project_gru_keras.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "from keras.models import Sequential\n",
    "from keras.layers import LSTM, GRU, Dense, Dropout, TimeDistributed, BatchNormalization\n",
    "from keras import regularizers\n",
    "import numpy as np\n",
    "import h5py\n",
    "import matplotlib.pyplot as plt\n",
    "from data_utils import get_data\n",
    "from sklearn import preprocessing\n",
    "from functools import reduce\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X_train: (2458, 1000, 22) \n",
      "y_train: (2458,) \n",
      "X_val: (100, 1000, 22) \n",
      "y_val: (100,) \n",
      "X_test: (0, 1000, 22) \n",
      "y_test: (0,) \n"
     ]
    }
   ],
   "source": [
    "# Load data from all .mat files, combine them, eliminate EOG signals, shuffle and \n",
    "# seperate training data, validation data and testing data.\n",
    "# Also do mean subtraction on x.\n",
    "\n",
    "data = get_data('project_datasets',num_validation=100, num_test=0, subtract_mean=True, subtract_axis=1,\n",
    "                transpose=True)\n",
    "for k in data.keys():\n",
    "    print('{}: {} '.format(k, data[k].shape))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# substract data from list\n",
    "X_train = data.get('X_train')\n",
    "y_train = data.get('y_train')\n",
    "X_val = data.get('X_val')\n",
    "y_val = data.get('y_val')\n",
    "#X_test = data.get('X_test')\n",
    "#y_test = data.get('y_test')\n",
    "\n",
    "# get data dimension\n",
    "N_train, T_train, C_train = data.get('X_train').shape\n",
    "N_val, T_val, C_val = data.get('X_val').shape\n",
    "#N_test, T_test, C_test = data.get('X_test').shape\n",
    "\n",
    "# add dummy zeros for y classification\n",
    "lb = preprocessing.LabelBinarizer()\n",
    "lb.fit([0, 1, 2, 3,])\n",
    "lb.classes_\n",
    "y_train = lb.transform(y_train)\n",
    "y_val = lb.transform(y_val)\n",
    "#y_test = lb.transform(y_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[0 1 0 0]\n",
      " [0 0 1 0]\n",
      " [0 0 0 1]\n",
      " [0 0 1 0]\n",
      " [1 0 0 0]\n",
      " [0 0 0 1]\n",
      " [0 0 1 0]\n",
      " [1 0 0 0]\n",
      " [0 0 1 0]\n",
      " [0 0 1 0]\n",
      " [0 0 0 1]\n",
      " [0 0 0 1]\n",
      " [0 0 0 1]\n",
      " [0 0 0 1]\n",
      " [0 0 1 0]\n",
      " [0 0 0 1]\n",
      " [0 1 0 0]\n",
      " [0 0 0 1]\n",
      " [0 0 0 1]\n",
      " [0 0 0 1]\n",
      " [0 1 0 0]\n",
      " [0 0 1 0]\n",
      " [0 1 0 0]\n",
      " [1 0 0 0]\n",
      " [1 0 0 0]\n",
      " [1 0 0 0]\n",
      " [0 0 0 1]\n",
      " [0 1 0 0]\n",
      " [1 0 0 0]\n",
      " [0 0 0 1]\n",
      " [1 0 0 0]\n",
      " [0 0 1 0]\n",
      " [0 0 0 1]\n",
      " [1 0 0 0]\n",
      " [1 0 0 0]\n",
      " [0 0 0 1]\n",
      " [1 0 0 0]\n",
      " [0 1 0 0]\n",
      " [1 0 0 0]\n",
      " [0 1 0 0]\n",
      " [1 0 0 0]\n",
      " [0 0 1 0]\n",
      " [0 1 0 0]\n",
      " [0 1 0 0]\n",
      " [1 0 0 0]\n",
      " [0 0 0 1]\n",
      " [0 0 0 1]\n",
      " [0 1 0 0]\n",
      " [1 0 0 0]\n",
      " [1 0 0 0]\n",
      " [1 0 0 0]\n",
      " [0 0 1 0]\n",
      " [0 0 0 1]\n",
      " [0 0 0 1]\n",
      " [0 0 1 0]\n",
      " [0 0 0 1]\n",
      " [0 0 1 0]\n",
      " [0 0 1 0]\n",
      " [0 0 0 1]\n",
      " [0 0 0 1]\n",
      " [0 0 0 1]\n",
      " [1 0 0 0]\n",
      " [0 1 0 0]\n",
      " [1 0 0 0]\n",
      " [1 0 0 0]\n",
      " [1 0 0 0]\n",
      " [1 0 0 0]\n",
      " [1 0 0 0]\n",
      " [0 0 0 1]\n",
      " [0 0 0 1]\n",
      " [0 1 0 0]\n",
      " [0 0 1 0]\n",
      " [0 0 0 1]\n",
      " [1 0 0 0]\n",
      " [0 0 1 0]\n",
      " [0 0 0 1]\n",
      " [1 0 0 0]\n",
      " [0 0 1 0]\n",
      " [0 0 0 1]\n",
      " [0 0 0 1]\n",
      " [0 0 0 1]\n",
      " [1 0 0 0]\n",
      " [0 1 0 0]\n",
      " [0 1 0 0]\n",
      " [1 0 0 0]\n",
      " [0 0 1 0]\n",
      " [0 1 0 0]\n",
      " [0 1 0 0]\n",
      " [0 0 1 0]\n",
      " [0 1 0 0]\n",
      " [0 1 0 0]\n",
      " [1 0 0 0]\n",
      " [0 1 0 0]\n",
      " [0 1 0 0]\n",
      " [0 0 1 0]\n",
      " [0 0 1 0]\n",
      " [0 1 0 0]\n",
      " [1 0 0 0]\n",
      " [0 0 0 1]\n",
      " [1 0 0 0]]\n"
     ]
    }
   ],
   "source": [
    "print(y_val)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# # normalize data across the time sequence dimension only\n",
    "# scaler = preprocessing.MinMaxScaler()\n",
    "# for i in range(N_train):\n",
    "#     scaler.fit(X_train[i,:,:])\n",
    "#     X_train[i,:,:] = scaler.transform(X_train[i,:,:])\n",
    "    \n",
    "# for i in range(N_val):\n",
    "#     scaler.fit(X_val[i,:,:])\n",
    "#     X_val[i,:,:] = scaler.transform(X_val[i,:,:])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X_train_sub:  (2458, 500, 22)\n",
      "X_val_sub:  (100, 500, 22)\n"
     ]
    }
   ],
   "source": [
    "def downsample(data, mul=10, enableTrunc = False):\n",
    "    assert data.shape[0] % mul == 0 or  enableTrunc, \"Can't downsample without data truncation.\\\n",
    "                                                \\n Set another value of mul or set enableTrunc = True\"\n",
    "    outdata = []\n",
    "    for i in range(data.shape[0]//mul):\n",
    "        outdata.append(np.mean(data[i*mul:(i+1)*mul,...], axis=0, keepdims=True))\n",
    "    outdata = reduce(lambda a,b: np.append(a,b,0), outdata)\n",
    "    \n",
    "    return outdata\n",
    "span = 2\n",
    "X_train_sub = np.expand_dims(downsample(X_train[0,:,:], mul=span),axis=0)\n",
    "for n in np.arange(1,X_train.shape[0]):\n",
    "    X_train_sub = np.append(X_train_sub,\n",
    "                            np.expand_dims(downsample(X_train[n,:,:], mul=span),axis=0),axis=0)\n",
    "\n",
    "X_val_sub = np.expand_dims(downsample(X_val[0,:,:], mul=span),axis=0)\n",
    "for n in np.arange(1,X_val.shape[0]):\n",
    "    X_val_sub = np.append(X_val_sub,\n",
    "                            np.expand_dims(downsample(X_val[n,:,:], mul=span),axis=0),axis=0)\n",
    "\n",
    "print('X_train_sub: ', X_train_sub.shape)\n",
    "print('X_val_sub: ', X_val_sub.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Expected input batch shape: (batch_size, timesteps, data_dim)\n",
    "# Note that we have to provide the full batch_input_shape since the network is stateful.\n",
    "# the sample of index i in batch k is the follow-up for the sample i in batch k-1.\n",
    "\n",
    "# perhaps should try masking layer\n",
    "\n",
    "data_dim = C_train\n",
    "timesteps = T_train//span\n",
    "num_classes = 4\n",
    "batch_size = 100\n",
    "num_epoch = 100\n",
    "\n",
    "model = Sequential()\n",
    "model.add(GRU(20, use_bias=False, return_sequences=True, stateful=True,\n",
    "          batch_input_shape=(batch_size, timesteps, data_dim),\n",
    "          kernel_regularizer=regularizers.l2(0.01),\n",
    "          recurrent_regularizer=regularizers.l2(0.01)))\n",
    "model.add(GRU(16, use_bias=False, return_sequences=True, stateful=True,\n",
    "          kernel_regularizer=regularizers.l2(0.01),\n",
    "          recurrent_regularizer=regularizers.l2(0.01)))\n",
    "model.add(GRU(12, use_bias=False, return_sequences=True, stateful=True,\n",
    "          kernel_regularizer=regularizers.l2(0.01),\n",
    "          recurrent_regularizer=regularizers.l2(0.01)))\n",
    "model.add(GRU(8, use_bias=False, stateful=True,\n",
    "          kernel_regularizer=regularizers.l2(0.01),\n",
    "          recurrent_regularizer=regularizers.l2(0.01)))\n",
    "model.add(Dense(num_classes, activation='softmax'))\n",
    "\n",
    "model.compile(loss='categorical_crossentropy',\n",
    "              optimizer='adam',\n",
    "              metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 2400 samples, validate on 100 samples\n",
      "Epoch 1/100\n",
      "2400/2400 [==============================] - 32s 14ms/step - loss: 2.7996 - acc: 0.2483 - val_loss: 2.6459 - val_acc: 0.2700\n",
      "Epoch 2/100\n",
      "2400/2400 [==============================] - 30s 13ms/step - loss: 2.5198 - acc: 0.2650 - val_loss: 2.4064 - val_acc: 0.2000\n",
      "Epoch 3/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 2.2987 - acc: 0.2708 - val_loss: 2.2118 - val_acc: 0.2100\n",
      "Epoch 4/100\n",
      "2400/2400 [==============================] - 30s 12ms/step - loss: 2.1214 - acc: 0.2817 - val_loss: 2.0555 - val_acc: 0.2200\n",
      "Epoch 5/100\n",
      "2400/2400 [==============================] - 30s 13ms/step - loss: 1.9791 - acc: 0.2938 - val_loss: 1.9289 - val_acc: 0.2200\n",
      "Epoch 6/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.8646 - acc: 0.2967 - val_loss: 1.8266 - val_acc: 0.2200\n",
      "Epoch 7/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.7725 - acc: 0.3017 - val_loss: 1.7441 - val_acc: 0.2800\n",
      "Epoch 8/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.6983 - acc: 0.3004 - val_loss: 1.6774 - val_acc: 0.2600\n",
      "Epoch 9/100\n",
      "2400/2400 [==============================] - 30s 13ms/step - loss: 1.6387 - acc: 0.2992 - val_loss: 1.6232 - val_acc: 0.2600\n",
      "Epoch 10/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.5906 - acc: 0.2946 - val_loss: 1.5794 - val_acc: 0.2800\n",
      "Epoch 11/100\n",
      "2400/2400 [==============================] - 30s 12ms/step - loss: 1.5519 - acc: 0.3017 - val_loss: 1.5439 - val_acc: 0.2700\n",
      "Epoch 12/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.5208 - acc: 0.3017 - val_loss: 1.5152 - val_acc: 0.2400\n",
      "Epoch 13/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.4958 - acc: 0.3008 - val_loss: 1.4917 - val_acc: 0.2300\n",
      "Epoch 14/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.4755 - acc: 0.2971 - val_loss: 1.4727 - val_acc: 0.2100\n",
      "Epoch 15/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.4592 - acc: 0.2925 - val_loss: 1.4572 - val_acc: 0.2100\n",
      "Epoch 16/100\n",
      "2400/2400 [==============================] - 28s 12ms/step - loss: 1.4460 - acc: 0.2908 - val_loss: 1.4447 - val_acc: 0.1900\n",
      "Epoch 17/100\n",
      "2400/2400 [==============================] - 30s 13ms/step - loss: 1.4353 - acc: 0.2867 - val_loss: 1.4345 - val_acc: 0.2100\n",
      "Epoch 18/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.4265 - acc: 0.2833 - val_loss: 1.4261 - val_acc: 0.2100\n",
      "Epoch 19/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.4194 - acc: 0.2854 - val_loss: 1.4192 - val_acc: 0.1900\n",
      "Epoch 20/100\n",
      "2400/2400 [==============================] - 30s 12ms/step - loss: 1.4136 - acc: 0.2867 - val_loss: 1.4137 - val_acc: 0.1900\n",
      "Epoch 21/100\n",
      "2400/2400 [==============================] - 30s 12ms/step - loss: 1.4088 - acc: 0.2846 - val_loss: 1.4090 - val_acc: 0.1700\n",
      "Epoch 22/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.4048 - acc: 0.2775 - val_loss: 1.4052 - val_acc: 0.1700\n",
      "Epoch 23/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.4015 - acc: 0.2792 - val_loss: 1.4019 - val_acc: 0.1800\n",
      "Epoch 24/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.3988 - acc: 0.2796 - val_loss: 1.3993 - val_acc: 0.2000\n",
      "Epoch 25/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.3966 - acc: 0.2675 - val_loss: 1.3970 - val_acc: 0.2100\n",
      "Epoch 26/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.3947 - acc: 0.2633 - val_loss: 1.3952 - val_acc: 0.2000\n",
      "Epoch 27/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.3932 - acc: 0.2583 - val_loss: 1.3936 - val_acc: 0.1900\n",
      "Epoch 28/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.3919 - acc: 0.2583 - val_loss: 1.3923 - val_acc: 0.1900\n",
      "Epoch 29/100\n",
      "2400/2400 [==============================] - 30s 13ms/step - loss: 1.3909 - acc: 0.2525 - val_loss: 1.3913 - val_acc: 0.2000\n",
      "Epoch 30/100\n",
      "2400/2400 [==============================] - 30s 13ms/step - loss: 1.3900 - acc: 0.2529 - val_loss: 1.3904 - val_acc: 0.2000\n",
      "Epoch 31/100\n",
      "2400/2400 [==============================] - 30s 12ms/step - loss: 1.3893 - acc: 0.2525 - val_loss: 1.3897 - val_acc: 0.2000\n",
      "Epoch 32/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.3887 - acc: 0.2525 - val_loss: 1.3891 - val_acc: 0.2000\n",
      "Epoch 33/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.3882 - acc: 0.2525 - val_loss: 1.3886 - val_acc: 0.2000\n",
      "Epoch 34/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.3878 - acc: 0.2525 - val_loss: 1.3882 - val_acc: 0.2000\n",
      "Epoch 35/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.3875 - acc: 0.2525 - val_loss: 1.3879 - val_acc: 0.2000\n",
      "Epoch 36/100\n",
      "2400/2400 [==============================] - 30s 12ms/step - loss: 1.3873 - acc: 0.2525 - val_loss: 1.3877 - val_acc: 0.2000\n",
      "Epoch 37/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.3871 - acc: 0.2525 - val_loss: 1.3875 - val_acc: 0.2000\n",
      "Epoch 38/100\n",
      "2400/2400 [==============================] - 32s 13ms/step - loss: 1.3869 - acc: 0.2525 - val_loss: 1.3874 - val_acc: 0.2000\n",
      "Epoch 39/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.3868 - acc: 0.2525 - val_loss: 1.3872 - val_acc: 0.2000\n",
      "Epoch 40/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.3867 - acc: 0.2525 - val_loss: 1.3872 - val_acc: 0.2000\n",
      "Epoch 41/100\n",
      "2400/2400 [==============================] - 30s 13ms/step - loss: 1.3866 - acc: 0.2525 - val_loss: 1.3871 - val_acc: 0.2000\n",
      "Epoch 42/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.3865 - acc: 0.2525 - val_loss: 1.3870 - val_acc: 0.2000\n",
      "Epoch 43/100\n",
      "2400/2400 [==============================] - 30s 12ms/step - loss: 1.3865 - acc: 0.2525 - val_loss: 1.3870 - val_acc: 0.2000\n",
      "Epoch 44/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.3864 - acc: 0.2525 - val_loss: 1.3869 - val_acc: 0.2000\n",
      "Epoch 45/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.3864 - acc: 0.2525 - val_loss: 1.3869 - val_acc: 0.2000\n",
      "Epoch 46/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.3864 - acc: 0.2525 - val_loss: 1.3869 - val_acc: 0.2000\n",
      "Epoch 47/100\n",
      "2400/2400 [==============================] - 30s 13ms/step - loss: 1.3864 - acc: 0.2525 - val_loss: 1.3869 - val_acc: 0.2000\n",
      "Epoch 48/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 49/100\n",
      "2400/2400 [==============================] - 30s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 50/100\n",
      "2400/2400 [==============================] - 32s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 51/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 52/100\n",
      "2400/2400 [==============================] - 30s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 53/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 54/100\n",
      "2400/2400 [==============================] - 32s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 55/100\n",
      "2400/2400 [==============================] - 30s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 56/100\n",
      "2400/2400 [==============================] - 30s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 57/100\n",
      "2400/2400 [==============================] - 30s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 58/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 59/100\n",
      "2400/2400 [==============================] - 30s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 60/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 61/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 62/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 63/100\n",
      "2400/2400 [==============================] - 30s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 64/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 65/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 66/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 67/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 68/100\n",
      "2400/2400 [==============================] - 30s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 69/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 70/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 71/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 72/100\n",
      "2400/2400 [==============================] - 30s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 73/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 74/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 75/100\n",
      "2400/2400 [==============================] - 30s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 76/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 77/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 78/100\n",
      "2400/2400 [==============================] - 30s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 79/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 80/100\n",
      "2400/2400 [==============================] - 30s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 81/100\n",
      "2400/2400 [==============================] - 32s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 82/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 83/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 84/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 85/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 86/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 87/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 88/100\n",
      "2400/2400 [==============================] - 30s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 89/100\n",
      "2400/2400 [==============================] - 30s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 90/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 91/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 92/100\n",
      "2400/2400 [==============================] - 30s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 93/100\n",
      "2400/2400 [==============================] - 32s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 94/100\n",
      "2400/2400 [==============================] - 30s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 95/100\n",
      "2400/2400 [==============================] - 29s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 96/100\n",
      "2400/2400 [==============================] - 30s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 97/100\n",
      "2400/2400 [==============================] - 31s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 98/100\n",
      "2400/2400 [==============================] - 30s 13ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 99/100\n",
      "2400/2400 [==============================] - 30s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n",
      "Epoch 100/100\n",
      "2400/2400 [==============================] - 30s 12ms/step - loss: 1.3863 - acc: 0.2525 - val_loss: 1.3868 - val_acc: 0.2000\n"
     ]
    }
   ],
   "source": [
    "# train the data with validation\n",
    "history = model.fit(X_train_sub[0:int(N_train/batch_size)*batch_size, :, :],\n",
    "                    y_train[0:int(N_train/batch_size)*batch_size, :],\n",
    "                    batch_size=batch_size, epochs=num_epoch, shuffle=False,\n",
    "                    validation_data=(X_val_sub, y_val))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "dict_keys(['val_loss', 'val_acc', 'loss', 'acc'])\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f2aa482aa58>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f2aa464ae48>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# list all data in history\n",
    "print(history.history.keys())\n",
    "\n",
    "# summarize history for accuracy\n",
    "plt.plot(history.history['acc'])\n",
    "plt.plot(history.history['val_acc'])\n",
    "plt.title('model accuracy')\n",
    "plt.ylabel('accuracy')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'val'], loc='upper left')\n",
    "plt.show()\n",
    "\n",
    "# summarize history for loss\n",
    "plt.plot(history.history['loss'],'o')\n",
    "plt.plot(history.history['val_loss'],'o')\n",
    "plt.title('model loss')\n",
    "plt.ylabel('loss')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'val'], loc='upper left')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}