>1295.489990</td>\n",
       "      <td>1266.890015</td>\n",
       "      <td>1292.619995</td>\n",
       "      <td>1292.619995</td>\n",
       "      <td>13560000</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1292.619995</td>\n",
       "      <td>1292.619995</td>\n",
       "      <td>1292.619995</td>\n",
       "      <td>1292.619995</td>\n",
       "      <td>1292.619995</td>\n",
       "      <td>1292.619995</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1985-01-30</td>\n",
       "      <td>1297.369995</td>\n",
       "      <td>1305.099976</td>\n",
       "      <td>1278.930054</td>\n",
       "      <td>1287.880005</td>\n",
       "      <td>1287.880005</td>\n",
       "      <td>16820000</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1291.039998</td>\n",
       "      <td>1291.758179</td>\n",
       "      <td>1292.255380</td>\n",
       "      <td>1292.495258</td>\n",
       "      <td>1292.544757</td>\n",
       "      <td>1292.572831</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>1985-01-31</td>\n",
       "      <td>1283.239990</td>\n",
       "      <td>1293.400024</td>\n",
       "      <td>1272.640015</td>\n",
       "      <td>1286.770020</td>\n",
       "      <td>1286.770020</td>\n",
       "      <td>14070000</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1289.616672</td>\n",
       "      <td>1290.851241</td>\n",
       "      <td>1291.833430</td>\n",
       "      <td>1292.344594</td>\n",
       "      <td>1292.453095</td>\n",
       "      <td>1292.515092</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>1985-02-01</td>\n",
       "      <td>1276.939941</td>\n",
       "      <td>1286.109985</td>\n",
       "      <td>1269.770020</td>\n",
       "      <td>1277.719971</td>\n",
       "      <td>1277.719971</td>\n",
       "      <td>10980000</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1285.651105</td>\n",
       "      <td>1288.463737</td>\n",
       "      <td>1290.747779</td>\n",
       "      <td>1291.959736</td>\n",
       "      <td>1292.219235</td>\n",
       "      <td>1292.367876</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>1985-02-04</td>\n",
       "      <td>1272.079956</td>\n",
       "      <td>1294.939941</td>\n",
       "      <td>1268.989990</td>\n",
       "      <td>1290.079956</td>\n",
       "      <td>1290.079956</td>\n",
       "      <td>11630000</td>\n",
       "      <td>1287.013989</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1287.127389</td>\n",
       "      <td>1288.757595</td>\n",
       "      <td>1290.696408</td>\n",
       "      <td>1291.910268</td>\n",
       "      <td>1292.185279</td>\n",
       "      <td>1292.345111</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 21 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "         Date         Open         High          Low        Close  \\\n",
       "0  1985-01-29  1277.719971  1295.489990  1266.890015  1292.619995   \n",
       "1  1985-01-30  1297.369995  1305.099976  1278.930054  1287.880005   \n",
       "2  1985-01-31  1283.239990  1293.400024  1272.640015  1286.770020   \n",
       "3  1985-02-01  1276.939941  1286.109985  1269.770020  1277.719971   \n",
       "4  1985-02-04  1272.079956  1294.939941  1268.989990  1290.079956   \n",
       "\n",
       "     Adj Close    Volume        sma_5  sma_10  sma_25  ...  sma_125  sma_200  \\\n",
       "0  1292.619995  13560000          NaN     NaN     NaN  ...      NaN      NaN   \n",
       "1  1287.880005  16820000          NaN     NaN     NaN  ...      NaN      NaN   \n",
       "2  1286.770020  14070000          NaN     NaN     NaN  ...      NaN      NaN   \n",
       "3  1277.719971  10980000          NaN     NaN     NaN  ...      NaN      NaN   \n",
       "4  1290.079956  11630000  1287.013989     NaN     NaN  ...      NaN      NaN   \n",
       "\n",
       "   %K  %D        ema_5       ema_10       ema_25       ema_75      ema_125  \\\n",
       "0 NaN NaN  1292.619995  1292.619995  1292.619995  1292.619995  1292.619995   \n",
       "1 NaN NaN  1291.039998  1291.758179  1292.255380  1292.495258  1292.544757   \n",
       "2 NaN NaN  1289.616672  1290.851241  1291.833430  1292.344594  1292.453095   \n",
       "3 NaN NaN  1285.651105  1288.463737  1290.747779  1291.959736  1292.219235   \n",
       "4 NaN NaN  1287.127389  1288.757595  1290.696408  1291.910268  1292.185279   \n",
       "\n",
       "       ema_200  \n",
       "0  1292.619995  \n",
       "1  1292.572831  \n",
       "2  1292.515092  \n",
       "3  1292.367876  \n",
       "4  1292.345111  \n",
       "\n",
       "[5 rows x 21 columns]"
      ]
     },
     "execution_count": 68,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df = pd.read_csv('./data/DJI_fe.csv')\n",
    "df = df.drop('Unnamed: 0', axis=1)\n",
    "df = df.drop('Unnamed: 0.1', axis=1)\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "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"
    },
    {
     "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": [
    "utils.plot_raw(df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [],
   "source": [
    "df = df.dropna()\n",
    "df = df.drop('Date', axis=1)\n",
    "df = df.drop('Adj Close', axis=1)\n",
    "df = df.drop('Volume', axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['Open', 'Close', 'High', 'Low']\n",
      "4\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/2two2/Documents/MultimodelStockPrediction/utils.py:44: FutureWarning: Method .as_matrix will be removed in a future version. Use .values instead.\n",
      "  data_raw = stock.as_matrix()\n"
     ]
    }
   ],
   "source": [
    "cols = ['Open', 'Close', 'High', 'Low']\n",
    "# range_ma = [5, 10, 25, 75, 125, 200]\n",
    "# for ma in range_ma:\n",
    "#     cols.append('sma_' + str(ma))\n",
    "#     cols.append('ema_' + str(ma))\n",
    "\n",
    "print(cols)\n",
    "print(len(cols))\n",
    "\n",
    "\n",
    "df_norm = utils.normalize_data(df,cols)\n",
    "\n",
    "x_train, y_train, x_val, y_val, x_test, y_test = utils.load_data(df_norm[cols])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "x_train.shape =  (6835, 19, 4)\n",
      "y_train.shape =  (6835, 4)\n",
      "x_val.shape =  (854, 19, 4)\n",
      "y_val.shape =  (854, 4)\n",
      "x_test.shape =  (854, 19, 4)\n",
      "y_test.shape =  (854, 4)\n"
     ]
    }
   ],
   "source": [
    "print('x_train.shape = ', x_train.shape)\n",
    "print('y_train.shape = ', y_train.shape)\n",
    "print('x_val.shape = ', x_val.shape)\n",
    "print('y_val.shape = ', y_val.shape)\n",
    "print('x_test.shape = ', x_test.shape)\n",
    "print('y_test.shape = ', y_test.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {},
   "outputs": [],
   "source": [
    "weight_path_model = \"saved_model/model_seq2word.h5\"\n",
    "callbacks_list_model = [\n",
    "    ModelCheckpoint(\n",
    "             weight_path_model,\n",
    "            save_best_only=True,\n",
    "            save_weights_only=True,\n",
    "            monitor='val_loss',\n",
    "            mode='min',\n",
    "            verbose=1\n",
    "        )\n",
    "]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {},
   "outputs": [],
   "source": [
    "# ! seq2word not real seq2seq\n",
    "def seq2word_model():\n",
    "    seq_len = 20 - 1\n",
    "    \n",
    "    # Encoder\n",
    "    encoder_inputs = Input(shape=(19, 4))\n",
    "    encoder = LSTM(100, return_state=True, name='encoder')\n",
    "    encoder_outputs, state_h, state_c = encoder(encoder_inputs)\n",
    "    states = [state_h, state_c]\n",
    "    \n",
    "    # Decoder\n",
    "    reshaper = Reshape((1, 100), name='reshaper')\n",
    "    decoder = LSTM(100, return_sequences=True, return_state=True,\n",
    "                  name='decoder')\n",
    "    \n",
    "    # Fully-connected\n",
    "    fc_output = Dense(4, activation='linear', \n",
    "              name='fc')\n",
    "    \n",
    "    # Reshape output\n",
    "    reshaper_output = Reshape((4,), name='reshaper_output')\n",
    "    \n",
    "    inputs = reshaper(encoder_outputs)\n",
    "    outputs, h, c = decoder(inputs, initial_state=states)\n",
    "        \n",
    "    inputs = outputs\n",
    "    states = [h, c]\n",
    "\n",
    "    outputs = reshaper_output(fc_output(outputs))\n",
    "\n",
    "    model = Model(inputs=encoder_inputs, outputs=outputs)\n",
    "    \n",
    "    model.compile(optimizer=Adam(learning_rate=lr, decay=decay),\n",
    "                loss='mse',\n",
    "                metrics=['mse'])   \n",
    "    \n",
    "#     model.summary()\n",
    "    \n",
    "    return model\n",
    "\n",
    "def seq2seq_model():\n",
    "    seq_len = 20 - 1\n",
    "    \n",
    "    # Encoder\n",
    "    encoder_inputs = Input(shape=(19, 4))\n",
    "    encoder = LSTM(100, return_state=True, name='encoder')\n",
    "    encoder_outputs, state_h, state_c = encoder(encoder_inputs)\n",
    "    states = [state_h, state_c]\n",
    "    \n",
    "    # Decoder\n",
    "    reshaper = Reshape((1, 100), name='reshaper')\n",
    "    decoder = LSTM(100, return_sequences=True, return_state=True,\n",
    "                  name='decoder')\n",
    "    \n",
    "    # Fully-connected\n",
    "    fc_output = Dense(4, activation='linear', \n",
    "              name='fc')\n",
    "\n",
    "    inputs = reshaper(encoder_outputs)\n",
    "\n",
    "    all_outputs = list()\n",
    "    \n",
    "    for _ in range(seq_len):\n",
    "        outputs, h, c = decoder(inputs, initial_state=states)\n",
    "        \n",
    "        inputs = outputs\n",
    "        states = [h, c]\n",
    "        \n",
    "        outputs = fc_output(outputs)\n",
    "        all_outputs.append(outputs)\n",
    "        \n",
    "    decoder_outputs = Lambda(lambda x: concatenate(x, axis=1))(all_outputs)\n",
    "    model = Model(inputs=encoder_inputs, outputs=decoder_outputs)\n",
    "    \n",
    "    model.compile(optimizer=Adam(learning_rate=lr, decay=decay),\n",
    "                loss='mse',\n",
    "                metrics=['mse'])   \n",
    "    \n",
    "#     model.summary()\n",
    "    \n",
    "    return model\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(6835, 4)"
      ]
     },
     "execution_count": 79,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_1 (InputLayer)            [(None, 19, 4)]      0                                            \n",
      "__________________________________________________________________________________________________\n",
      "encoder (LSTM)                  [(None, 100), (None, 42000       input_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "reshaper (Reshape)              (None, 1, 100)       0           encoder[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "decoder (LSTM)                  [(None, 1, 100), (No 80400       reshaper[0][0]                   \n",
      "                                                                 encoder[0][1]                    \n",
      "                                                                 encoder[0][2]                    \n",
      "__________________________________________________________________________________________________\n",
      "fc (Dense)                      (None, 1, 4)         404         decoder[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "reshaper_output (Reshape)       (None, 4)            0           fc[0][0]                         \n",
      "==================================================================================================\n",
      "Total params: 122,804\n",
      "Trainable params: 122,804\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "Train on 6835 samples, validate on 854 samples\n",
      "Epoch 1/100\n",
      "6144/6835 [=========================>....] - ETA: 1s - loss: 0.0347 - mse: 0.0347\n",
      "Epoch 00001: val_loss improved from inf to 0.01586, saving model to saved_model/model_seq2word.h5\n",
      "6835/6835 [==============================] - 12s 2ms/sample - loss: 0.0313 - mse: 0.0313 - val_loss: 0.0159 - val_mse: 0.0159\n",
      "Epoch 2/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 0.0052 - mse: 0.0052\n",
      "Epoch 00002: val_loss improved from 0.01586 to 0.01028, saving model to saved_model/model_seq2word.h5\n",
      "6835/6835 [==============================] - 2s 327us/sample - loss: 0.0048 - mse: 0.0048 - val_loss: 0.0103 - val_mse: 0.0103\n",
      "Epoch 3/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 0.0026 - mse: 0.0026\n",
      "Epoch 00003: val_loss improved from 0.01028 to 0.00990, saving model to saved_model/model_seq2word.h5\n",
      "6835/6835 [==============================] - 3s 399us/sample - loss: 0.0026 - mse: 0.0026 - val_loss: 0.0099 - val_mse: 0.0099\n",
      "Epoch 4/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 9.5815e-04 - mse: 9.5816e-04\n",
      "Epoch 00004: val_loss improved from 0.00990 to 0.00103, saving model to saved_model/model_seq2word.h5\n",
      "6835/6835 [==============================] - 3s 417us/sample - loss: 9.5335e-04 - mse: 9.5335e-04 - val_loss: 0.0010 - val_mse: 0.0010\n",
      "Epoch 5/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 7.9808e-04 - mse: 7.9808e-04\n",
      "Epoch 00005: val_loss improved from 0.00103 to 0.00077, saving model to saved_model/model_seq2word.h5\n",
      "6835/6835 [==============================] - 2s 268us/sample - loss: 7.5177e-04 - mse: 7.5177e-04 - val_loss: 7.7037e-04 - val_mse: 7.7037e-04\n",
      "Epoch 6/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7477e-04 - mse: 3.7477e-04\n",
      "Epoch 00006: val_loss did not improve from 0.00077\n",
      "6835/6835 [==============================] - 2s 255us/sample - loss: 3.7863e-04 - mse: 3.7863e-04 - val_loss: 0.0018 - val_mse: 0.0018\n",
      "Epoch 7/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 2.1419e-04 - mse: 2.1419e-04\n",
      "Epoch 00007: val_loss improved from 0.00077 to 0.00016, saving model to saved_model/model_seq2word.h5\n",
      "6835/6835 [==============================] - 2s 270us/sample - loss: 2.0730e-04 - mse: 2.0730e-04 - val_loss: 1.6488e-04 - val_mse: 1.6488e-04\n",
      "Epoch 8/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 1.5375e-04 - mse: 1.5375e-04\n",
      "Epoch 00008: val_loss improved from 0.00016 to 0.00013, saving model to saved_model/model_seq2word.h5\n",
      "6835/6835 [==============================] - 2s 304us/sample - loss: 1.4955e-04 - mse: 1.4955e-04 - val_loss: 1.2549e-04 - val_mse: 1.2549e-04\n",
      "Epoch 9/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 7.5475e-05 - mse: 7.5475e-05\n",
      "Epoch 00009: val_loss did not improve from 0.00013\n",
      "6835/6835 [==============================] - 2s 308us/sample - loss: 7.5216e-05 - mse: 7.5216e-05 - val_loss: 1.9604e-04 - val_mse: 1.9604e-04\n",
      "Epoch 10/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 6.2374e-05 - mse: 6.2374e-05\n",
      "Epoch 00010: val_loss improved from 0.00013 to 0.00010, saving model to saved_model/model_seq2word.h5\n",
      "6835/6835 [==============================] - 2s 353us/sample - loss: 6.0728e-05 - mse: 6.0728e-05 - val_loss: 1.0329e-04 - val_mse: 1.0329e-04\n",
      "Epoch 11/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 4.9289e-05 - mse: 4.9289e-05\n",
      "Epoch 00011: val_loss did not improve from 0.00010\n",
      "6835/6835 [==============================] - 2s 284us/sample - loss: 4.9312e-05 - mse: 4.9312e-05 - val_loss: 1.0680e-04 - val_mse: 1.0680e-04\n",
      "Epoch 12/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 4.4418e-05 - mse: 4.4418e-05\n",
      "Epoch 00012: val_loss improved from 0.00010 to 0.00008, saving model to saved_model/model_seq2word.h5\n",
      "6835/6835 [==============================] - 2s 270us/sample - loss: 4.4002e-05 - mse: 4.4002e-05 - val_loss: 7.9820e-05 - val_mse: 7.9820e-05\n",
      "Epoch 13/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 4.2095e-05 - mse: 4.2095e-05\n",
      "Epoch 00013: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 260us/sample - loss: 4.2438e-05 - mse: 4.2438e-05 - val_loss: 1.1080e-04 - val_mse: 1.1080e-04\n",
      "Epoch 14/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 4.1530e-05 - mse: 4.1530e-05\n",
      "Epoch 00014: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 254us/sample - loss: 4.1876e-05 - mse: 4.1876e-05 - val_loss: 1.0043e-04 - val_mse: 1.0043e-04\n",
      "Epoch 15/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 4.0086e-05 - mse: 4.0086e-05\n",
      "Epoch 00015: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 287us/sample - loss: 4.1141e-05 - mse: 4.1141e-05 - val_loss: 9.0281e-05 - val_mse: 9.0281e-05\n",
      "Epoch 16/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 4.0923e-05 - mse: 4.0923e-05\n",
      "Epoch 00016: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 336us/sample - loss: 4.0989e-05 - mse: 4.0989e-05 - val_loss: 9.3823e-05 - val_mse: 9.3823e-05\n",
      "Epoch 17/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 4.1078e-05 - mse: 4.1078e-05\n",
      "Epoch 00017: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 353us/sample - loss: 4.0779e-05 - mse: 4.0779e-05 - val_loss: 9.4260e-05 - val_mse: 9.4260e-05\n",
      "Epoch 18/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 4.0888e-05 - mse: 4.0888e-05\n",
      "Epoch 00018: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 399us/sample - loss: 4.0596e-05 - mse: 4.0596e-05 - val_loss: 8.6961e-05 - val_mse: 8.6961e-05\n",
      "Epoch 19/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 4.0584e-05 - mse: 4.0584e-05\n",
      "Epoch 00019: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 297us/sample - loss: 4.0482e-05 - mse: 4.0482e-05 - val_loss: 8.5601e-05 - val_mse: 8.5601e-05\n",
      "Epoch 20/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.9295e-05 - mse: 3.9295e-05\n",
      "Epoch 00020: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 269us/sample - loss: 4.0334e-05 - mse: 4.0334e-05 - val_loss: 8.6484e-05 - val_mse: 8.6484e-05\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 21/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 4.0678e-05 - mse: 4.0678e-05\n",
      "Epoch 00021: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 368us/sample - loss: 4.0245e-05 - mse: 4.0245e-05 - val_loss: 8.4592e-05 - val_mse: 8.4592e-05\n",
      "Epoch 22/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.9501e-05 - mse: 3.9501e-05\n",
      "Epoch 00022: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 273us/sample - loss: 4.0168e-05 - mse: 4.0168e-05 - val_loss: 8.1285e-05 - val_mse: 8.1285e-05\n",
      "Epoch 23/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.9452e-05 - mse: 3.9452e-05\n",
      "Epoch 00023: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 259us/sample - loss: 4.0019e-05 - mse: 4.0019e-05 - val_loss: 8.3385e-05 - val_mse: 8.3385e-05\n",
      "Epoch 24/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.9398e-05 - mse: 3.9398e-05\n",
      "Epoch 00024: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 309us/sample - loss: 3.9947e-05 - mse: 3.9947e-05 - val_loss: 8.1843e-05 - val_mse: 8.1843e-05\n",
      "Epoch 25/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.9899e-05 - mse: 3.9899e-05\n",
      "Epoch 00025: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 348us/sample - loss: 3.9827e-05 - mse: 3.9827e-05 - val_loss: 7.9927e-05 - val_mse: 7.9927e-05\n",
      "Epoch 26/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 4.0031e-05 - mse: 4.0031e-05\n",
      "Epoch 00026: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 338us/sample - loss: 3.9740e-05 - mse: 3.9740e-05 - val_loss: 7.9918e-05 - val_mse: 7.9918e-05\n",
      "Epoch 27/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.9418e-05 - mse: 3.9418e-05\n",
      "Epoch 00027: val_loss improved from 0.00008 to 0.00008, saving model to saved_model/model_seq2word.h5\n",
      "6835/6835 [==============================] - 2s 328us/sample - loss: 3.9655e-05 - mse: 3.9655e-05 - val_loss: 7.9372e-05 - val_mse: 7.9372e-05\n",
      "Epoch 28/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.9550e-05 - mse: 3.9550e-05\n",
      "Epoch 00028: val_loss improved from 0.00008 to 0.00008, saving model to saved_model/model_seq2word.h5\n",
      "6835/6835 [==============================] - 2s 290us/sample - loss: 3.9563e-05 - mse: 3.9563e-05 - val_loss: 7.8498e-05 - val_mse: 7.8498e-05\n",
      "Epoch 29/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.9416e-05 - mse: 3.9416e-05\n",
      "Epoch 00029: val_loss improved from 0.00008 to 0.00008, saving model to saved_model/model_seq2word.h5\n",
      "6835/6835 [==============================] - 2s 308us/sample - loss: 3.9477e-05 - mse: 3.9477e-05 - val_loss: 7.8283e-05 - val_mse: 7.8283e-05\n",
      "Epoch 30/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.9031e-05 - mse: 3.9031e-05\n",
      "Epoch 00030: val_loss improved from 0.00008 to 0.00008, saving model to saved_model/model_seq2word.h5\n",
      "6835/6835 [==============================] - 3s 380us/sample - loss: 3.9392e-05 - mse: 3.9392e-05 - val_loss: 7.7926e-05 - val_mse: 7.7926e-05\n",
      "Epoch 31/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.9615e-05 - mse: 3.9615e-05\n",
      "Epoch 00031: val_loss improved from 0.00008 to 0.00008, saving model to saved_model/model_seq2word.h5\n",
      "6835/6835 [==============================] - 3s 406us/sample - loss: 3.9312e-05 - mse: 3.9312e-05 - val_loss: 7.7214e-05 - val_mse: 7.7214e-05\n",
      "Epoch 32/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.9543e-05 - mse: 3.9543e-05\n",
      "Epoch 00032: val_loss improved from 0.00008 to 0.00008, saving model to saved_model/model_seq2word.h5\n",
      "6835/6835 [==============================] - 3s 451us/sample - loss: 3.9238e-05 - mse: 3.9238e-05 - val_loss: 7.6936e-05 - val_mse: 7.6936e-05\n",
      "Epoch 33/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.9185e-05 - mse: 3.9185e-05\n",
      "Epoch 00033: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 440us/sample - loss: 3.9163e-05 - mse: 3.9163e-05 - val_loss: 7.7073e-05 - val_mse: 7.7073e-05\n",
      "Epoch 34/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.8680e-05 - mse: 3.8680e-05\n",
      "Epoch 00034: val_loss improved from 0.00008 to 0.00008, saving model to saved_model/model_seq2word.h5\n",
      "6835/6835 [==============================] - 4s 542us/sample - loss: 3.9088e-05 - mse: 3.9088e-05 - val_loss: 7.6175e-05 - val_mse: 7.6175e-05\n",
      "Epoch 35/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.9041e-05 - mse: 3.9041e-05\n",
      "Epoch 00035: val_loss improved from 0.00008 to 0.00008, saving model to saved_model/model_seq2word.h5\n",
      "6835/6835 [==============================] - 3s 412us/sample - loss: 3.9014e-05 - mse: 3.9014e-05 - val_loss: 7.6166e-05 - val_mse: 7.6166e-05\n",
      "Epoch 36/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.9626e-05 - mse: 3.9626e-05\n",
      "Epoch 00036: val_loss improved from 0.00008 to 0.00008, saving model to saved_model/model_seq2word.h5\n",
      "6835/6835 [==============================] - 2s 298us/sample - loss: 3.8943e-05 - mse: 3.8943e-05 - val_loss: 7.6000e-05 - val_mse: 7.6000e-05\n",
      "Epoch 37/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.8854e-05 - mse: 3.8854e-05\n",
      "Epoch 00037: val_loss improved from 0.00008 to 0.00008, saving model to saved_model/model_seq2word.h5\n",
      "6835/6835 [==============================] - 2s 256us/sample - loss: 3.8891e-05 - mse: 3.8891e-05 - val_loss: 7.5932e-05 - val_mse: 7.5932e-05\n",
      "Epoch 38/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.9496e-05 - mse: 3.9495e-05\n",
      "Epoch 00038: val_loss improved from 0.00008 to 0.00008, saving model to saved_model/model_seq2word.h5\n",
      "6835/6835 [==============================] - 2s 325us/sample - loss: 3.8796e-05 - mse: 3.8796e-05 - val_loss: 7.5659e-05 - val_mse: 7.5659e-05\n",
      "Epoch 39/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7234e-05 - mse: 3.7234e-05\n",
      "Epoch 00039: val_loss improved from 0.00008 to 0.00008, saving model to saved_model/model_seq2word.h5\n",
      "6835/6835 [==============================] - 2s 337us/sample - loss: 3.8749e-05 - mse: 3.8749e-05 - val_loss: 7.5598e-05 - val_mse: 7.5598e-05\n",
      "Epoch 40/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.8263e-05 - mse: 3.8263e-05\n",
      "Epoch 00040: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 257us/sample - loss: 3.8676e-05 - mse: 3.8676e-05 - val_loss: 7.5604e-05 - val_mse: 7.5604e-05\n",
      "Epoch 41/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.8517e-05 - mse: 3.8517e-05\n",
      "Epoch 00041: val_loss improved from 0.00008 to 0.00008, saving model to saved_model/model_seq2word.h5\n",
      "6835/6835 [==============================] - 2s 261us/sample - loss: 3.8617e-05 - mse: 3.8617e-05 - val_loss: 7.5481e-05 - val_mse: 7.5481e-05\n",
      "Epoch 42/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.8206e-05 - mse: 3.8206e-05\n",
      "Epoch 00042: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 317us/sample - loss: 3.8552e-05 - mse: 3.8552e-05 - val_loss: 7.5518e-05 - val_mse: 7.5518e-05\n",
      "Epoch 43/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.6975e-05 - mse: 3.6975e-05\n",
      "Epoch 00043: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 353us/sample - loss: 3.8499e-05 - mse: 3.8499e-05 - val_loss: 7.5716e-05 - val_mse: 7.5716e-05\n",
      "Epoch 44/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7846e-05 - mse: 3.7846e-05\n",
      "Epoch 00044: val_loss improved from 0.00008 to 0.00008, saving model to saved_model/model_seq2word.h5\n",
      "6835/6835 [==============================] - 2s 354us/sample - loss: 3.8457e-05 - mse: 3.8457e-05 - val_loss: 7.5451e-05 - val_mse: 7.5451e-05\n",
      "Epoch 45/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.8706e-05 - mse: 3.8706e-05\n",
      "Epoch 00045: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 253us/sample - loss: 3.8375e-05 - mse: 3.8375e-05 - val_loss: 7.5810e-05 - val_mse: 7.5810e-05\n",
      "Epoch 46/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.8356e-05 - mse: 3.8356e-05\n",
      "Epoch 00046: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 246us/sample - loss: 3.8332e-05 - mse: 3.8332e-05 - val_loss: 7.5869e-05 - val_mse: 7.5869e-05\n",
      "Epoch 47/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.8088e-05 - mse: 3.8088e-05\n",
      "Epoch 00047: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 244us/sample - loss: 3.8269e-05 - mse: 3.8269e-05 - val_loss: 7.5806e-05 - val_mse: 7.5806e-05\n",
      "Epoch 48/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7960e-05 - mse: 3.7960e-05\n",
      "Epoch 00048: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 241us/sample - loss: 3.8234e-05 - mse: 3.8234e-05 - val_loss: 7.5731e-05 - val_mse: 7.5731e-05\n",
      "Epoch 49/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.8351e-05 - mse: 3.8351e-05\n",
      "Epoch 00049: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 245us/sample - loss: 3.8159e-05 - mse: 3.8159e-05 - val_loss: 7.6792e-05 - val_mse: 7.6792e-05\n",
      "Epoch 50/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.8234e-05 - mse: 3.8234e-05\n",
      "Epoch 00050: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 239us/sample - loss: 3.8127e-05 - mse: 3.8127e-05 - val_loss: 7.6293e-05 - val_mse: 7.6293e-05\n",
      "Epoch 51/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.8038e-05 - mse: 3.8038e-05\n",
      "Epoch 00051: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 370us/sample - loss: 3.8066e-05 - mse: 3.8066e-05 - val_loss: 7.6307e-05 - val_mse: 7.6307e-05\n",
      "Epoch 52/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7890e-05 - mse: 3.7890e-05\n",
      "Epoch 00052: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 324us/sample - loss: 3.8021e-05 - mse: 3.8021e-05 - val_loss: 7.6359e-05 - val_mse: 7.6359e-05\n",
      "Epoch 53/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7967e-05 - mse: 3.7967e-05\n",
      "Epoch 00053: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 351us/sample - loss: 3.7993e-05 - mse: 3.7993e-05 - val_loss: 7.6876e-05 - val_mse: 7.6876e-05\n",
      "Epoch 54/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.8407e-05 - mse: 3.8407e-05\n",
      "Epoch 00054: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 455us/sample - loss: 3.7926e-05 - mse: 3.7926e-05 - val_loss: 7.8225e-05 - val_mse: 7.8225e-05\n",
      "Epoch 55/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.8457e-05 - mse: 3.8457e-05\n",
      "Epoch 00055: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 303us/sample - loss: 3.7889e-05 - mse: 3.7889e-05 - val_loss: 7.7293e-05 - val_mse: 7.7293e-05\n",
      "Epoch 56/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.8323e-05 - mse: 3.8323e-05\n",
      "Epoch 00056: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 480us/sample - loss: 3.7826e-05 - mse: 3.7826e-05 - val_loss: 7.6773e-05 - val_mse: 7.6773e-05\n",
      "Epoch 57/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7602e-05 - mse: 3.7602e-05\n",
      "Epoch 00057: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 469us/sample - loss: 3.7812e-05 - mse: 3.7812e-05 - val_loss: 7.7817e-05 - val_mse: 7.7817e-05\n",
      "Epoch 58/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7434e-05 - mse: 3.7434e-05\n",
      "Epoch 00058: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 410us/sample - loss: 3.7749e-05 - mse: 3.7749e-05 - val_loss: 7.8121e-05 - val_mse: 7.8121e-05\n",
      "Epoch 59/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.8156e-05 - mse: 3.8156e-05\n",
      "Epoch 00059: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 381us/sample - loss: 3.7710e-05 - mse: 3.7710e-05 - val_loss: 7.8288e-05 - val_mse: 7.8288e-05\n",
      "Epoch 60/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.6938e-05 - mse: 3.6938e-05\n",
      "Epoch 00060: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 386us/sample - loss: 3.7682e-05 - mse: 3.7682e-05 - val_loss: 7.9368e-05 - val_mse: 7.9368e-05\n",
      "Epoch 61/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.8182e-05 - mse: 3.8182e-05\n",
      "Epoch 00061: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 350us/sample - loss: 3.7649e-05 - mse: 3.7649e-05 - val_loss: 7.8505e-05 - val_mse: 7.8505e-05\n",
      "Epoch 62/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7575e-05 - mse: 3.7575e-05\n",
      "Epoch 00062: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 404us/sample - loss: 3.7601e-05 - mse: 3.7601e-05 - val_loss: 7.9935e-05 - val_mse: 7.9935e-05\n",
      "Epoch 63/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.6709e-05 - mse: 3.6709e-05\n",
      "Epoch 00063: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 393us/sample - loss: 3.7552e-05 - mse: 3.7552e-05 - val_loss: 7.9770e-05 - val_mse: 7.9770e-05\n",
      "Epoch 64/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7235e-05 - mse: 3.7235e-05\n",
      "Epoch 00064: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 501us/sample - loss: 3.7522e-05 - mse: 3.7522e-05 - val_loss: 7.9303e-05 - val_mse: 7.9303e-05\n",
      "Epoch 65/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.6973e-05 - mse: 3.6973e-05\n",
      "Epoch 00065: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 410us/sample - loss: 3.7481e-05 - mse: 3.7481e-05 - val_loss: 8.1166e-05 - val_mse: 8.1166e-05\n",
      "Epoch 66/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7484e-05 - mse: 3.7484e-05\n",
      "Epoch 00066: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 4s 541us/sample - loss: 3.7445e-05 - mse: 3.7445e-05 - val_loss: 8.0489e-05 - val_mse: 8.0489e-05\n",
      "Epoch 67/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7102e-05 - mse: 3.7102e-05\n",
      "Epoch 00067: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 316us/sample - loss: 3.7420e-05 - mse: 3.7420e-05 - val_loss: 8.1463e-05 - val_mse: 8.1463e-05\n",
      "Epoch 68/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7865e-05 - mse: 3.7865e-05\n",
      "Epoch 00068: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 297us/sample - loss: 3.7364e-05 - mse: 3.7364e-05 - val_loss: 8.0570e-05 - val_mse: 8.0570e-05\n",
      "Epoch 69/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7450e-05 - mse: 3.7450e-05\n",
      "Epoch 00069: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 236us/sample - loss: 3.7351e-05 - mse: 3.7351e-05 - val_loss: 8.1299e-05 - val_mse: 8.1299e-05\n",
      "Epoch 70/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7510e-05 - mse: 3.7510e-05\n",
      "Epoch 00070: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 327us/sample - loss: 3.7292e-05 - mse: 3.7292e-05 - val_loss: 8.3466e-05 - val_mse: 8.3466e-05\n",
      "Epoch 71/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7411e-05 - mse: 3.7411e-05\n",
      "Epoch 00071: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 263us/sample - loss: 3.7279e-05 - mse: 3.7279e-05 - val_loss: 8.2978e-05 - val_mse: 8.2978e-05\n",
      "Epoch 72/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7384e-05 - mse: 3.7384e-05\n",
      "Epoch 00072: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 487us/sample - loss: 3.7237e-05 - mse: 3.7237e-05 - val_loss: 8.1804e-05 - val_mse: 8.1804e-05\n",
      "Epoch 73/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7477e-05 - mse: 3.7477e-05\n",
      "Epoch 00073: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 438us/sample - loss: 3.7211e-05 - mse: 3.7211e-05 - val_loss: 8.3577e-05 - val_mse: 8.3577e-05\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 74/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7573e-05 - mse: 3.7573e-05\n",
      "Epoch 00074: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 465us/sample - loss: 3.7172e-05 - mse: 3.7172e-05 - val_loss: 8.3457e-05 - val_mse: 8.3457e-05\n",
      "Epoch 75/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7647e-05 - mse: 3.7647e-05\n",
      "Epoch 00075: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 392us/sample - loss: 3.7141e-05 - mse: 3.7141e-05 - val_loss: 8.3862e-05 - val_mse: 8.3862e-05\n",
      "Epoch 76/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.6512e-05 - mse: 3.6512e-05\n",
      "Epoch 00076: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 340us/sample - loss: 3.7112e-05 - mse: 3.7112e-05 - val_loss: 8.4832e-05 - val_mse: 8.4832e-05\n",
      "Epoch 77/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.6569e-05 - mse: 3.6569e-05\n",
      "Epoch 00077: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 471us/sample - loss: 3.7084e-05 - mse: 3.7084e-05 - val_loss: 8.4774e-05 - val_mse: 8.4774e-05\n",
      "Epoch 78/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7287e-05 - mse: 3.7287e-05\n",
      "Epoch 00078: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 488us/sample - loss: 3.7058e-05 - mse: 3.7058e-05 - val_loss: 8.5510e-05 - val_mse: 8.5510e-05\n",
      "Epoch 79/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7320e-05 - mse: 3.7320e-05\n",
      "Epoch 00079: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 474us/sample - loss: 3.7021e-05 - mse: 3.7021e-05 - val_loss: 8.4824e-05 - val_mse: 8.4824e-05\n",
      "Epoch 80/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.6931e-05 - mse: 3.6931e-05\n",
      "Epoch 00080: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 4s 521us/sample - loss: 3.7021e-05 - mse: 3.7021e-05 - val_loss: 8.5188e-05 - val_mse: 8.5188e-05\n",
      "Epoch 81/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7128e-05 - mse: 3.7128e-05\n",
      "Epoch 00081: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 4s 534us/sample - loss: 3.7012e-05 - mse: 3.7012e-05 - val_loss: 8.8004e-05 - val_mse: 8.8004e-05\n",
      "Epoch 82/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.6507e-05 - mse: 3.6507e-05\n",
      "Epoch 00082: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 448us/sample - loss: 3.6949e-05 - mse: 3.6949e-05 - val_loss: 8.5395e-05 - val_mse: 8.5395e-05\n",
      "Epoch 83/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.5686e-05 - mse: 3.5686e-05\n",
      "Epoch 00083: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 392us/sample - loss: 3.6921e-05 - mse: 3.6921e-05 - val_loss: 8.7067e-05 - val_mse: 8.7067e-05\n",
      "Epoch 84/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.6895e-05 - mse: 3.6895e-05\n",
      "Epoch 00084: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 410us/sample - loss: 3.6888e-05 - mse: 3.6888e-05 - val_loss: 8.7208e-05 - val_mse: 8.7208e-05\n",
      "Epoch 85/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.6182e-05 - mse: 3.6182e-05\n",
      "Epoch 00085: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 378us/sample - loss: 3.6861e-05 - mse: 3.6861e-05 - val_loss: 8.8351e-05 - val_mse: 8.8351e-05\n",
      "Epoch 86/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7108e-05 - mse: 3.7108e-05\n",
      "Epoch 00086: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 373us/sample - loss: 3.6845e-05 - mse: 3.6845e-05 - val_loss: 8.7640e-05 - val_mse: 8.7640e-05\n",
      "Epoch 87/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.6862e-05 - mse: 3.6862e-05\n",
      "Epoch 00087: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 288us/sample - loss: 3.6820e-05 - mse: 3.6820e-05 - val_loss: 8.9768e-05 - val_mse: 8.9768e-05\n",
      "Epoch 88/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.6547e-05 - mse: 3.6547e-05\n",
      "Epoch 00088: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 253us/sample - loss: 3.6785e-05 - mse: 3.6785e-05 - val_loss: 8.8472e-05 - val_mse: 8.8472e-05\n",
      "Epoch 89/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7756e-05 - mse: 3.7756e-05\n",
      "Epoch 00089: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 236us/sample - loss: 3.6778e-05 - mse: 3.6778e-05 - val_loss: 8.9108e-05 - val_mse: 8.9108e-05\n",
      "Epoch 90/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.6605e-05 - mse: 3.6605e-05\n",
      "Epoch 00090: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 319us/sample - loss: 3.6745e-05 - mse: 3.6745e-05 - val_loss: 9.0621e-05 - val_mse: 9.0621e-05\n",
      "Epoch 91/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.6716e-05 - mse: 3.6716e-05\n",
      "Epoch 00091: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 330us/sample - loss: 3.6724e-05 - mse: 3.6724e-05 - val_loss: 8.9091e-05 - val_mse: 8.9091e-05\n",
      "Epoch 92/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.6917e-05 - mse: 3.6917e-05\n",
      "Epoch 00092: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 334us/sample - loss: 3.6709e-05 - mse: 3.6709e-05 - val_loss: 9.0508e-05 - val_mse: 9.0508e-05\n",
      "Epoch 93/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.6482e-05 - mse: 3.6482e-05\n",
      "Epoch 00093: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 369us/sample - loss: 3.6677e-05 - mse: 3.6677e-05 - val_loss: 8.9292e-05 - val_mse: 8.9292e-05\n",
      "Epoch 94/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.6487e-05 - mse: 3.6487e-05\n",
      "Epoch 00094: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 295us/sample - loss: 3.6644e-05 - mse: 3.6644e-05 - val_loss: 9.2452e-05 - val_mse: 9.2452e-05\n",
      "Epoch 95/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.6671e-05 - mse: 3.6671e-05\n",
      "Epoch 00095: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 371us/sample - loss: 3.6657e-05 - mse: 3.6657e-05 - val_loss: 9.2990e-05 - val_mse: 9.2990e-05\n",
      "Epoch 96/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.7121e-05 - mse: 3.7121e-05\n",
      "Epoch 00096: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 4s 527us/sample - loss: 3.6589e-05 - mse: 3.6589e-05 - val_loss: 9.0678e-05 - val_mse: 9.0678e-05\n",
      "Epoch 97/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.6918e-05 - mse: 3.6918e-05\n",
      "Epoch 00097: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 434us/sample - loss: 3.6584e-05 - mse: 3.6584e-05 - val_loss: 9.1386e-05 - val_mse: 9.1386e-05\n",
      "Epoch 98/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.6932e-05 - mse: 3.6932e-05\n",
      "Epoch 00098: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 2s 312us/sample - loss: 3.6553e-05 - mse: 3.6553e-05 - val_loss: 9.3435e-05 - val_mse: 9.3435e-05\n",
      "Epoch 99/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.6578e-05 - mse: 3.6578e-05\n",
      "Epoch 00099: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 4s 568us/sample - loss: 3.6534e-05 - mse: 3.6534e-05 - val_loss: 9.3191e-05 - val_mse: 9.3191e-05\n",
      "Epoch 100/100\n",
      "6144/6835 [=========================>....] - ETA: 0s - loss: 3.6114e-05 - mse: 3.6114e-05\n",
      "Epoch 00100: val_loss did not improve from 0.00008\n",
      "6835/6835 [==============================] - 3s 475us/sample - loss: 3.6515e-05 - mse: 3.6515e-05 - val_loss: 9.3892e-05 - val_mse: 9.3892e-05\n"
     ]
    }
   ],
   "source": [
    "K.clear_session()\n",
    "\n",
    "# predict\n",
    "# model_word = seq2word_model()\n",
    "\n",
    "# history = model_word.fit(x_train, y_train, epochs=epochs,\n",
    "#                    validation_data=(x_val, y_val), \n",
    "#                    batch_size=batch_size, callbacks=callbacks_list_model)\n",
    "\n",
    "model_word = seq2word_model()\n",
    "\n",
    "history = model_word.fit(x_train, y_train, epochs=epochs,\n",
    "                   validation_data=(x_val, y_val), \n",
    "                   batch_size=batch_size, callbacks=callbacks_list_model)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 81,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_train_pred = model_seq.predict(x_train)\n",
    "y_val_pred = model_seq.predict(x_val)\n",
    "y_test_pred = model_seq.predict(x_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "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": [
    "# utils.plot_result(y_train, y_val, y_test, y_train_pred, y_val_pred, y_test_pred, ft=0)\n",
    "# y_train_pred = y_train_pred.reshape(-1, 4)\n",
    "# y_train_pred = y_train_pred[:, 0]\n",
    "# print('shape : ', y_train_pred.shape)\n",
    "# y_train = y_train.reshape(-1, 4)\n",
    "# y_train = y_train[:, 0]\n",
    "\n",
    "plt.plot(y_train_pred_trans, label='pred')\n",
    "plt.plot(y_train, label='actual')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/site-packages/ipykernel_launcher.py:2: FutureWarning: Method .as_matrix will be removed in a future version. Use .values instead.\n",
      "  \n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([[1, 4],\n",
       "       [2, 5],\n",
       "       [3, 6]])"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df = pd.DataFrame({\"eiei\":[1,2,3], \"yeah\":[4,5,6]})\n",
    "mat = df.as_matrix()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "x = np.array([[[1,2,3],[4,5,6]],\n",
    "[[7,8,9],[10,11,12]]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[[ 1,  2,  3],\n",
       "        [ 4,  5,  6]],\n",
       "\n",
       "       [[ 7,  8,  9],\n",
       "        [10, 11, 12]]])"
      ]
     },
     "execution_count": 50,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(2, 2, 3)"
      ]
     },
     "execution_count": 51,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 4,  5,  6],\n",
       "       [10, 11, 12]])"
      ]
     },
     "execution_count": 52,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x[:,-1,:]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.6.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}