LSTM Autoencoder_mse_kld(bfs)-for ordinal.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import glob, os\n",
    "from keras import layers as ly\n",
    "from keras.models import Sequential, load_model, Model\n",
    "from keras.layers import LSTM, Dense, RepeatVector, TimeDistributed\n",
    "from keras.preprocessing import sequence\n",
    "from sklearn.model_selection import train_test_split\n",
    "from keras.optimizers import Adam, SGD\n",
    "from keras.layers import Input, Lambda\n",
    "from keras import losses\n",
    "from keras.models import model_from_json\n",
    "import keras.backend.tensorflow_backend as K\n",
    "from keras.callbacks import ModelCheckpoint, EarlyStopping\n",
    "import string, json\n",
    "from keras.utils import plot_model\n",
    "import random as rd"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "#dir = './sequence/*'\n",
    "dir = './create/datasets/seq2/*'\n",
    "alpha = list(string.ascii_uppercase)\n",
    "chr2index = {alpha[i]:i for i in range(len(alpha))}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "def chr2OH(alphabet):\n",
    "    oh = [0 for i in range(len(alpha))]\n",
    "    index = chr2index[alphabet]\n",
    "    oh[index] = 1\n",
    "    return oh"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [],
   "source": [
    "#idx = [i for i in range(1, 101)]\n",
    "#rd.shuffle(idx)\n",
    "train_idx = [48, 12, 92, 60, 97, 28, 10, 90, 85, 94, 57, 54, 83, 88, 42, 51, 53, 84, 21, 8, 65, 39, 78, 18, 76, 14, 3, 70, 4, 64, 72, 62, 71, 59, 26, 99, 2, 37, 58, 46, 31, 1, 49, 5, 89, 75, 56, 73, 32, 11, 100, 20, 22, 91, 96, 17, 93, 35, 19, 13, 82, 87, 98, 95, 29, 81, 66, 67, 52, 24] #train_idx = idx[:70]\n",
    "test_idx = [47, 6, 43, 80, 33, 61, 15, 38, 79, 30, 34, 77, 36, 45, 25, 23, 68, 63, 86, 41, 9, 16, 44, 69, 27, 7, 40, 74, 55, 50]#test_idx = idx[70:]\n",
    "#print(train_idx, test_idx)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "# file read\n",
    "all_names = []\n",
    "all_data = []\n",
    "sequence_length = []\n",
    "alpha = list(string.ascii_uppercase)\n",
    "data_length = len(glob.glob(dir))\n",
    "file_predix = './create/datasets/seq2/group'\n",
    "for index in range(1, data_length+1):\n",
    "    #print(index)\n",
    "    prefix = file_predix + str(index) + \"/\"\n",
    "    #files = glob.glob(filename)\n",
    "    #print(files)\n",
    "    \n",
    "    for idx in [i for i in range(1, 101)]:\n",
    "        filename = prefix + str(index) + \"graph\" + str(idx)\n",
    "        files = glob.glob(filename + \"-*\")\n",
    "        for file in files:\n",
    "            datasets = []\n",
    "            all_names.append(file.split('/')[-1].replace('.txt', ''))\n",
    "            for rf in open(file, 'r'):\n",
    "                (u, v, w) = rf[1:-2].split(', ')\n",
    "                #datasets.append(chr2OH(u[1:-1]) + chr2OH(v[1:-1]) +[float(w)])\n",
    "                datasets.append([chr2index[u[1:-1]], chr2index[v[1:-1]], float(w)])\n",
    "            sequence_length.append(len(datasets))\n",
    "            all_data.append(datasets)\n",
    "#print(all_names)\n",
    "all_data = np.array([np.array(arr) for arr in all_data])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'\\n#x_train, train_name = all_data, all_names\\nx_train,  tmp, train_name, tmp_name = train_test_split(all_data, all_names, test_size = 0.1)\\nx_train = np.concatenate((x_train, tmp))\\ntrain_name = np.concatenate((train_name, tmp_name))\\n'"
      ]
     },
     "execution_count": 53,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "\n",
    "x_train, x_test, train_name, test_name = train_test_split(all_data, all_names, test_size=0.3)\n",
    "x_test, x_val, test_name, val_name = train_test_split(x_test, test_name, test_size=0.33)\n",
    "'''\n",
    "#x_train, train_name = all_data, all_names\n",
    "x_train,  tmp, train_name, tmp_name = train_test_split(all_data, all_names, test_size = 0.1)\n",
    "x_train = np.concatenate((x_train, tmp))\n",
    "train_name = np.concatenate((train_name, tmp_name))\n",
    "'''"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [],
   "source": [
    "tr_names= []\n",
    "for name in train_name:\n",
    "    tr_names.append(name.split('graph')[0])\n",
    "    #tr_names.append(name.split('-')[0].replace('graph', ''))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [],
   "source": [
    "max_sequence_length = max(sequence_length)\n",
    "n_features = 3\n",
    "batch_size = 32\n",
    "epochs = 200\n",
    "steps_per_epoch = len(x_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "13"
      ]
     },
     "execution_count": 56,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "max_sequence_length"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [],
   "source": [
    "def custom_loss(y_true, y_pred):\n",
    "    loss1 = losses.mean_squared_error(y_true, y_pred)\n",
    "    loss2 = losses.kld(y_true, y_pred) # categorical_crossentropy\n",
    "    return loss1 * 0.7 + loss2 * 0.3\n",
    "\n",
    "def custom_loss1(y_true, y_pred):\n",
    "    loss1 = losses.mean_squared_error(y_true, y_pred)\n",
    "    loss2 = losses.categorical_crossentropy(y_true, y_pred) # categorical_crossentropy\n",
    "    return loss1 * 0.7 + loss2 * 0.3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [],
   "source": [
    "def val_generator(x_val):\n",
    "    idx = 0\n",
    "    while True:\n",
    "        yield np.array([x_val[idx]]), np.array([x_val[idx]])\n",
    "        idx += 1\n",
    "        if idx >= len(x_val):\n",
    "            idx = 0\n",
    "\n",
    "def train_generator(x_train):\n",
    "    idx = 0\n",
    "    while True:\n",
    "        yield np.array([x_train[idx]]), np.array([x_train[idx]])\n",
    "        idx += 1\n",
    "        if idx >= len(x_train):\n",
    "            idx = 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "import random as rd\n",
    "rd.shuffle(x_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "def repeat_vector(args):\n",
    "    layer_to_repeat = args[0]\n",
    "    sequence_layer = args[1]\n",
    "    return RepeatVector(K.shape(sequence_layer)[1])(layer_to_repeat)\n",
    "    \n",
    "\n",
    "inputs = Input(shape=(None, 3)) #53\n",
    "encoded = LSTM(128, return_sequences=True)(inputs)  #activation 안적으면 tanh\n",
    "encoded = LSTM(64)(encoded)\n",
    "\n",
    "decoded = Lambda(repeat_vector, output_shape=(None, 64)) ([encoded, inputs]) # inputs의 shape[1] 만큼 encoded 를 반복 생성\n",
    "\n",
    "decoded = LSTM(64, return_sequences=True)(decoded)\n",
    "decoded = LSTM(128, return_sequences=True)(decoded)\n",
    "decoded = TimeDistributed(Dense(3))(decoded) #53"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/200\n",
      "4297/4297 [==============================] - 55s 13ms/step - loss: 304.2329 - val_loss: 191.9525\n",
      "Epoch 2/200\n",
      "4297/4297 [==============================] - 61s 14ms/step - loss: 201.8652 - val_loss: 189.9991\n",
      "Epoch 3/200\n",
      "4297/4297 [==============================] - 68s 16ms/step - loss: 550.6699 - val_loss: 934.3474\n",
      "Epoch 4/200\n",
      "4297/4297 [==============================] - 68s 16ms/step - loss: 900.5533 - val_loss: 786.0828\n",
      "Epoch 5/200\n",
      "4297/4297 [==============================] - 64s 15ms/step - loss: 783.2092 - val_loss: 713.7300\n",
      "Epoch 6/200\n",
      "4297/4297 [==============================] - 68s 16ms/step - loss: 725.0057 - val_loss: 681.7232\n",
      "Epoch 7/200\n",
      "4297/4297 [==============================] - 67s 16ms/step - loss: 698.2780 - val_loss: 669.1175\n",
      "Epoch 8/200\n",
      "4297/4297 [==============================] - 67s 16ms/step - loss: 686.6161 - val_loss: 664.7795\n",
      "Epoch 9/200\n",
      "4297/4297 [==============================] - 66s 15ms/step - loss: 681.6504 - val_loss: 663.6381\n",
      "Epoch 10/200\n",
      "4297/4297 [==============================] - 64s 15ms/step - loss: 679.5518 - val_loss: 663.5997\n",
      "Epoch 11/200\n",
      "4297/4297 [==============================] - 65s 15ms/step - loss: 678.6619 - val_loss: 663.8629\n",
      "Epoch 12/200\n",
      "4297/4297 [==============================] - 65s 15ms/step - loss: 678.2790 - val_loss: 664.1497\n",
      "Epoch 13/200\n",
      "4295/4297 [============================>.] - ETA: 0s - loss: 678.0276"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-61-a73e798b8140>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[0mmse_ae\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mModel\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0minputs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mdecoded\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      2\u001b[0m \u001b[0mmse_ae\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcompile\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mloss\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m'mse'\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0moptimizer\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m'adam'\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;31m#decay=0.9\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 3\u001b[0;31m \u001b[0mmse_hist\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmse_ae\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mfit_generator\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtrain_generator\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx_train\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mepochs\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m200\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0msteps_per_epoch\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0msteps_per_epoch\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mverbose\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mvalidation_steps\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx_val\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mvalidation_data\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mval_generator\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx_val\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      4\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      5\u001b[0m \u001b[0;31m#mse_hist = mse_ae.fit_generator(train_generator(x_train), epochs=200, steps_per_epoch=steps_per_epoch, verbose=1, validation_data=val_generator(x_val))\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/.local/lib/python3.7/site-packages/keras/legacy/interfaces.py\u001b[0m in \u001b[0;36mwrapper\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m     89\u001b[0m                 warnings.warn('Update your `' + object_name + '` call to the ' +\n\u001b[1;32m     90\u001b[0m                               'Keras 2 API: ' + signature, stacklevel=2)\n\u001b[0;32m---> 91\u001b[0;31m             \u001b[0;32mreturn\u001b[0m \u001b[0mfunc\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0margs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mkwargs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     92\u001b[0m         \u001b[0mwrapper\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_original_function\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mfunc\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     93\u001b[0m         \u001b[0;32mreturn\u001b[0m \u001b[0mwrapper\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/.local/lib/python3.7/site-packages/keras/engine/training.py\u001b[0m in \u001b[0;36mfit_generator\u001b[0;34m(self, generator, steps_per_epoch, epochs, verbose, callbacks, validation_data, validation_steps, validation_freq, class_weight, max_queue_size, workers, use_multiprocessing, shuffle, initial_epoch)\u001b[0m\n\u001b[1;32m   1656\u001b[0m             \u001b[0muse_multiprocessing\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0muse_multiprocessing\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1657\u001b[0m             \u001b[0mshuffle\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mshuffle\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1658\u001b[0;31m             initial_epoch=initial_epoch)\n\u001b[0m\u001b[1;32m   1659\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1660\u001b[0m     \u001b[0;34m@\u001b[0m\u001b[0minterfaces\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mlegacy_generator_methods_support\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/.local/lib/python3.7/site-packages/keras/engine/training_generator.py\u001b[0m in \u001b[0;36mfit_generator\u001b[0;34m(model, generator, steps_per_epoch, epochs, verbose, callbacks, validation_data, validation_steps, validation_freq, class_weight, max_queue_size, workers, use_multiprocessing, shuffle, initial_epoch)\u001b[0m\n\u001b[1;32m    235\u001b[0m                             \u001b[0mvalidation_steps\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    236\u001b[0m                             \u001b[0mcallbacks\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcallbacks\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 237\u001b[0;31m                             workers=0)\n\u001b[0m\u001b[1;32m    238\u001b[0m                     \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    239\u001b[0m                         \u001b[0;31m# No need for try/except because\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/.local/lib/python3.7/site-packages/keras/legacy/interfaces.py\u001b[0m in \u001b[0;36mwrapper\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m     89\u001b[0m                 warnings.warn('Update your `' + object_name + '` call to the ' +\n\u001b[1;32m     90\u001b[0m                               'Keras 2 API: ' + signature, stacklevel=2)\n\u001b[0;32m---> 91\u001b[0;31m             \u001b[0;32mreturn\u001b[0m \u001b[0mfunc\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0margs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mkwargs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     92\u001b[0m         \u001b[0mwrapper\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_original_function\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mfunc\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     93\u001b[0m         \u001b[0;32mreturn\u001b[0m \u001b[0mwrapper\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/.local/lib/python3.7/site-packages/keras/engine/training.py\u001b[0m in \u001b[0;36mevaluate_generator\u001b[0;34m(self, generator, steps, callbacks, max_queue_size, workers, use_multiprocessing, verbose)\u001b[0m\n\u001b[1;32m   1715\u001b[0m             \u001b[0mworkers\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mworkers\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1716\u001b[0m             \u001b[0muse_multiprocessing\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0muse_multiprocessing\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1717\u001b[0;31m             verbose=verbose)\n\u001b[0m\u001b[1;32m   1718\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1719\u001b[0m     \u001b[0;34m@\u001b[0m\u001b[0minterfaces\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mlegacy_generator_methods_support\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/.local/lib/python3.7/site-packages/keras/engine/training_generator.py\u001b[0m in \u001b[0;36mevaluate_generator\u001b[0;34m(model, generator, steps, callbacks, max_queue_size, workers, use_multiprocessing, verbose)\u001b[0m\n\u001b[1;32m    383\u001b[0m             \u001b[0mbatch_logs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m{\u001b[0m\u001b[0;34m'batch'\u001b[0m\u001b[0;34m:\u001b[0m \u001b[0msteps_done\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m'size'\u001b[0m\u001b[0;34m:\u001b[0m \u001b[0mbatch_size\u001b[0m\u001b[0;34m}\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    384\u001b[0m             \u001b[0mcallbacks\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_call_batch_hook\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'test'\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m'begin'\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0msteps_done\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mbatch_logs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 385\u001b[0;31m             \u001b[0mouts\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtest_on_batch\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0msample_weight\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0msample_weight\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    386\u001b[0m             \u001b[0mouts\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mto_list\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mouts\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    387\u001b[0m             \u001b[0mouts_per_batch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mappend\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mouts\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/.local/lib/python3.7/site-packages/keras/engine/training.py\u001b[0m in \u001b[0;36mtest_on_batch\u001b[0;34m(self, x, y, sample_weight)\u001b[0m\n\u001b[1;32m   1486\u001b[0m             \u001b[0mins\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mx\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0my\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0msample_weights\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1487\u001b[0m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_make_test_function\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1488\u001b[0;31m         \u001b[0moutputs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtest_function\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mins\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   1489\u001b[0m         \u001b[0;32mreturn\u001b[0m \u001b[0munpack_singleton\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0moutputs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1490\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/.local/lib/python3.7/site-packages/keras/backend/tensorflow_backend.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, inputs)\u001b[0m\n\u001b[1;32m   2977\u001b[0m                     \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_legacy_call\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0minputs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   2978\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 2979\u001b[0;31m             \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_call\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0minputs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   2980\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   2981\u001b[0m             \u001b[0;32mif\u001b[0m \u001b[0mpy_any\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mis_tensor\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0mx\u001b[0m \u001b[0;32min\u001b[0m \u001b[0minputs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/.local/lib/python3.7/site-packages/keras/backend/tensorflow_backend.py\u001b[0m in \u001b[0;36m_call\u001b[0;34m(self, inputs)\u001b[0m\n\u001b[1;32m   2935\u001b[0m             \u001b[0mfetched\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_callable_fn\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0marray_vals\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mrun_metadata\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mrun_metadata\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   2936\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 2937\u001b[0;31m             \u001b[0mfetched\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_callable_fn\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0marray_vals\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   2938\u001b[0m         \u001b[0;32mreturn\u001b[0m \u001b[0mfetched\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0moutputs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   2939\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/anaconda3/envs/graph/lib/python3.7/site-packages/tensorflow/python/client/session.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m   1456\u001b[0m         ret = tf_session.TF_SessionRunCallable(self._session._session,\n\u001b[1;32m   1457\u001b[0m                                                \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_handle\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0margs\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1458\u001b[0;31m                                                run_metadata_ptr)\n\u001b[0m\u001b[1;32m   1459\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mrun_metadata\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1460\u001b[0m           \u001b[0mproto_data\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtf_session\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mTF_GetBuffer\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mrun_metadata_ptr\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "mse_ae = Model(inputs, decoded)\n",
    "mse_ae.compile(loss='mse', optimizer='adam')#decay=0.9\n",
    "mse_hist = mse_ae.fit_generator(train_generator(x_train), epochs=200, steps_per_epoch=steps_per_epoch, verbose=1, validation_steps=len(x_val), validation_data=val_generator(x_val))\n",
    "\n",
    "#mse_hist = mse_ae.fit_generator(train_generator(x_train), epochs=200, steps_per_epoch=steps_per_epoch, verbose=1, validation_data=val_generator(x_val))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def model_save(model, filename):\n",
    "    mode_json = model.to_json()\n",
    "    with open('./recent/' + filename + '.json', 'w') as file:\n",
    "        file.write(model_json)\n",
    "    model.save_weights('./recent/weights_'+filename+'.h5')\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "W0524 22:45:52.586362 140523506476800 deprecation.py:323] From /home/minji/anaconda3/envs/graph/lib/python3.7/site-packages/tensorflow/python/ops/math_grad.py:1250: add_dispatch_support.<locals>.wrapper (from tensorflow.python.ops.array_ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Use tf.where in 2.0, which has the same broadcast rule as np.where\n",
      "W0524 22:45:53.931516 140523506476800 deprecation_wrapper.py:119] From /home/minji/.local/lib/python3.7/site-packages/keras/backend/tensorflow_backend.py:1033: The name tf.assign_add is deprecated. Please use tf.compat.v1.assign_add instead.\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/200\n",
      "4299/4299 [==============================] - 51s 12ms/step - loss: 278.2203\n",
      "Epoch 2/200\n",
      "2530/4299 [================>.............] - ETA: 20s - loss: 732.4828"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-28-2df49beec82b>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[0mkld_ae\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mModel\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0minputs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mdecoded\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      2\u001b[0m \u001b[0mkld_ae\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcompile\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mloss\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcustom_loss\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0moptimizer\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m'adam'\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;31m#decay=0.9\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 3\u001b[0;31m \u001b[0mkld_hist\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mkld_ae\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mfit_generator\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtrain_generator\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx_train\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mepochs\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m200\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0msteps_per_epoch\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0msteps_per_epoch\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mverbose\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m~/.local/lib/python3.7/site-packages/keras/legacy/interfaces.py\u001b[0m in \u001b[0;36mwrapper\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m     89\u001b[0m                 warnings.warn('Update your `' + object_name + '` call to the ' +\n\u001b[1;32m     90\u001b[0m                               'Keras 2 API: ' + signature, stacklevel=2)\n\u001b[0;32m---> 91\u001b[0;31m             \u001b[0;32mreturn\u001b[0m \u001b[0mfunc\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0margs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mkwargs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     92\u001b[0m         \u001b[0mwrapper\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_original_function\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mfunc\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     93\u001b[0m         \u001b[0;32mreturn\u001b[0m \u001b[0mwrapper\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/.local/lib/python3.7/site-packages/keras/engine/training.py\u001b[0m in \u001b[0;36mfit_generator\u001b[0;34m(self, generator, steps_per_epoch, epochs, verbose, callbacks, validation_data, validation_steps, validation_freq, class_weight, max_queue_size, workers, use_multiprocessing, shuffle, initial_epoch)\u001b[0m\n\u001b[1;32m   1656\u001b[0m             \u001b[0muse_multiprocessing\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0muse_multiprocessing\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1657\u001b[0m             \u001b[0mshuffle\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mshuffle\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1658\u001b[0;31m             initial_epoch=initial_epoch)\n\u001b[0m\u001b[1;32m   1659\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1660\u001b[0m     \u001b[0;34m@\u001b[0m\u001b[0minterfaces\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mlegacy_generator_methods_support\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/.local/lib/python3.7/site-packages/keras/engine/training_generator.py\u001b[0m in \u001b[0;36mfit_generator\u001b[0;34m(model, generator, steps_per_epoch, epochs, verbose, callbacks, validation_data, validation_steps, validation_freq, class_weight, max_queue_size, workers, use_multiprocessing, shuffle, initial_epoch)\u001b[0m\n\u001b[1;32m    213\u001b[0m                 outs = model.train_on_batch(x, y,\n\u001b[1;32m    214\u001b[0m                                             \u001b[0msample_weight\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0msample_weight\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 215\u001b[0;31m                                             class_weight=class_weight)\n\u001b[0m\u001b[1;32m    216\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    217\u001b[0m                 \u001b[0mouts\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mto_list\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mouts\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/.local/lib/python3.7/site-packages/keras/engine/training.py\u001b[0m in \u001b[0;36mtrain_on_batch\u001b[0;34m(self, x, y, sample_weight, class_weight)\u001b[0m\n\u001b[1;32m   1447\u001b[0m             \u001b[0mins\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mx\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0my\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0msample_weights\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1448\u001b[0m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_make_train_function\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1449\u001b[0;31m         \u001b[0moutputs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtrain_function\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mins\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   1450\u001b[0m         \u001b[0;32mreturn\u001b[0m \u001b[0munpack_singleton\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0moutputs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1451\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/.local/lib/python3.7/site-packages/keras/backend/tensorflow_backend.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, inputs)\u001b[0m\n\u001b[1;32m   2977\u001b[0m                     \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_legacy_call\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0minputs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   2978\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 2979\u001b[0;31m             \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_call\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0minputs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   2980\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   2981\u001b[0m             \u001b[0;32mif\u001b[0m \u001b[0mpy_any\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mis_tensor\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0mx\u001b[0m \u001b[0;32min\u001b[0m \u001b[0minputs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/.local/lib/python3.7/site-packages/keras/backend/tensorflow_backend.py\u001b[0m in \u001b[0;36m_call\u001b[0;34m(self, inputs)\u001b[0m\n\u001b[1;32m   2935\u001b[0m             \u001b[0mfetched\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_callable_fn\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0marray_vals\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mrun_metadata\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mrun_metadata\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   2936\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 2937\u001b[0;31m             \u001b[0mfetched\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_callable_fn\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0marray_vals\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   2938\u001b[0m         \u001b[0;32mreturn\u001b[0m \u001b[0mfetched\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0moutputs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   2939\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/anaconda3/envs/graph/lib/python3.7/site-packages/tensorflow/python/client/session.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m   1456\u001b[0m         ret = tf_session.TF_SessionRunCallable(self._session._session,\n\u001b[1;32m   1457\u001b[0m                                                \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_handle\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0margs\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1458\u001b[0;31m                                                run_metadata_ptr)\n\u001b[0m\u001b[1;32m   1459\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mrun_metadata\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1460\u001b[0m           \u001b[0mproto_data\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtf_session\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mTF_GetBuffer\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mrun_metadata_ptr\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "kld_ae = Model(inputs, decoded)\n",
    "kld_ae.compile(loss=custom_loss, optimizer='adam')#decay=0.9\n",
    "#kld_hist = kld_ae.fit_generator(train_generator(x_train), epochs=200, steps_per_epoch=steps_per_epoch, verbose=1, validation_steps=len(x_val), validation_data=val_generator(x_val))\n",
    "kld_hist = kld_ae.fit_generator(train_generator(x_train), epochs=200, steps_per_epoch=steps_per_epoch, verbose=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "# MODEL SAVE\n",
    "model_json = cce_ae.to_json()\n",
    "filename = 'cce_no_weight_model' #input('filename: ') \n",
    "with open('models/original/cce_no_weight_model.json', 'w') as file:\n",
    "    file.write(model_json)\n",
    "cce_ae.save_weights('models/weights_' +  filename + '.h5')\n",
    "\n",
    "\n",
    "with open('models/original/cce_no_weight_model_history.json', 'w') as f:\n",
    "    json.dump(cce_hist.history, f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "cce_inputs = Input(shape=(None, 53))\n",
    "cce_encoded = LSTM(128, return_sequences=True)(cce_inputs)  #activation 안적으면 tanh\n",
    "cce_encoded = LSTM(64)(cce_encoded)\n",
    "\n",
    "cce_decoded = Lambda(repeat_vector, output_shape=(None, 64)) ([cce_encoded, cce_inputs]) # inputs의 shape[1] 만큼 encoded 를 반복 생성\n",
    "\n",
    "cce_decoded = LSTM(64, return_sequences=True)(cce_decoded)\n",
    "cce_decoded = LSTM(128, return_sequences=True)(cce_decoded)\n",
    "cce_decoded = TimeDistributed(Dense(53))(cce_decoded)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/200\n",
      "4297/4297 [==============================] - 56s 13ms/step - loss: 2.1219 - val_loss: 2.2727\n",
      "Epoch 2/200\n",
      "4297/4297 [==============================] - 55s 13ms/step - loss: 2.1063 - val_loss: 2.2738\n",
      "Epoch 3/200\n",
      "4297/4297 [==============================] - 55s 13ms/step - loss: 2.1037 - val_loss: 2.2656\n",
      "Epoch 4/200\n",
      "4297/4297 [==============================] - 54s 13ms/step - loss: 2.1022 - val_loss: 2.2711\n",
      "Epoch 5/200\n",
      "4297/4297 [==============================] - 56s 13ms/step - loss: 2.1013 - val_loss: 2.2705\n",
      "Epoch 6/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.1006 - val_loss: 2.2698\n",
      "Epoch 7/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.1001 - val_loss: 2.2689\n",
      "Epoch 8/200\n",
      "4297/4297 [==============================] - 60s 14ms/step - loss: 2.0996 - val_loss: 2.2685\n",
      "Epoch 9/200\n",
      "4297/4297 [==============================] - 62s 14ms/step - loss: 2.0992 - val_loss: 2.2683\n",
      "Epoch 10/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0989 - val_loss: 2.2683\n",
      "Epoch 11/200\n",
      "4297/4297 [==============================] - 58s 14ms/step - loss: 2.0987 - val_loss: 2.2681\n",
      "Epoch 12/200\n",
      "4297/4297 [==============================] - 58s 13ms/step - loss: 2.0985 - val_loss: 2.2680\n",
      "Epoch 13/200\n",
      "4297/4297 [==============================] - 49s 11ms/step - loss: 2.0981 - val_loss: 2.2679\n",
      "Epoch 14/200\n",
      "4297/4297 [==============================] - 54s 12ms/step - loss: 2.0979 - val_loss: 2.2677\n",
      "Epoch 15/200\n",
      "4297/4297 [==============================] - 56s 13ms/step - loss: 2.0978 - val_loss: 2.2672\n",
      "Epoch 16/200\n",
      "4297/4297 [==============================] - 54s 13ms/step - loss: 2.0977 - val_loss: 2.2668\n",
      "Epoch 17/200\n",
      "4297/4297 [==============================] - 49s 11ms/step - loss: 2.0976 - val_loss: 2.2663\n",
      "Epoch 18/200\n",
      "4297/4297 [==============================] - 48s 11ms/step - loss: 2.0975 - val_loss: 2.2659\n",
      "Epoch 19/200\n",
      "4297/4297 [==============================] - 48s 11ms/step - loss: 2.0974 - val_loss: 2.2656\n",
      "Epoch 20/200\n",
      "4297/4297 [==============================] - 48s 11ms/step - loss: 2.0973 - val_loss: 2.2653\n",
      "Epoch 21/200\n",
      "4297/4297 [==============================] - 50s 12ms/step - loss: 2.0972 - val_loss: 2.2650\n",
      "Epoch 22/200\n",
      "4297/4297 [==============================] - 50s 12ms/step - loss: 2.0972 - val_loss: 2.2648\n",
      "Epoch 23/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0971 - val_loss: 2.2646\n",
      "Epoch 24/200\n",
      "4297/4297 [==============================] - 54s 13ms/step - loss: 2.0971 - val_loss: 2.2644\n",
      "Epoch 25/200\n",
      "4297/4297 [==============================] - 50s 12ms/step - loss: 2.0970 - val_loss: 2.2643\n",
      "Epoch 26/200\n",
      "4297/4297 [==============================] - 49s 11ms/step - loss: 2.0970 - val_loss: 2.2642\n",
      "Epoch 27/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0969 - val_loss: 2.2641\n",
      "Epoch 28/200\n",
      "4297/4297 [==============================] - 50s 12ms/step - loss: 2.0969 - val_loss: 2.2641\n",
      "Epoch 29/200\n",
      "4297/4297 [==============================] - 50s 12ms/step - loss: 2.0968 - val_loss: 2.2640\n",
      "Epoch 30/200\n",
      "4297/4297 [==============================] - 50s 12ms/step - loss: 2.0968 - val_loss: 2.2639\n",
      "Epoch 31/200\n",
      "4297/4297 [==============================] - 50s 12ms/step - loss: 2.0967 - val_loss: 2.2639\n",
      "Epoch 32/200\n",
      "4297/4297 [==============================] - 56s 13ms/step - loss: 2.0967 - val_loss: 2.2638\n",
      "Epoch 33/200\n",
      "4297/4297 [==============================] - 53s 12ms/step - loss: 2.0967 - val_loss: 2.2637\n",
      "Epoch 34/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0966 - val_loss: 2.2637\n",
      "Epoch 35/200\n",
      "4297/4297 [==============================] - 53s 12ms/step - loss: 2.0966 - val_loss: 2.2637\n",
      "Epoch 36/200\n",
      "4297/4297 [==============================] - 48s 11ms/step - loss: 2.0966 - val_loss: 2.2636\n",
      "Epoch 37/200\n",
      "4297/4297 [==============================] - 54s 13ms/step - loss: 2.0965 - val_loss: 2.2636\n",
      "Epoch 38/200\n",
      "4297/4297 [==============================] - 53s 12ms/step - loss: 2.0965 - val_loss: 2.2635\n",
      "Epoch 39/200\n",
      "4297/4297 [==============================] - 49s 11ms/step - loss: 2.0965 - val_loss: 2.2635\n",
      "Epoch 40/200\n",
      "4297/4297 [==============================] - 49s 11ms/step - loss: 2.0964 - val_loss: 2.2634\n",
      "Epoch 41/200\n",
      "4297/4297 [==============================] - 49s 11ms/step - loss: 2.0964 - val_loss: 2.2634\n",
      "Epoch 42/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0964 - val_loss: 2.2633\n",
      "Epoch 43/200\n",
      "4297/4297 [==============================] - 55s 13ms/step - loss: 2.0964 - val_loss: 2.2633\n",
      "Epoch 44/200\n",
      "4297/4297 [==============================] - 54s 13ms/step - loss: 2.0963 - val_loss: 2.2632\n",
      "Epoch 45/200\n",
      "4297/4297 [==============================] - 54s 13ms/step - loss: 2.0963 - val_loss: 2.2632\n",
      "Epoch 46/200\n",
      "4297/4297 [==============================] - 55s 13ms/step - loss: 2.0963 - val_loss: 2.2632\n",
      "Epoch 47/200\n",
      "4297/4297 [==============================] - 55s 13ms/step - loss: 2.0963 - val_loss: 2.2631\n",
      "Epoch 48/200\n",
      "4297/4297 [==============================] - 55s 13ms/step - loss: 2.0962 - val_loss: 2.2631\n",
      "Epoch 49/200\n",
      "4297/4297 [==============================] - 54s 13ms/step - loss: 2.0962 - val_loss: 2.2631\n",
      "Epoch 50/200\n",
      "4297/4297 [==============================] - 54s 13ms/step - loss: 2.0962 - val_loss: 2.2631\n",
      "Epoch 51/200\n",
      "4297/4297 [==============================] - 55s 13ms/step - loss: 2.0962 - val_loss: 2.2630\n",
      "Epoch 52/200\n",
      "4297/4297 [==============================] - 55s 13ms/step - loss: 2.0962 - val_loss: 2.2630\n",
      "Epoch 53/200\n",
      "4297/4297 [==============================] - 55s 13ms/step - loss: 2.0961 - val_loss: 2.2630\n",
      "Epoch 54/200\n",
      "4297/4297 [==============================] - 66s 15ms/step - loss: 2.0961 - val_loss: 2.2629\n",
      "Epoch 55/200\n",
      "4297/4297 [==============================] - 60s 14ms/step - loss: 2.0961 - val_loss: 2.2629\n",
      "Epoch 56/200\n",
      "4297/4297 [==============================] - 60s 14ms/step - loss: 2.0961 - val_loss: 2.2629\n",
      "Epoch 57/200\n",
      "4297/4297 [==============================] - 60s 14ms/step - loss: 2.0961 - val_loss: 2.2629\n",
      "Epoch 58/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0961 - val_loss: 2.2629\n",
      "Epoch 59/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0960 - val_loss: 2.2628\n",
      "Epoch 60/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0960 - val_loss: 2.2628\n",
      "Epoch 61/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0960 - val_loss: 2.2628\n",
      "Epoch 62/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0960 - val_loss: 2.2628\n",
      "Epoch 63/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0960 - val_loss: 2.2628\n",
      "Epoch 64/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0960 - val_loss: 2.2627\n",
      "Epoch 65/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0960 - val_loss: 2.2627\n",
      "Epoch 66/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0959 - val_loss: 2.2627\n",
      "Epoch 67/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0959 - val_loss: 2.2627\n",
      "Epoch 68/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0959 - val_loss: 2.2627\n",
      "Epoch 69/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0959 - val_loss: 2.2627\n",
      "Epoch 70/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0959 - val_loss: 2.2626\n",
      "Epoch 71/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0959 - val_loss: 2.2626\n",
      "Epoch 72/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0959 - val_loss: 2.2626\n",
      "Epoch 73/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0959 - val_loss: 2.2626\n",
      "Epoch 74/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0958 - val_loss: 2.2626\n",
      "Epoch 75/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0958 - val_loss: 2.2626\n",
      "Epoch 76/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0958 - val_loss: 2.2625\n",
      "Epoch 77/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0958 - val_loss: 2.2625\n",
      "Epoch 78/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0958 - val_loss: 2.2625\n",
      "Epoch 79/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0958 - val_loss: 2.2625\n",
      "Epoch 80/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0958 - val_loss: 2.2625\n",
      "Epoch 81/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0958 - val_loss: 2.2625\n",
      "Epoch 82/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0958 - val_loss: 2.2625\n",
      "Epoch 83/200\n",
      "4297/4297 [==============================] - 60s 14ms/step - loss: 2.0957 - val_loss: 2.2625\n",
      "Epoch 84/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0957 - val_loss: 2.2625\n",
      "Epoch 85/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0957 - val_loss: 2.2624\n",
      "Epoch 86/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0957 - val_loss: 2.2624\n",
      "Epoch 87/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0957 - val_loss: 2.2624\n",
      "Epoch 88/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0957 - val_loss: 2.2624\n",
      "Epoch 89/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0957 - val_loss: 2.2624\n",
      "Epoch 90/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0957 - val_loss: 2.2624\n",
      "Epoch 91/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0957 - val_loss: 2.2624\n",
      "Epoch 92/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0957 - val_loss: 2.2624\n",
      "Epoch 93/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0957 - val_loss: 2.2624\n",
      "Epoch 94/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0956 - val_loss: 2.2623\n",
      "Epoch 95/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0956 - val_loss: 2.2623\n",
      "Epoch 96/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0956 - val_loss: 2.2623\n",
      "Epoch 97/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0956 - val_loss: 2.2623\n",
      "Epoch 98/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0956 - val_loss: 2.2623\n",
      "Epoch 99/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0956 - val_loss: 2.2623\n",
      "Epoch 100/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0956 - val_loss: 2.2623\n",
      "Epoch 101/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0956 - val_loss: 2.2623\n",
      "Epoch 102/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0956 - val_loss: 2.2623\n",
      "Epoch 103/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0956 - val_loss: 2.2623\n",
      "Epoch 104/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0956 - val_loss: 2.2622\n",
      "Epoch 105/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0956 - val_loss: 2.2622\n",
      "Epoch 106/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0955 - val_loss: 2.2622\n",
      "Epoch 107/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0955 - val_loss: 2.2622\n",
      "Epoch 108/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0955 - val_loss: 2.2622\n",
      "Epoch 109/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0955 - val_loss: 2.2622\n",
      "Epoch 110/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0955 - val_loss: 2.2622\n",
      "Epoch 111/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0955 - val_loss: 2.2622\n",
      "Epoch 112/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0955 - val_loss: 2.2622\n",
      "Epoch 113/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0955 - val_loss: 2.2622\n",
      "Epoch 114/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0955 - val_loss: 2.2622\n",
      "Epoch 115/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0955 - val_loss: 2.2622\n",
      "Epoch 116/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0955 - val_loss: 2.2622\n",
      "Epoch 117/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0955 - val_loss: 2.2622\n",
      "Epoch 118/200\n",
      "4297/4297 [==============================] - 59s 14ms/step - loss: 2.0955 - val_loss: 2.2622\n",
      "Epoch 119/200\n",
      "4297/4297 [==============================] - 54s 13ms/step - loss: 2.0955 - val_loss: 2.2621\n",
      "Epoch 120/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0955 - val_loss: 2.2621\n",
      "Epoch 121/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0955 - val_loss: 2.2621\n",
      "Epoch 122/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0954 - val_loss: 2.2621\n",
      "Epoch 123/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0954 - val_loss: 2.2621\n",
      "Epoch 124/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0954 - val_loss: 2.2621\n",
      "Epoch 125/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0954 - val_loss: 2.2621\n",
      "Epoch 126/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0954 - val_loss: 2.2621\n",
      "Epoch 127/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0954 - val_loss: 2.2621\n",
      "Epoch 128/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0954 - val_loss: 2.2621\n",
      "Epoch 129/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0954 - val_loss: 2.2621\n",
      "Epoch 130/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0954 - val_loss: 2.2621\n",
      "Epoch 131/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0954 - val_loss: 2.2621\n",
      "Epoch 132/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0954 - val_loss: 2.2621\n",
      "Epoch 133/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0954 - val_loss: 2.2621\n",
      "Epoch 134/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0954 - val_loss: 2.2621\n",
      "Epoch 135/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0954 - val_loss: 2.2621\n",
      "Epoch 136/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0954 - val_loss: 2.2621\n",
      "Epoch 137/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0954 - val_loss: 2.2621\n",
      "Epoch 138/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0954 - val_loss: 2.2621\n",
      "Epoch 139/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0954 - val_loss: 2.2620\n",
      "Epoch 140/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0953 - val_loss: 2.2620\n",
      "Epoch 141/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0953 - val_loss: 2.2620\n",
      "Epoch 142/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0953 - val_loss: 2.2620\n",
      "Epoch 143/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0953 - val_loss: 2.2620\n",
      "Epoch 144/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0953 - val_loss: 2.2620\n",
      "Epoch 145/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0953 - val_loss: 2.2620\n",
      "Epoch 146/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0953 - val_loss: 2.2620\n",
      "Epoch 147/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0953 - val_loss: 2.2620\n",
      "Epoch 148/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0953 - val_loss: 2.2620\n",
      "Epoch 149/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0953 - val_loss: 2.2620\n",
      "Epoch 150/200\n",
      "4297/4297 [==============================] - 53s 12ms/step - loss: 2.0953 - val_loss: 2.2620\n",
      "Epoch 151/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0953 - val_loss: 2.2620\n",
      "Epoch 152/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0953 - val_loss: 2.2620\n",
      "Epoch 153/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0953 - val_loss: 2.2620\n",
      "Epoch 154/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0953 - val_loss: 2.2620\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 155/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0953 - val_loss: 2.2620\n",
      "Epoch 156/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0953 - val_loss: 2.2620\n",
      "Epoch 157/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0953 - val_loss: 2.2620\n",
      "Epoch 158/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0953 - val_loss: 2.2620\n",
      "Epoch 159/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0953 - val_loss: 2.2620\n",
      "Epoch 160/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0953 - val_loss: 2.2620\n",
      "Epoch 161/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0953 - val_loss: 2.2620\n",
      "Epoch 162/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0953 - val_loss: 2.2620\n",
      "Epoch 163/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0952 - val_loss: 2.2620\n",
      "Epoch 164/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0952 - val_loss: 2.2620\n",
      "Epoch 165/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0952 - val_loss: 2.2620\n",
      "Epoch 166/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0952 - val_loss: 2.2620\n",
      "Epoch 167/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0952 - val_loss: 2.2620\n",
      "Epoch 168/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0952 - val_loss: 2.2620\n",
      "Epoch 169/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0952 - val_loss: 2.2620\n",
      "Epoch 170/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0952 - val_loss: 2.2620\n",
      "Epoch 171/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0952 - val_loss: 2.2620\n",
      "Epoch 172/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0952 - val_loss: 2.2620\n",
      "Epoch 173/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0952 - val_loss: 2.2620\n",
      "Epoch 174/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0952 - val_loss: 2.2621\n",
      "Epoch 175/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0952 - val_loss: 2.2621\n",
      "Epoch 176/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0952 - val_loss: 2.2621\n",
      "Epoch 177/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0952 - val_loss: 2.2621\n",
      "Epoch 178/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0952 - val_loss: 2.2621\n",
      "Epoch 179/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0952 - val_loss: 2.2621\n",
      "Epoch 180/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0952 - val_loss: 2.2621\n",
      "Epoch 181/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0952 - val_loss: 2.2621\n",
      "Epoch 182/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0952 - val_loss: 2.2621\n",
      "Epoch 183/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0952 - val_loss: 2.2621\n",
      "Epoch 184/200\n",
      "4297/4297 [==============================] - 51s 12ms/step - loss: 2.0952 - val_loss: 2.2621\n",
      "Epoch 185/200\n",
      "4297/4297 [==============================] - 53s 12ms/step - loss: 2.0952 - val_loss: 2.2621\n",
      "Epoch 186/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0952 - val_loss: 2.2621\n",
      "Epoch 187/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0952 - val_loss: 2.2621\n",
      "Epoch 188/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0952 - val_loss: 2.2621\n",
      "Epoch 189/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0952 - val_loss: 2.2621\n",
      "Epoch 190/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0951 - val_loss: 2.2621\n",
      "Epoch 191/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0951 - val_loss: 2.2621\n",
      "Epoch 192/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0951 - val_loss: 2.2621\n",
      "Epoch 193/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0951 - val_loss: 2.2621\n",
      "Epoch 194/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0951 - val_loss: 2.2621\n",
      "Epoch 195/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0951 - val_loss: 2.2621\n",
      "Epoch 196/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0951 - val_loss: 2.2621\n",
      "Epoch 197/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0951 - val_loss: 2.2621\n",
      "Epoch 198/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0951 - val_loss: 2.2621\n",
      "Epoch 199/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0951 - val_loss: 2.2621\n",
      "Epoch 200/200\n",
      "4297/4297 [==============================] - 52s 12ms/step - loss: 2.0951 - val_loss: 2.2621\n"
     ]
    }
   ],
   "source": [
    "cce_ae = Model(inputs, decoded)\n",
    "cce_ae.compile(loss=custom_loss1, optimizer=Adam(lr=0.0002, decay=1e-3))#Adam(lr=0.003))#decay=0.9\n",
    "cce_hist = cce_ae.fit_generator(train_generator(x_train), epochs=200, steps_per_epoch=steps_per_epoch, verbose=1, validation_steps=len(x_val), validation_data=val_generator(x_val))\n",
    "#cce_hist = cce_ae.fit_generator(train_generator(x_train), epochs=200, steps_per_epoch=steps_per_epoch, verbose=1, validation_steps=len(x_val), validation_data=val_generator(x_val))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 81,
   "metadata": {},
   "outputs": [],
   "source": [
    "kld_json = kld_ae.to_json()\n",
    "mse_json = mse_ae.to_json()\n",
    "cce_json = cce_ae.to_json()\n",
    "models = [kld_ae, mse_ae, cce_ae]\n",
    "models_json = [kld_json, mse_json, cce_json]\n",
    "filenames = ['kld_oh_model', 'mse_oh_model', 'cce_oh_model']\n",
    "histories = [kld_hist, mse_hist, cce_hist]\n",
    "\n",
    "for filename, model, model_json, hist in zip(filenames, models, models_json,histories):\n",
    "    with open('models/' + filename + '.json', 'w') as file:\n",
    "        file.write(model_json)\n",
    "    model.save_weights('models/weights_' + filename + '.h5')\n",
    "    with open('models/' + filename + '_history.json', 'w') as file:\n",
    "        json.dump(hist.history, file)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def model_save(filename, model):\n",
    "    model_json = model.to_json()\n",
    "    with open('models/' + filename, 'w') as file:\n",
    "        file.write(model_json)\n",
    "    model.save_weights('models/weights_' + filename + '.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "# MODEL SAVE\n",
    "model_json = mse_ae.to_json()\n",
    "filename = '1mse_oh_model' #input('filename: ') \n",
    "with open('models/1mse_oh_model.json', 'w') as file:\n",
    "    file.write(model_json)\n",
    "mse_ae.save_weights('models/weights_' +  filename + '.h5')\n",
    "\n",
    "\n",
    "with open('models/1mse_oh_model_hist.json', 'w') as f:\n",
    "    json.dump(mse_hist.history, f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "with open('models/1mse_oh_model_hist.json', 'w') as f:\n",
    "    json.dump(mse_hist.history, f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "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": [
    "import matplotlib.pyplot as plt\n",
    "h1 = json.load(open('models/mse_oh_model_history.json', 'r'))\n",
    "\n",
    "fig, loss_ax = plt.subplots()\n",
    "\n",
    "loss_ax.plot(h1['val_loss'], 'b', label='val loss')\n",
    "loss_ax.plot(h1['loss'], 'r', label='train loss')\n",
    "loss_ax.set_xlabel('epoch')\n",
    "loss_ax.set_ylabel('loss')\n",
    "loss_ax.legend(loc='upper left')\n",
    "\n",
    "plt.title('MSE one-hot')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "loaded_model = lstm_autoencoder"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.17943959858127367\n"
     ]
    }
   ],
   "source": [
    "mean= 0\n",
    "for xt in x_test:\n",
    "    xt = xt.reshape(1, xt.shape[0], xt.shape[1])\n",
    "    out = mse_ae.predict(xt)\n",
    "    mean += ((xt-out)**2).mean(axis=None)\n",
    "print(mean/len(x_test))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "filename = \"mse_oh_model\"\n",
    "# MODEL LOAD\n",
    "mse_ae = model_from_json(open('models/' +filename + '.json').read())\n",
    "mse_ae.load_weights('models/weights_' + filename + '.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "encoder = Model(loaded_model.input, loaded_model.layers[3].output)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "test = x_test[0].reshape(1, x_test[0].shape[0], x_test[0].shape[1])\n",
    "latent_vector = []\n",
    "for x in x_test:\n",
    "    x = x.reshape(1, x.shape[0], x.shape[1])\n",
    "    latent_vector.append(encoder.predict(x)[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_path = 'model_save/mse_kld_models/weights' + '{epoch:02d}-{loss:.4f}.h5'\n",
    "early_stopping_callback = EarlyStopping(monitor='loss', patience=200)\n",
    "checkpoint_callback = ModelCheckpoint(model_path, monitor='loss', verbose = 1, save_best_only = True, save_weights_only = True, mode='min', period=100)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<IPython.core.display.Image object>"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "plot_model(loaded_model, to_file='./model.png', show_shapes=True, show_layer_names=True)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}