model_02_LSTM_base_action_predict.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Testing stock\n",
    "\n",
    "* 時間, `最近成交價`, `當盤成交量`, `累積成交量`, `最佳五檔（賣價）`, `最佳五檔（賣量）`, `最佳五檔（買價）`, `最佳五檔（買量）`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "import re\n",
    "import numpy as np\n",
    "\n",
    "def getDataset(path):\n",
    "    dataset = []\n",
    "    with open(path, 'r') as fp:\n",
    "        for i in range(500):\n",
    "            data = fp.readline()\n",
    "            if len(data) == 0:\n",
    "                break\n",
    "            data = data.replace('-', '0')\n",
    "            data = re.split(',|\\n', data)[1:-1]\n",
    "            data[3] = data[3].split('_')[0]\n",
    "            data[4] = data[4].split('_')[0]\n",
    "            data[5] = data[5].split('_')[0]\n",
    "            data[6] = data[6].split('_')[0]\n",
    "            try:\n",
    "                dataset.append(np.asarray(data, dtype=np.float))\n",
    "            except ValueError:\n",
    "                print(data)\n",
    "                \n",
    "    return np.asarray(dataset)\n",
    "\n",
    "def countDataset(path):\n",
    "    return len(getDataset(path))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# get all paths and summary"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "500 1901\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import os\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "dataset_path = \"./dataset/Taiwan_5s/\"\n",
    "\n",
    "min_count = 65535\n",
    "choose = 500\n",
    "input_days = 15\n",
    "output_days = 1\n",
    "\n",
    "summary = [0] * 100\n",
    "paths = []\n",
    "\n",
    "for dirPath, dirNames, fileNames in os.walk(dataset_path):\n",
    "    if dirPath == dataset_path:\n",
    "        continue\n",
    "    total = 0\n",
    "    for datapath in fileNames:\n",
    "        count = countDataset(dirPath + '/' + datapath)\n",
    "        summary[count // 10 - 1] += 1\n",
    "        if count == choose:\n",
    "            paths.append(dirPath + '/' + datapath)\n",
    "print(choose, len(paths))\n",
    "# plot results\n",
    "plt.plot(summary, color = 'blue', label = 'count')\n",
    "plt.title('summary')\n",
    "plt.xlabel('range * 10')\n",
    "plt.ylabel('count')\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# My own MinMaxScaler"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "class MinMaxScaler:\n",
    "    __min = 0.\n",
    "    __max = 1.\n",
    "    __range = 1.\n",
    "    __feature_range = (0, 1)\n",
    "    __scale = 1.\n",
    "    def __init__(self):\n",
    "        pass\n",
    "    def getScalerData(self, dataset, offset=0.1, feature_range=(0, 1)):\n",
    "        data_max = np.max(dataset)\n",
    "        data_min = np.min(dataset)\n",
    "        if len(dataset) == 1:\n",
    "            range_temp = dataset * offset\n",
    "        else:\n",
    "            range_temp = (data_max - data_min) * (1 + offset)\n",
    "        self.__min = data_max - range_temp\n",
    "        self.__max = data_min + range_temp\n",
    "        self.__range = self.__max - self.__min\n",
    "        self.__feature_range = feature_range\n",
    "        self.__scale = (feature_range[1] - feature_range[0]) / self.__range\n",
    "        return self.getTransformData(dataset)\n",
    "    def getTransformData(self, dataset):\n",
    "        return (dataset - self.__min) * self.__scale + self.__feature_range[0]\n",
    "    def getInverseData(self, scalerDataset):\n",
    "        return (scalerDataset - self.__feature_range[0]) / self.__scale + self.__min\n",
    "    def getParameter(self):\n",
    "        return self.__min, self.__max, self.__range, self.__feature_range, self.__scale\n",
    "    def updatePatameter(self, parameter):\n",
    "        self.__min, self.__max, self.__range, self.__feature_range, self.__scale = parameter"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# create dataset\n",
    "* 1. price use MinMaxScaler with offset 0.1\n",
    "* 2. trading volume using local MinMaxScaler\n",
    "* 3. best 1 sell price use MinMaxScaler in 1.\n",
    "* 4. best 1 buy price use MinMaxScaler in 1.\n",
    "* 5. if sell count > buy count than > 0.5"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.keras.utils import to_categorical\n",
    "\n",
    "def create_dataset(path, input_days=60, output_days=1):\n",
    "    dataset = getDataset(path)\n",
    "    price_sc = MinMaxScaler()\n",
    "    price_sc.getScalerData(np.array([dataset[0][0]]), offset=0.1, feature_range=(0, 1))\n",
    "    prices = price_sc.getTransformData(dataset[:, [0, 3, 5]])\n",
    "    output = []\n",
    "    label = []\n",
    "    for i in range(input_days, choose - output_days + 1):\n",
    "        temp = dataset[i-input_days:i]\n",
    "        # create total set\n",
    "        temp_sc = MinMaxScaler()\n",
    "        a = prices[i-input_days:i]\n",
    "        diff = prices[i, 0] - prices[i-1, 0]\n",
    "        if diff > 0:\n",
    "            y = 2\n",
    "        elif diff < 0:\n",
    "            y = 0\n",
    "        else :\n",
    "            y = 1\n",
    "        b = temp_sc.getScalerData(temp[:, 1:2])\n",
    "        sell_count = temp[:, 4:5]\n",
    "        buy_count = temp[:, 6:7]\n",
    "        temp_scale = np.absolute(sell_count - buy_count) / np.maximum(sell_count, buy_count) / 2\n",
    "        c = np.where(sell_count > buy_count, 0.5+temp_scale, 0.5-temp_scale)\n",
    "        output.append(np.append(np.append(a, b, axis=-1), c, axis=-1))\n",
    "        label.append(y)\n",
    "    output = np.asarray(output, dtype=np.float)\n",
    "    label = to_categorical(np.asarray(label, dtype=np.float), num_classes=3)\n",
    "    return output, label, price_sc.getParameter()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\DVLAB\\AppData\\anaconda3\\envs\\dlib\\lib\\site-packages\\ipykernel_launcher.py:20: RuntimeWarning: divide by zero encountered in double_scalars\n",
      "C:\\Users\\DVLAB\\AppData\\anaconda3\\envs\\dlib\\lib\\site-packages\\ipykernel_launcher.py:23: RuntimeWarning: invalid value encountered in multiply\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(720710, 15, 5) (720710, 3)\n"
     ]
    }
   ],
   "source": [
    "x = np.empty((0, input_days, 5), dtype=np.float)\n",
    "y = np.empty((0, 3), dtype=np.float)\n",
    "\n",
    "for i, path in enumerate(paths):\n",
    "    if i > 1500:\n",
    "        break\n",
    "    if i < 15 :\n",
    "        continue\n",
    "    x_temp, y_temp, sc_parameter = create_dataset(path=path, input_days=input_days, output_days=output_days)\n",
    "    x = np.append(x, x_temp, axis=0)\n",
    "    y = np.append(y, y_temp, axis=0)\n",
    "\n",
    "print(x.shape, y.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "x[np.isnan(x)] = 0"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Create model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "RADAM = False\n",
    "if RADAM:\n",
    "    from keras_radam import RAdam\n",
    "    from keras_lookahead import Lookahead\n",
    "    optimizer = Lookahead(RAdam())\n",
    "else:\n",
    "    optimizer = 'adam'\n",
    "    \n",
    "period = 25\n",
    "epochs = 250\n",
    "batch_size = 3000"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From C:\\Users\\DVLAB\\AppData\\anaconda3\\envs\\dlib\\lib\\site-packages\\tensorflow\\python\\ops\\resource_variable_ops.py:435: colocate_with (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Colocations handled automatically by placer.\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "cu_dnnlstm (CuDNNLSTM)       (None, 15, 50)            11400     \n",
      "_________________________________________________________________\n",
      "cu_dnnlstm_1 (CuDNNLSTM)     (None, 15, 50)            20400     \n",
      "_________________________________________________________________\n",
      "cu_dnnlstm_2 (CuDNNLSTM)     (None, 15, 50)            20400     \n",
      "_________________________________________________________________\n",
      "flatten (Flatten)            (None, 750)               0         \n",
      "_________________________________________________________________\n",
      "dense (Dense)                (None, 128)               96128     \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 40)                5160      \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 3)                 123       \n",
      "=================================================================\n",
      "Total params: 153,611\n",
      "Trainable params: 153,611\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "# LSTM Training\n",
    "from tensorflow.keras.models import Sequential\n",
    "from tensorflow.keras.layers import CuDNNLSTM, Dense, Dropout, Flatten\n",
    "from tensorflow.keras.utils import plot_model\n",
    "\n",
    "model = Sequential()\n",
    "\n",
    "model.add(CuDNNLSTM(units = 50, return_sequences = True, input_shape = (input_days, 5)))\n",
    "model.add(CuDNNLSTM(units = 50, return_sequences = True))\n",
    "model.add(CuDNNLSTM(units = 50, return_sequences = True))\n",
    "model.add(Flatten())\n",
    "model.add(Dense(units = 128, activation='relu'))\n",
    "model.add(Dense(units = 40, activation='relu'))\n",
    "model.add(Dense(units = 3, activation='softmax'))\n",
    "model.compile(optimizer = optimizer, loss = 'categorical_crossentropy')\n",
    "model.summary()\n",
    "plot_model(model, 'model.png')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Training the model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 576568 samples, validate on 144142 samples\n",
      "WARNING:tensorflow:From C:\\Users\\DVLAB\\AppData\\anaconda3\\envs\\dlib\\lib\\site-packages\\tensorflow\\python\\ops\\math_ops.py:3066: to_int32 (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Use tf.cast instead.\n",
      "Epoch 1/250\n",
      "576568/576568 [==============================] - 8s 13us/sample - loss: 0.7197 - val_loss: 0.7338\n",
      "Epoch 2/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.7076 - val_loss: 0.7305\n",
      "Epoch 3/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.7053 - val_loss: 0.7294\n",
      "Epoch 4/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.7035 - val_loss: 0.7267\n",
      "Epoch 5/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.7028 - val_loss: 0.7260\n",
      "Epoch 6/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.7017 - val_loss: 0.7256\n",
      "Epoch 7/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.7010 - val_loss: 0.7253\n",
      "Epoch 8/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.7008 - val_loss: 0.7249\n",
      "Epoch 9/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.7008 - val_loss: 0.7249\n",
      "Epoch 10/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.7004 - val_loss: 0.7240\n",
      "Epoch 11/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.7004 - val_loss: 0.7247\n",
      "Epoch 12/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.6997 - val_loss: 0.7236\n",
      "Epoch 13/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.6995 - val_loss: 0.7249\n",
      "Epoch 14/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.6994 - val_loss: 0.7232\n",
      "Epoch 15/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.6991 - val_loss: 0.7231\n",
      "Epoch 16/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.6986 - val_loss: 0.7217\n",
      "Epoch 17/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.6983 - val_loss: 0.7223\n",
      "Epoch 18/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.6981 - val_loss: 0.7212\n",
      "Epoch 19/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.6978 - val_loss: 0.7216\n",
      "Epoch 20/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.6979 - val_loss: 0.7208\n",
      "Epoch 21/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.6975 - val_loss: 0.7220\n",
      "Epoch 22/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.6973 - val_loss: 0.7219\n",
      "Epoch 23/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.6970 - val_loss: 0.7210\n",
      "Epoch 24/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.6969 - val_loss: 0.7199\n",
      "Epoch 25/250\n",
      "564000/576568 [============================>.] - ETA: 0s - loss: 0.6964\n",
      "Epoch 00025: saving model to ./model/model_02_check_point/cp-0025.ckpt\n",
      "WARNING:tensorflow:This model was compiled with a Keras optimizer (<tensorflow.python.keras.optimizers.Adam object at 0x000001E0F0C24D30>) but is being saved in TensorFlow format with `save_weights`. The model's weights will be saved, but unlike with TensorFlow optimizers in the TensorFlow format the optimizer's state will not be saved.\n",
      "\n",
      "Consider using a TensorFlow optimizer from `tf.train`.\n",
      "WARNING:tensorflow:From C:\\Users\\DVLAB\\AppData\\anaconda3\\envs\\dlib\\lib\\site-packages\\tensorflow\\python\\keras\\engine\\network.py:1436: update_checkpoint_state (from tensorflow.python.training.checkpoint_management) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Use tf.train.CheckpointManager to manage checkpoints rather than manually editing the Checkpoint proto.\n",
      "576568/576568 [==============================] - 3s 6us/sample - loss: 0.6967 - val_loss: 0.7202\n",
      "Epoch 26/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.6961 - val_loss: 0.7196\n",
      "Epoch 27/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.6956 - val_loss: 0.7222\n",
      "Epoch 28/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.6900 - val_loss: 0.6938\n",
      "Epoch 29/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.6343 - val_loss: 0.6263\n",
      "Epoch 30/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.6025 - val_loss: 0.6502\n",
      "Epoch 31/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5961 - val_loss: 0.6004\n",
      "Epoch 32/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5895 - val_loss: 0.5984\n",
      "Epoch 33/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5863 - val_loss: 0.6023\n",
      "Epoch 34/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5838 - val_loss: 0.6021\n",
      "Epoch 35/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5821 - val_loss: 0.5938\n",
      "Epoch 36/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5794 - val_loss: 0.6064\n",
      "Epoch 37/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5800 - val_loss: 0.5946\n",
      "Epoch 38/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5785 - val_loss: 0.5926\n",
      "Epoch 39/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5781 - val_loss: 0.5876\n",
      "Epoch 40/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5757 - val_loss: 0.5896\n",
      "Epoch 41/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5758 - val_loss: 0.5893\n",
      "Epoch 42/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5737 - val_loss: 0.5894\n",
      "Epoch 43/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5741 - val_loss: 0.5863\n",
      "Epoch 44/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5722 - val_loss: 0.5891\n",
      "Epoch 45/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5720 - val_loss: 0.5862\n",
      "Epoch 46/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5717 - val_loss: 0.5873\n",
      "Epoch 47/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5708 - val_loss: 0.5871\n",
      "Epoch 48/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5699 - val_loss: 0.5861\n",
      "Epoch 49/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5693 - val_loss: 0.5860\n",
      "Epoch 50/250\n",
      "576000/576568 [============================>.] - ETA: 0s - loss: 0.5694\n",
      "Epoch 00050: saving model to ./model/model_02_check_point/cp-0050.ckpt\n",
      "WARNING:tensorflow:This model was compiled with a Keras optimizer (<tensorflow.python.keras.optimizers.Adam object at 0x000001E0F0C24D30>) but is being saved in TensorFlow format with `save_weights`. The model's weights will be saved, but unlike with TensorFlow optimizers in the TensorFlow format the optimizer's state will not be saved.\n",
      "\n",
      "Consider using a TensorFlow optimizer from `tf.train`.\n",
      "576568/576568 [==============================] - 3s 6us/sample - loss: 0.5694 - val_loss: 0.5832\n",
      "Epoch 51/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5680 - val_loss: 0.5838\n",
      "Epoch 52/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5677 - val_loss: 0.5883\n",
      "Epoch 53/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5673 - val_loss: 0.5886\n",
      "Epoch 54/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5665 - val_loss: 0.5905\n",
      "Epoch 55/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5669 - val_loss: 0.5924\n",
      "Epoch 56/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5658 - val_loss: 0.5881\n",
      "Epoch 57/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5653 - val_loss: 0.5853\n",
      "Epoch 58/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5650 - val_loss: 0.5863\n",
      "Epoch 59/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5644 - val_loss: 0.5892\n",
      "Epoch 60/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5626 - val_loss: 0.5919\n",
      "Epoch 61/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5637 - val_loss: 0.5849\n",
      "Epoch 62/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5623 - val_loss: 0.5926\n",
      "Epoch 63/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5614 - val_loss: 0.5865\n",
      "Epoch 64/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5617 - val_loss: 0.5913\n",
      "Epoch 65/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5607 - val_loss: 0.5917\n",
      "Epoch 66/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5606 - val_loss: 0.5884\n",
      "Epoch 67/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5594 - val_loss: 0.6063\n",
      "Epoch 68/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5608 - val_loss: 0.5891\n",
      "Epoch 69/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5587 - val_loss: 0.5868\n",
      "Epoch 70/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5577 - val_loss: 0.5870\n",
      "Epoch 71/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5573 - val_loss: 0.5937\n",
      "Epoch 72/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5561 - val_loss: 0.5947\n",
      "Epoch 73/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5556 - val_loss: 0.5895\n",
      "Epoch 74/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5554 - val_loss: 0.5888\n",
      "Epoch 75/250\n",
      "567000/576568 [============================>.] - ETA: 0s - loss: 0.5541\n",
      "Epoch 00075: saving model to ./model/model_02_check_point/cp-0075.ckpt\n",
      "WARNING:tensorflow:This model was compiled with a Keras optimizer (<tensorflow.python.keras.optimizers.Adam object at 0x000001E0F0C24D30>) but is being saved in TensorFlow format with `save_weights`. The model's weights will be saved, but unlike with TensorFlow optimizers in the TensorFlow format the optimizer's state will not be saved.\n",
      "\n",
      "Consider using a TensorFlow optimizer from `tf.train`.\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5541 - val_loss: 0.5883\n",
      "Epoch 76/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5537 - val_loss: 0.5907\n",
      "Epoch 77/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5527 - val_loss: 0.5916\n",
      "Epoch 78/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5523 - val_loss: 0.5957\n",
      "Epoch 79/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5511 - val_loss: 0.5926\n",
      "Epoch 80/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5509 - val_loss: 0.5937\n",
      "Epoch 81/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5492 - val_loss: 0.5909\n",
      "Epoch 82/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5488 - val_loss: 0.6021\n",
      "Epoch 83/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5487 - val_loss: 0.5956\n",
      "Epoch 84/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5468 - val_loss: 0.5964\n",
      "Epoch 85/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5468 - val_loss: 0.6106\n",
      "Epoch 86/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5457 - val_loss: 0.5974\n",
      "Epoch 87/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5451 - val_loss: 0.5945\n",
      "Epoch 88/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5428 - val_loss: 0.5939\n",
      "Epoch 89/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5425 - val_loss: 0.5987\n",
      "Epoch 90/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5420 - val_loss: 0.5967\n",
      "Epoch 91/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5411 - val_loss: 0.6014\n",
      "Epoch 92/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5400 - val_loss: 0.6001\n",
      "Epoch 93/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5390 - val_loss: 0.6016\n",
      "Epoch 94/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5380 - val_loss: 0.6051\n",
      "Epoch 95/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5373 - val_loss: 0.6036\n",
      "Epoch 96/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5360 - val_loss: 0.6027\n",
      "Epoch 97/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5347 - val_loss: 0.6038\n",
      "Epoch 98/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5337 - val_loss: 0.6047\n",
      "Epoch 99/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5323 - val_loss: 0.6050\n",
      "Epoch 100/250\n",
      "576000/576568 [============================>.] - ETA: 0s - loss: 0.5309\n",
      "Epoch 00100: saving model to ./model/model_02_check_point/cp-0100.ckpt\n",
      "WARNING:tensorflow:This model was compiled with a Keras optimizer (<tensorflow.python.keras.optimizers.Adam object at 0x000001E0F0C24D30>) but is being saved in TensorFlow format with `save_weights`. The model's weights will be saved, but unlike with TensorFlow optimizers in the TensorFlow format the optimizer's state will not be saved.\n",
      "\n",
      "Consider using a TensorFlow optimizer from `tf.train`.\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5309 - val_loss: 0.6076\n",
      "Epoch 101/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5312 - val_loss: 0.6105\n",
      "Epoch 102/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5297 - val_loss: 0.6072\n",
      "Epoch 103/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5287 - val_loss: 0.6183\n",
      "Epoch 104/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5268 - val_loss: 0.6187\n",
      "Epoch 105/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5270 - val_loss: 0.6157\n",
      "Epoch 106/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5262 - val_loss: 0.6149\n",
      "Epoch 107/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5251 - val_loss: 0.6150\n",
      "Epoch 108/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5231 - val_loss: 0.6166\n",
      "Epoch 109/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5217 - val_loss: 0.6247\n",
      "Epoch 110/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5216 - val_loss: 0.6193\n",
      "Epoch 111/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5210 - val_loss: 0.6184\n",
      "Epoch 112/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5186 - val_loss: 0.6243\n",
      "Epoch 113/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5170 - val_loss: 0.6194\n",
      "Epoch 114/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5167 - val_loss: 0.6360\n",
      "Epoch 115/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5171 - val_loss: 0.6233\n",
      "Epoch 116/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5141 - val_loss: 0.6267\n",
      "Epoch 117/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5133 - val_loss: 0.6446\n",
      "Epoch 118/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5128 - val_loss: 0.6314\n",
      "Epoch 119/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5114 - val_loss: 0.6288\n",
      "Epoch 120/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5109 - val_loss: 0.6298\n",
      "Epoch 121/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5099 - val_loss: 0.6327\n",
      "Epoch 122/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5076 - val_loss: 0.6317\n",
      "Epoch 123/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5068 - val_loss: 0.6368\n",
      "Epoch 124/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5058 - val_loss: 0.6393\n",
      "Epoch 125/250\n",
      "573000/576568 [============================>.] - ETA: 0s - loss: 0.5045\n",
      "Epoch 00125: saving model to ./model/model_02_check_point/cp-0125.ckpt\n",
      "WARNING:tensorflow:This model was compiled with a Keras optimizer (<tensorflow.python.keras.optimizers.Adam object at 0x000001E0F0C24D30>) but is being saved in TensorFlow format with `save_weights`. The model's weights will be saved, but unlike with TensorFlow optimizers in the TensorFlow format the optimizer's state will not be saved.\n",
      "\n",
      "Consider using a TensorFlow optimizer from `tf.train`.\n",
      "576568/576568 [==============================] - 3s 6us/sample - loss: 0.5046 - val_loss: 0.6360\n",
      "Epoch 126/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5036 - val_loss: 0.6411\n",
      "Epoch 127/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5030 - val_loss: 0.6402\n",
      "Epoch 128/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5025 - val_loss: 0.6446\n",
      "Epoch 129/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4996 - val_loss: 0.6431\n",
      "Epoch 130/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.5012 - val_loss: 0.6436\n",
      "Epoch 131/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4998 - val_loss: 0.6452\n",
      "Epoch 132/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4982 - val_loss: 0.6461\n",
      "Epoch 133/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4967 - val_loss: 0.6480\n",
      "Epoch 134/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4953 - val_loss: 0.6493\n",
      "Epoch 135/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4947 - val_loss: 0.6561\n",
      "Epoch 136/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4954 - val_loss: 0.6525\n",
      "Epoch 137/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4922 - val_loss: 0.6506\n",
      "Epoch 138/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4911 - val_loss: 0.6575\n",
      "Epoch 139/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4899 - val_loss: 0.6577\n",
      "Epoch 140/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4909 - val_loss: 0.6577\n",
      "Epoch 141/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4895 - val_loss: 0.6613\n",
      "Epoch 142/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4896 - val_loss: 0.6642\n",
      "Epoch 143/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4859 - val_loss: 0.6731\n",
      "Epoch 144/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4863 - val_loss: 0.6639\n",
      "Epoch 145/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4852 - val_loss: 0.6655\n",
      "Epoch 146/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4846 - val_loss: 0.6666\n",
      "Epoch 147/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4826 - val_loss: 0.6728\n",
      "Epoch 148/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4822 - val_loss: 0.6728\n",
      "Epoch 149/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4805 - val_loss: 0.6718\n",
      "Epoch 150/250\n",
      "567000/576568 [============================>.] - ETA: 0s - loss: 0.4792\n",
      "Epoch 00150: saving model to ./model/model_02_check_point/cp-0150.ckpt\n",
      "WARNING:tensorflow:This model was compiled with a Keras optimizer (<tensorflow.python.keras.optimizers.Adam object at 0x000001E0F0C24D30>) but is being saved in TensorFlow format with `save_weights`. The model's weights will be saved, but unlike with TensorFlow optimizers in the TensorFlow format the optimizer's state will not be saved.\n",
      "\n",
      "Consider using a TensorFlow optimizer from `tf.train`.\n",
      "576568/576568 [==============================] - 3s 6us/sample - loss: 0.4792 - val_loss: 0.6739\n",
      "Epoch 151/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4797 - val_loss: 0.6815\n",
      "Epoch 152/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4785 - val_loss: 0.6787\n",
      "Epoch 153/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4789 - val_loss: 0.6798\n",
      "Epoch 154/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4782 - val_loss: 0.6803\n",
      "Epoch 155/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4757 - val_loss: 0.6815\n",
      "Epoch 156/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4744 - val_loss: 0.6849\n",
      "Epoch 157/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4738 - val_loss: 0.6895\n",
      "Epoch 158/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4722 - val_loss: 0.6853\n",
      "Epoch 159/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4710 - val_loss: 0.6873\n",
      "Epoch 160/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4706 - val_loss: 0.6923\n",
      "Epoch 161/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4721 - val_loss: 0.6954\n",
      "Epoch 162/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4692 - val_loss: 0.6922\n",
      "Epoch 163/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4683 - val_loss: 0.7020\n",
      "Epoch 164/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4681 - val_loss: 0.6954\n",
      "Epoch 165/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4680 - val_loss: 0.6970\n",
      "Epoch 166/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4653 - val_loss: 0.6947\n",
      "Epoch 167/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4652 - val_loss: 0.7036\n",
      "Epoch 168/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4650 - val_loss: 0.7103\n",
      "Epoch 169/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4625 - val_loss: 0.7078\n",
      "Epoch 170/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4640 - val_loss: 0.7033\n",
      "Epoch 171/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4633 - val_loss: 0.7060\n",
      "Epoch 172/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4620 - val_loss: 0.7105\n",
      "Epoch 173/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4606 - val_loss: 0.7152\n",
      "Epoch 174/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4580 - val_loss: 0.7128\n",
      "Epoch 175/250\n",
      "576000/576568 [============================>.] - ETA: 0s - loss: 0.4588\n",
      "Epoch 00175: saving model to ./model/model_02_check_point/cp-0175.ckpt\n",
      "WARNING:tensorflow:This model was compiled with a Keras optimizer (<tensorflow.python.keras.optimizers.Adam object at 0x000001E0F0C24D30>) but is being saved in TensorFlow format with `save_weights`. The model's weights will be saved, but unlike with TensorFlow optimizers in the TensorFlow format the optimizer's state will not be saved.\n",
      "\n",
      "Consider using a TensorFlow optimizer from `tf.train`.\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4589 - val_loss: 0.7187\n",
      "Epoch 176/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4580 - val_loss: 0.7170\n",
      "Epoch 177/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4591 - val_loss: 0.7139\n",
      "Epoch 178/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4580 - val_loss: 0.7180\n",
      "Epoch 179/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4544 - val_loss: 0.7238\n",
      "Epoch 180/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4558 - val_loss: 0.7222\n",
      "Epoch 181/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4538 - val_loss: 0.7180\n",
      "Epoch 182/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4541 - val_loss: 0.7213\n",
      "Epoch 183/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4531 - val_loss: 0.7299\n",
      "Epoch 184/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4506 - val_loss: 0.7302\n",
      "Epoch 185/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4523 - val_loss: 0.7327\n",
      "Epoch 186/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4517 - val_loss: 0.7321\n",
      "Epoch 187/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4511 - val_loss: 0.7267\n",
      "Epoch 188/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4487 - val_loss: 0.7389\n",
      "Epoch 189/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4503 - val_loss: 0.7332\n",
      "Epoch 190/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4484 - val_loss: 0.7342\n",
      "Epoch 191/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4471 - val_loss: 0.7407\n",
      "Epoch 192/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4452 - val_loss: 0.7344\n",
      "Epoch 193/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4453 - val_loss: 0.7486\n",
      "Epoch 194/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4446 - val_loss: 0.7511\n",
      "Epoch 195/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4451 - val_loss: 0.7411\n",
      "Epoch 196/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4447 - val_loss: 0.7493\n",
      "Epoch 197/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4442 - val_loss: 0.7474\n",
      "Epoch 198/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4408 - val_loss: 0.7534\n",
      "Epoch 199/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4412 - val_loss: 0.7543\n",
      "Epoch 200/250\n",
      "567000/576568 [============================>.] - ETA: 0s - loss: 0.4399\n",
      "Epoch 00200: saving model to ./model/model_02_check_point/cp-0200.ckpt\n",
      "WARNING:tensorflow:This model was compiled with a Keras optimizer (<tensorflow.python.keras.optimizers.Adam object at 0x000001E0F0C24D30>) but is being saved in TensorFlow format with `save_weights`. The model's weights will be saved, but unlike with TensorFlow optimizers in the TensorFlow format the optimizer's state will not be saved.\n",
      "\n",
      "Consider using a TensorFlow optimizer from `tf.train`.\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4400 - val_loss: 0.7581\n",
      "Epoch 201/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4413 - val_loss: 0.7549\n",
      "Epoch 202/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4429 - val_loss: 0.7578\n",
      "Epoch 203/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4400 - val_loss: 0.7610\n",
      "Epoch 204/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4386 - val_loss: 0.7595\n",
      "Epoch 205/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4388 - val_loss: 0.7591\n",
      "Epoch 206/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4369 - val_loss: 0.7621\n",
      "Epoch 207/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4374 - val_loss: 0.7642\n",
      "Epoch 208/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4354 - val_loss: 0.7690\n",
      "Epoch 209/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4357 - val_loss: 0.7684\n",
      "Epoch 210/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4362 - val_loss: 0.7692\n",
      "Epoch 211/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4351 - val_loss: 0.7806\n",
      "Epoch 212/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4332 - val_loss: 0.7739\n",
      "Epoch 213/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4342 - val_loss: 0.7686\n",
      "Epoch 214/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4324 - val_loss: 0.7776\n",
      "Epoch 215/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4319 - val_loss: 0.7763\n",
      "Epoch 216/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4300 - val_loss: 0.7871\n",
      "Epoch 217/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4324 - val_loss: 0.7813\n",
      "Epoch 218/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4299 - val_loss: 0.7862\n",
      "Epoch 219/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4304 - val_loss: 0.7856\n",
      "Epoch 220/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4282 - val_loss: 0.7933\n",
      "Epoch 221/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4291 - val_loss: 0.7842\n",
      "Epoch 222/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4283 - val_loss: 0.7867\n",
      "Epoch 223/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4283 - val_loss: 0.7926\n",
      "Epoch 224/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4259 - val_loss: 0.7993\n",
      "Epoch 225/250\n",
      "576000/576568 [============================>.] - ETA: 0s - loss: 0.4258\n",
      "Epoch 00225: saving model to ./model/model_02_check_point/cp-0225.ckpt\n",
      "WARNING:tensorflow:This model was compiled with a Keras optimizer (<tensorflow.python.keras.optimizers.Adam object at 0x000001E0F0C24D30>) but is being saved in TensorFlow format with `save_weights`. The model's weights will be saved, but unlike with TensorFlow optimizers in the TensorFlow format the optimizer's state will not be saved.\n",
      "\n",
      "Consider using a TensorFlow optimizer from `tf.train`.\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4258 - val_loss: 0.7996\n",
      "Epoch 226/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4279 - val_loss: 0.7968\n",
      "Epoch 227/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4250 - val_loss: 0.7993\n",
      "Epoch 228/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4252 - val_loss: 0.8061\n",
      "Epoch 229/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4250 - val_loss: 0.8093\n",
      "Epoch 230/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4242 - val_loss: 0.7989\n",
      "Epoch 231/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4232 - val_loss: 0.8114\n",
      "Epoch 232/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4231 - val_loss: 0.8017\n",
      "Epoch 233/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4227 - val_loss: 0.8044\n",
      "Epoch 234/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4223 - val_loss: 0.8054\n",
      "Epoch 235/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4215 - val_loss: 0.8078\n",
      "Epoch 236/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4192 - val_loss: 0.8101\n",
      "Epoch 237/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4197 - val_loss: 0.8173\n",
      "Epoch 238/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4203 - val_loss: 0.8146\n",
      "Epoch 239/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4203 - val_loss: 0.8131\n",
      "Epoch 240/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4178 - val_loss: 0.8146\n",
      "Epoch 241/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4181 - val_loss: 0.8150\n",
      "Epoch 242/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4185 - val_loss: 0.8201\n",
      "Epoch 243/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4184 - val_loss: 0.8273\n",
      "Epoch 244/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4157 - val_loss: 0.8341\n",
      "Epoch 245/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4160 - val_loss: 0.8274\n",
      "Epoch 246/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4165 - val_loss: 0.8227\n",
      "Epoch 247/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4164 - val_loss: 0.8328\n",
      "Epoch 248/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4152 - val_loss: 0.8369\n",
      "Epoch 249/250\n",
      "576568/576568 [==============================] - 3s 5us/sample - loss: 0.4151 - val_loss: 0.8359\n",
      "Epoch 250/250\n",
      "567000/576568 [============================>.] - ETA: 0s - loss: 0.4155\n",
      "Epoch 00250: saving model to ./model/model_02_check_point/cp-0250.ckpt\n",
      "WARNING:tensorflow:This model was compiled with a Keras optimizer (<tensorflow.python.keras.optimizers.Adam object at 0x000001E0F0C24D30>) but is being saved in TensorFlow format with `save_weights`. The model's weights will be saved, but unlike with TensorFlow optimizers in the TensorFlow format the optimizer's state will not be saved.\n",
      "\n",
      "Consider using a TensorFlow optimizer from `tf.train`.\n",
      "576568/576568 [==============================] - 3s 6us/sample - loss: 0.4156 - val_loss: 0.8259\n"
     ]
    }
   ],
   "source": [
    "from tensorflow.keras.callbacks import ModelCheckpoint\n",
    "import os\n",
    "name = 'model_02'\n",
    "checkpoint_file = './model/' + name + '_check_point/cp-{epoch:04d}.ckpt'\n",
    "try:\n",
    "    os.mkdir('./model/' + name + '_check_point/')\n",
    "except:\n",
    "    pass\n",
    "\n",
    "# create callback function\n",
    "cp_callback = ModelCheckpoint(checkpoint_file, save_weights_only=True, verbose=1, period=period)\n",
    "\n",
    "# train the model\n",
    "train = model.fit(x, y, epochs=epochs, batch_size=batch_size, callbacks=[cp_callback], validation_split=0.2)\n",
    "\n",
    "# save model\n",
    "model.save('./model/' + name + '.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "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": [
    "plt.plot(train.history['loss'])\n",
    "plt.plot(train.history['val_loss'])\n",
    "plt.title('model loss')\n",
    "plt.ylabel('loss')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'test'], loc='upper right')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(485,) [1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1\n",
      " 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1\n",
      " 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1\n",
      " 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1\n",
      " 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 1 1 1 1 1 1 1 1 1 1 1 1 1\n",
      " 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1\n",
      " 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 1 1 1 1 1\n",
      " 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1\n",
      " 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 1 1\n",
      " 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1\n",
      " 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1\n",
      " 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1\n",
      " 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1\n",
      " 1 1 1 1]\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\DVLAB\\AppData\\anaconda3\\envs\\dlib\\lib\\site-packages\\ipykernel_launcher.py:20: RuntimeWarning: divide by zero encountered in double_scalars\n",
      "C:\\Users\\DVLAB\\AppData\\anaconda3\\envs\\dlib\\lib\\site-packages\\ipykernel_launcher.py:23: RuntimeWarning: invalid value encountered in multiply\n"
     ]
    }
   ],
   "source": [
    "model.load_weights('./model/model_02_check_point/cp-{epoch:04d}.ckpt'.format(epoch=50))\n",
    "test_x, test_y, sc_parameter = create_dataset(path=paths[1550], input_days=input_days, output_days=output_days)\n",
    "test_x[np.isnan(test_x)] = 0\n",
    "pred = np.argmax(model.predict(test_x), axis=-1)\n",
    "print(pred.shape, pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(485,)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "test_sc = MinMaxScaler()\n",
    "test_sc.updatePatameter(sc_parameter)\n",
    "test_real = test_x[:, -1, 0]\n",
    "test_real = test_sc.getInverseData(test_real)\n",
    "output_action = pred\n",
    "print(test_real.shape)\n",
    "\n",
    "# Visualising the results\n",
    "plot = test_real\n",
    "sell = np.where(output_action==0)\n",
    "do_nothing = np.where(output_action==1)\n",
    "buy = np.where(output_action==2)\n",
    "plt.plot(plot, color = 'blue', label = 'Real')\n",
    "plt.scatter(sell, plot[sell], color = 'green', label = 'sell')\n",
    "plt.scatter(do_nothing, plot[do_nothing], color = 'gray', label = 'do nothing')\n",
    "plt.scatter(buy, plot[buy], color = 'red', label = 'buy')\n",
    "# plt.plot(output_prices, color = 'blue', label = 'Predicted Google Stock Price')\n",
    "plt.title('Prediction')\n",
    "plt.xlabel('Time(days)')\n",
    "plt.ylabel('Real')\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [conda env:dlib]",
   "language": "python",
   "name": "conda-env-dlib-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}