tsforecast_dilated_cnn.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [],
   "source": [
    "from __future__ import print_function, division\n",
    "\n",
    "import numpy as np\n",
    "from keras.utils import plot_model\n",
    "from keras.layers import *\n",
    "from keras.models import Sequential, Model\n",
    "from keras import optimizers\n",
    "\n",
    "%matplotlib inline\n",
    "import matplotlib\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "__date__ = '2018-10-25'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "def create_dcnn_model(window_size, kernel_size=2, n_filters=4, n_input_series=1, n_outputs=1):\n",
    "#     model = Sequential((\n",
    "#         # The first conv layer learns `n_filters` filters (aka kernels), each of size ``(kernel_size, n_input_series)``.\n",
    "#         # Its output will have shape (None, window_size - kernel_size + 1, n_filters), i.e., for each position in\n",
    "#         # the input timeseries, the activation of each filter at that position.\n",
    "#         Conv1D(n_filters=n_filters, kernel_size=kernel_size, activation='relu', input_shape=(window_size, n_input_series)),\n",
    "#         MaxPooling1D(),     # Downsample the output of convolution by 2X.\n",
    "#         Convolution1D(n_filters=n_filters, kernel_size=kernel_size, activation='relu'),\n",
    "#         MaxPooling1D(),\n",
    "#         Flatten(),\n",
    "#         Dense(nb_outputs, activation='linear'),     # For binary classification, change the activation to 'sigmoid'\n",
    "#     ))\n",
    "    input_layer = Input(shape=(window_size, n_input_series), name='input_1')\n",
    "    c1_layer = Conv1D(filters=n_filters, kernel_size=kernel_size, dilation_rate=1, \n",
    "                      padding='causal', activation='relu', name='conv1d_1')(input_layer)\n",
    "    c2_layer = Conv1D(filters=n_filters, kernel_size=kernel_size, dilation_rate=2, \n",
    "                      padding='causal', activation='relu', name='conv1d_2')(c1_layer)\n",
    "    c3_layer = Conv1D(filters=n_filters, kernel_size=kernel_size, dilation_rate=4, \n",
    "                      padding='causal', activation='relu', name='conv1d_3')(c2_layer)\n",
    "    c4_layer = concatenate([c1_layer, c3_layer])\n",
    "    #conv_out = Conv1D(8, 1, activation='relu')(c3_layer)\n",
    "    output = Flatten()(c4_layer)\n",
    "    output = Dense(n_outputs, activation='linear')(output)\n",
    "    model = Model(inputs=input_layer, outputs=output)\n",
    "    adam = optimizers.Adam(lr=0.01)\n",
    "    model.compile(loss='mse', optimizer=adam, metrics=['mae'])\n",
    "    # To perform (binary) classification instead:\n",
    "    # model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['binary_accuracy'])\n",
    "    return model\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_1 (InputLayer)            (None, 10, 1)        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "conv1d_1 (Conv1D)               (None, 10, 4)        12          input_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "conv1d_2 (Conv1D)               (None, 10, 4)        36          conv1d_1[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "conv1d_3 (Conv1D)               (None, 10, 4)        36          conv1d_2[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_10 (Concatenate)    (None, 10, 8)        0           conv1d_1[0][0]                   \n",
      "                                                                 conv1d_3[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "flatten_9 (Flatten)             (None, 80)           0           concatenate_10[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "dense_5 (Dense)                 (None, 1)            81          flatten_9[0][0]                  \n",
      "==================================================================================================\n",
      "Total params: 165\n",
      "Trainable params: 165\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model = create_dcnn_model(window_size=10)\n",
    "model.summary()\n",
    "plot_model(model, to_file='dilated_cnn.png') # Run 'pip install pydot' before plotting if this package is missing"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<IPython.core.display.Image object>"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from keras.utils.vis_utils import model_to_dot\n",
    "from IPython.display import Image\n",
    "\n",
    "Image(model_to_dot(model, show_shapes=True).create(prog='dot', format='png'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "def make_timeseries_instances(timeseries, window_size):\n",
    "    \"\"\"Make input features and prediction targets from a `timeseries` for use in machine learning.\n",
    "\n",
    "    :return: A tuple of `(X, y, q)`.  `X` are the inputs to a predictor, a 3D ndarray with shape\n",
    "      ``(timeseries.shape[0] - window_size, window_size, timeseries.shape[1] or 1)``.  For each row of `X`, the\n",
    "      corresponding row of `y` is the next value in the timeseries.  The `q` or query is the last instance, what you would use\n",
    "      to predict a hypothetical next (unprovided) value in the `timeseries`.\n",
    "    :param ndarray timeseries: Either a simple vector, or a matrix of shape ``(timestep, series_num)``, i.e., time is axis 0 (the\n",
    "      row) and the series is axis 1 (the column).\n",
    "    :param int window_size: The number of samples to use as input prediction features (also called the lag or lookback).\n",
    "    \"\"\"\n",
    "    timeseries = np.asarray(timeseries)\n",
    "    assert 0 < window_size < timeseries.shape[0]\n",
    "    X = np.atleast_3d(np.array([timeseries[start:start + window_size] for start in range(0, timeseries.shape[0] - window_size)]))\n",
    "    y = timeseries[window_size:]\n",
    "    q = np.atleast_3d([timeseries[-window_size:]])\n",
    "    return X, y, q"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "def evaluate_timeseries(timeseries, window_size):\n",
    "    \"\"\"Create a 1D CNN regressor to predict the next value in a `timeseries` using the preceding `window_size` elements\n",
    "    as input features and evaluate its performance.\n",
    "\n",
    "    :param ndarray timeseries: Timeseries data with time increasing down the rows (the leading dimension/axis).\n",
    "    :param int window_size: The number of previous timeseries values to use to predict the next.\n",
    "    \"\"\"\n",
    "    filter_length = 5\n",
    "    nb_filter = 4\n",
    "    timeseries = np.atleast_2d(timeseries)\n",
    "    if timeseries.shape[0] == 1:\n",
    "        timeseries = timeseries.T       # Convert 1D vectors to 2D column vectors\n",
    "\n",
    "    nb_samples, nb_series = timeseries.shape\n",
    "    print('\\n\\nTimeseries ({} samples by {} series):\\n'.format(nb_samples, nb_series), timeseries)\n",
    "    model = create_dcnn_model(window_size=window_size, kernel_size=filter_length, n_input_series=nb_series, n_outputs=nb_series, n_filters=nb_filter)\n",
    "    print('\\n\\nModel with input size {}, output size {}, {} conv filters of length {}'.format(model.input_shape, model.output_shape, nb_filter, filter_length))\n",
    "    model.summary()\n",
    "\n",
    "    X, y, q = make_timeseries_instances(timeseries, window_size)\n",
    "    print('\\n\\nInput features:', X, '\\n\\nOutput labels:', y, '\\n\\nQuery vector:', q, sep='\\n')\n",
    "    test_size = int(0.01 * nb_samples)           # In real life you'd want to use 0.2 - 0.5\n",
    "    X_train, X_test, y_train, y_test = X[:-test_size], X[-test_size:], y[:-test_size], y[-test_size:]\n",
    "    model.fit(X_train, y_train, nb_epoch=25, batch_size=2, validation_data=(X_test, y_test))\n",
    "\n",
    "    pred = model.predict(X_test)\n",
    "    print('\\n\\nactual', 'predicted', sep='\\t')\n",
    "    for actual, predicted in zip(y_test, pred.squeeze()):\n",
    "        print(actual.squeeze(), predicted, sep='\\t')\n",
    "    print('next', model.predict(q).squeeze(), sep='\\t')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Simple single timeseries vector prediction\n",
      "\n",
      "\n",
      "Timeseries (1000 samples by 1 series):\n",
      " [[  0]\n",
      " [  1]\n",
      " [  2]\n",
      " ...\n",
      " [997]\n",
      " [998]\n",
      " [999]]\n",
      "\n",
      "\n",
      "Model with input size (None, 50, 1), output size (None, 1), 4 conv filters of length 5\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_1 (InputLayer)            (None, 50, 1)        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "conv1d_1 (Conv1D)               (None, 50, 4)        24          input_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "conv1d_2 (Conv1D)               (None, 50, 4)        84          conv1d_1[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "conv1d_3 (Conv1D)               (None, 50, 4)        84          conv1d_2[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_13 (Concatenate)    (None, 50, 8)        0           conv1d_1[0][0]                   \n",
      "                                                                 conv1d_3[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "flatten_12 (Flatten)            (None, 400)          0           concatenate_13[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "dense_8 (Dense)                 (None, 1)            401         flatten_12[0][0]                 \n",
      "==================================================================================================\n",
      "Total params: 593\n",
      "Trainable params: 593\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "\n",
      "\n",
      "Input features:\n",
      "[[[  0]\n",
      "  [  1]\n",
      "  [  2]\n",
      "  ...\n",
      "  [ 47]\n",
      "  [ 48]\n",
      "  [ 49]]\n",
      "\n",
      " [[  1]\n",
      "  [  2]\n",
      "  [  3]\n",
      "  ...\n",
      "  [ 48]\n",
      "  [ 49]\n",
      "  [ 50]]\n",
      "\n",
      " [[  2]\n",
      "  [  3]\n",
      "  [  4]\n",
      "  ...\n",
      "  [ 49]\n",
      "  [ 50]\n",
      "  [ 51]]\n",
      "\n",
      " ...\n",
      "\n",
      " [[947]\n",
      "  [948]\n",
      "  [949]\n",
      "  ...\n",
      "  [994]\n",
      "  [995]\n",
      "  [996]]\n",
      "\n",
      " [[948]\n",
      "  [949]\n",
      "  [950]\n",
      "  ...\n",
      "  [995]\n",
      "  [996]\n",
      "  [997]]\n",
      "\n",
      " [[949]\n",
      "  [950]\n",
      "  [951]\n",
      "  ...\n",
      "  [996]\n",
      "  [997]\n",
      "  [998]]]\n",
      "\n",
      "\n",
      "Output labels:\n",
      "[[ 50]\n",
      " [ 51]\n",
      " [ 52]\n",
      " ...\n",
      " [997]\n",
      " [998]\n",
      " [999]]\n",
      "\n",
      "\n",
      "Query vector:\n",
      "[[[950]\n",
      "  [951]\n",
      "  [952]\n",
      "  ...\n",
      "  [997]\n",
      "  [998]\n",
      "  [999]]]\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/anaconda/envs/py35/lib/python3.5/site-packages/ipykernel/__main__.py:24: UserWarning: The `nb_epoch` argument in `fit` has been renamed `epochs`.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 940 samples, validate on 10 samples\n",
      "Epoch 1/25\n",
      "940/940 [==============================] - 1s 2ms/step - loss: 2857.7504 - mean_absolute_error: 25.0098 - val_loss: 1.2942 - val_mean_absolute_error: 1.1353\n",
      "Epoch 2/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 581.0652 - mean_absolute_error: 18.1277 - val_loss: 0.1533 - val_mean_absolute_error: 0.3873\n",
      "Epoch 3/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 172.6792 - mean_absolute_error: 9.1360 - val_loss: 229.2880 - val_mean_absolute_error: 15.1421\n",
      "Epoch 4/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 17.2856 - mean_absolute_error: 2.1549 - val_loss: 0.2039 - val_mean_absolute_error: 0.4516\n",
      "Epoch 5/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 0.0845 - mean_absolute_error: 0.1818 - val_loss: 23.0162 - val_mean_absolute_error: 4.7975\n",
      "Epoch 6/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 1272.4199 - mean_absolute_error: 12.3838 - val_loss: 8.2175 - val_mean_absolute_error: 2.8666\n",
      "Epoch 7/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 1.1507 - mean_absolute_error: 0.7623 - val_loss: 16.3420 - val_mean_absolute_error: 4.0425\n",
      "Epoch 8/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 58.9122 - mean_absolute_error: 3.8163 - val_loss: 1223.4502 - val_mean_absolute_error: 34.9777\n",
      "Epoch 9/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 42.8473 - mean_absolute_error: 3.3691 - val_loss: 419.6487 - val_mean_absolute_error: 20.4853\n",
      "Epoch 10/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 970.7839 - mean_absolute_error: 12.9586 - val_loss: 13.6189 - val_mean_absolute_error: 3.6904\n",
      "Epoch 11/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 0.6841 - mean_absolute_error: 0.4789 - val_loss: 0.1294 - val_mean_absolute_error: 0.3597\n",
      "Epoch 12/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 0.6840 - mean_absolute_error: 0.4939 - val_loss: 0.6256 - val_mean_absolute_error: 0.7909\n",
      "Epoch 13/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 666.2777 - mean_absolute_error: 8.3751 - val_loss: 5114.2854 - val_mean_absolute_error: 71.5140\n",
      "Epoch 14/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 399.7670 - mean_absolute_error: 6.3022 - val_loss: 1.4088 - val_mean_absolute_error: 1.1869\n",
      "Epoch 15/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 0.2553 - mean_absolute_error: 0.2935 - val_loss: 0.0305 - val_mean_absolute_error: 0.1745\n",
      "Epoch 16/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 10.9024 - mean_absolute_error: 1.4152 - val_loss: 316.2648 - val_mean_absolute_error: 17.7838\n",
      "Epoch 17/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 323.5693 - mean_absolute_error: 10.1500 - val_loss: 2.2258 - val_mean_absolute_error: 1.4919\n",
      "Epoch 18/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 2.0898 - mean_absolute_error: 0.7805 - val_loss: 2.9434 - val_mean_absolute_error: 1.7156\n",
      "Epoch 19/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 0.6414 - mean_absolute_error: 0.5039 - val_loss: 0.2007 - val_mean_absolute_error: 0.4480\n",
      "Epoch 20/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 1.3732 - mean_absolute_error: 0.6511 - val_loss: 159.2402 - val_mean_absolute_error: 12.6190\n",
      "Epoch 21/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 146.9568 - mean_absolute_error: 5.5990 - val_loss: 0.0272 - val_mean_absolute_error: 0.1650\n",
      "Epoch 22/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 0.9958 - mean_absolute_error: 0.4144 - val_loss: 101.5839 - val_mean_absolute_error: 10.0788\n",
      "Epoch 23/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 2399.0278 - mean_absolute_error: 13.7282 - val_loss: 0.1312 - val_mean_absolute_error: 0.3622\n",
      "Epoch 24/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 6.0341 - mean_absolute_error: 1.6637 - val_loss: 0.2767 - val_mean_absolute_error: 0.5260\n",
      "Epoch 25/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 5.1899 - mean_absolute_error: 1.0854 - val_loss: 57.0282 - val_mean_absolute_error: 7.5517\n",
      "\n",
      "\n",
      "actual\tpredicted\n",
      "990\t982.4714\n",
      "991\t983.4663\n",
      "992\t984.4611\n",
      "993\t985.456\n",
      "994\t986.45087\n",
      "995\t987.44574\n",
      "996\t988.4407\n",
      "997\t989.4354\n",
      "998\t990.43036\n",
      "999\t991.42523\n",
      "next\t992.4201\n",
      "\n",
      "Multiple-input, multiple-output prediction\n",
      "\n",
      "\n",
      "Timeseries (1000 samples by 2 series):\n",
      " [[   0    0]\n",
      " [   1   -1]\n",
      " [   2   -2]\n",
      " ...\n",
      " [ 997 -997]\n",
      " [ 998 -998]\n",
      " [ 999 -999]]\n",
      "\n",
      "\n",
      "Model with input size (None, 50, 2), output size (None, 2), 4 conv filters of length 5\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_1 (InputLayer)            (None, 50, 2)        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "conv1d_1 (Conv1D)               (None, 50, 4)        44          input_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "conv1d_2 (Conv1D)               (None, 50, 4)        84          conv1d_1[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "conv1d_3 (Conv1D)               (None, 50, 4)        84          conv1d_2[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_14 (Concatenate)    (None, 50, 8)        0           conv1d_1[0][0]                   \n",
      "                                                                 conv1d_3[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "flatten_13 (Flatten)            (None, 400)          0           concatenate_14[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "dense_9 (Dense)                 (None, 2)            802         flatten_13[0][0]                 \n",
      "==================================================================================================\n",
      "Total params: 1,014\n",
      "Trainable params: 1,014\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "\n",
      "\n",
      "Input features:\n",
      "[[[   0    0]\n",
      "  [   1   -1]\n",
      "  [   2   -2]\n",
      "  ...\n",
      "  [  47  -47]\n",
      "  [  48  -48]\n",
      "  [  49  -49]]\n",
      "\n",
      " [[   1   -1]\n",
      "  [   2   -2]\n",
      "  [   3   -3]\n",
      "  ...\n",
      "  [  48  -48]\n",
      "  [  49  -49]\n",
      "  [  50  -50]]\n",
      "\n",
      " [[   2   -2]\n",
      "  [   3   -3]\n",
      "  [   4   -4]\n",
      "  ...\n",
      "  [  49  -49]\n",
      "  [  50  -50]\n",
      "  [  51  -51]]\n",
      "\n",
      " ...\n",
      "\n",
      " [[ 947 -947]\n",
      "  [ 948 -948]\n",
      "  [ 949 -949]\n",
      "  ...\n",
      "  [ 994 -994]\n",
      "  [ 995 -995]\n",
      "  [ 996 -996]]\n",
      "\n",
      " [[ 948 -948]\n",
      "  [ 949 -949]\n",
      "  [ 950 -950]\n",
      "  ...\n",
      "  [ 995 -995]\n",
      "  [ 996 -996]\n",
      "  [ 997 -997]]\n",
      "\n",
      " [[ 949 -949]\n",
      "  [ 950 -950]\n",
      "  [ 951 -951]\n",
      "  ...\n",
      "  [ 996 -996]\n",
      "  [ 997 -997]\n",
      "  [ 998 -998]]]\n",
      "\n",
      "\n",
      "Output labels:\n",
      "[[  50  -50]\n",
      " [  51  -51]\n",
      " [  52  -52]\n",
      " ...\n",
      " [ 997 -997]\n",
      " [ 998 -998]\n",
      " [ 999 -999]]\n",
      "\n",
      "\n",
      "Query vector:\n",
      "[[[ 950 -950]\n",
      "  [ 951 -951]\n",
      "  [ 952 -952]\n",
      "  ...\n",
      "  [ 997 -997]\n",
      "  [ 998 -998]\n",
      "  [ 999 -999]]]\n",
      "Train on 940 samples, validate on 10 samples\n",
      "Epoch 1/25\n",
      "940/940 [==============================] - 2s 2ms/step - loss: 3687.4059 - mean_absolute_error: 22.6885 - val_loss: 3.0273 - val_mean_absolute_error: 1.7378\n",
      "Epoch 2/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 0.4896 - mean_absolute_error: 0.4626 - val_loss: 0.0732 - val_mean_absolute_error: 0.2703\n",
      "Epoch 3/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 18.4851 - mean_absolute_error: 2.3765 - val_loss: 317.9837 - val_mean_absolute_error: 17.8213\n",
      "Epoch 4/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 1253.2264 - mean_absolute_error: 14.0771 - val_loss: 37.7021 - val_mean_absolute_error: 6.1348\n",
      "Epoch 5/25\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "940/940 [==============================] - 1s 1ms/step - loss: 177.9678 - mean_absolute_error: 7.0271 - val_loss: 30.9993 - val_mean_absolute_error: 4.5617\n",
      "Epoch 6/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 2673.1432 - mean_absolute_error: 20.3588 - val_loss: 0.2616 - val_mean_absolute_error: 0.5114\n",
      "Epoch 7/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 6080.0359 - mean_absolute_error: 20.8604 - val_loss: 276652.5000 - val_mean_absolute_error: 525.9745\n",
      "Epoch 8/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 1709.8570 - mean_absolute_error: 18.6420 - val_loss: 0.0459 - val_mean_absolute_error: 0.2117\n",
      "Epoch 9/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 0.5067 - mean_absolute_error: 0.2883 - val_loss: 6.8975e-05 - val_mean_absolute_error: 0.0083\n",
      "Epoch 10/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 0.0062 - mean_absolute_error: 0.0291 - val_loss: 0.0022 - val_mean_absolute_error: 0.0472\n",
      "Epoch 11/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 0.1186 - mean_absolute_error: 0.1840 - val_loss: 0.3474 - val_mean_absolute_error: 0.5894\n",
      "Epoch 12/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 1.6124 - mean_absolute_error: 0.6289 - val_loss: 1.8266 - val_mean_absolute_error: 1.3515\n",
      "Epoch 13/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 4847.9921 - mean_absolute_error: 27.0216 - val_loss: 784.7475 - val_mean_absolute_error: 28.0076\n",
      "Epoch 14/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 1679.4444 - mean_absolute_error: 15.5236 - val_loss: 4.7319e-05 - val_mean_absolute_error: 0.0063\n",
      "Epoch 15/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 0.1667 - mean_absolute_error: 0.0999 - val_loss: 0.0012 - val_mean_absolute_error: 0.0334\n",
      "Epoch 16/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 0.4383 - mean_absolute_error: 0.3051 - val_loss: 0.8318 - val_mean_absolute_error: 0.9120\n",
      "Epoch 17/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 862.2348 - mean_absolute_error: 8.9145 - val_loss: 4733.2828 - val_mean_absolute_error: 68.7983\n",
      "Epoch 18/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 96.0483 - mean_absolute_error: 3.1543 - val_loss: 1.2936e-04 - val_mean_absolute_error: 0.0112\n",
      "Epoch 19/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 1818.4319 - mean_absolute_error: 16.2767 - val_loss: 22.4711 - val_mean_absolute_error: 4.7400\n",
      "Epoch 20/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 26.0662 - mean_absolute_error: 2.3850 - val_loss: 0.9407 - val_mean_absolute_error: 0.9699\n",
      "Epoch 21/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 1740.9371 - mean_absolute_error: 12.2314 - val_loss: 10421.6244 - val_mean_absolute_error: 102.0858\n",
      "Epoch 22/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 643.8394 - mean_absolute_error: 14.5608 - val_loss: 5.3918 - val_mean_absolute_error: 2.3220\n",
      "Epoch 23/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 246.6817 - mean_absolute_error: 7.3403 - val_loss: 0.0076 - val_mean_absolute_error: 0.0869\n",
      "Epoch 24/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 437.2667 - mean_absolute_error: 8.8195 - val_loss: 116806.2219 - val_mean_absolute_error: 341.7677\n",
      "Epoch 25/25\n",
      "940/940 [==============================] - 1s 1ms/step - loss: 3361.1823 - mean_absolute_error: 32.8222 - val_loss: 0.6592 - val_mean_absolute_error: 0.8117\n",
      "\n",
      "\n",
      "actual\tpredicted\n",
      "[ 990 -990]\t[ 990.8288 -990.7952]\n",
      "[ 991 -991]\t[ 991.829  -991.7952]\n",
      "[ 992 -992]\t[ 992.82916 -992.79474]\n",
      "[ 993 -993]\t[ 993.8288 -993.795 ]\n",
      "[ 994 -994]\t[ 994.8291 -994.7947]\n",
      "[ 995 -995]\t[ 995.829   -995.79443]\n",
      "[ 996 -996]\t[ 996.8289 -996.7945]\n",
      "[ 997 -997]\t[ 997.8288 -997.7942]\n",
      "[ 998 -998]\t[ 998.82904 -998.7941 ]\n",
      "[ 999 -999]\t[ 999.82904 -999.7939 ]\n",
      "next\t[ 1000.829  -1000.7939]\n"
     ]
    }
   ],
   "source": [
    "def main():\n",
    "    \"\"\"Prepare input data, build model, evaluate.\"\"\"\n",
    "    np.set_printoptions(threshold=25)\n",
    "    ts_length = 1000\n",
    "    window_size = 50\n",
    "\n",
    "    print('\\nSimple single timeseries vector prediction')\n",
    "    timeseries = np.arange(ts_length)                   # The timeseries f(t) = t\n",
    "    evaluate_timeseries(timeseries, window_size)\n",
    "\n",
    "    print('\\nMultiple-input, multiple-output prediction')\n",
    "    timeseries = np.array([np.arange(ts_length), -np.arange(ts_length)]).T      # The timeseries f(t) = [t, -t]\n",
    "    evaluate_timeseries(timeseries, window_size)\n",
    "\n",
    "\n",
    "if __name__ == '__main__':\n",
    "    main()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [conda env:py35]",
   "language": "python",
   "name": "conda-env-py35-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.5.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}