Scratchpad.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "import numpy as np\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "import tensorflow as tf\n",
    "\n",
    "from keras import backend as K\n",
    "from keras.models import Sequential, Model\n",
    "from keras.layers import LSTM, Dense, SimpleRNN, Input, merge, Embedding, TimeDistributed\n",
    "from keras.objectives import sparse_categorical_crossentropy, mean_squared_error"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "## Code taken from http://www.rigtorp.se/2011/01/01/rolling-statistics-numpy.html\n",
    "\n",
    "def rolling_window(a, window):\n",
    "    shape = a.shape[:-1] + (a.shape[-1] - window + 1, window)\n",
    "    strides = a.strides + (a.strides[-1],)\n",
    "    return np.lib.stride_tricks.as_strided(a, shape=shape, strides=strides)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(196,)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x7fc4d6929da0>]"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYYAAAD8CAYAAABzTgP2AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nGv5/P1/AGd0SKqUAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fc4d69446a0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "X = np.arange(-10,10,0.1)\n",
    "X = np.cos(np.mean(rolling_window(X, 5), -1))\n",
    "#X = X[:-5+1]\n",
    "print(X.shape)\n",
    "plt.plot(X)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Replicating data generation from http://machinelearningmastery.com/time-series-prediction-lstm-recurrent-neural-networks-python-keras/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def create_dataset(dataset, look_back=1):\n",
    "    dataX, dataY = [], []\n",
    "    for i in range(len(dataset)-look_back-1):\n",
    "        a = dataset[i:(i+look_back)]\n",
    "        dataX.append(a)\n",
    "        dataY.append(dataset[i + look_back])\n",
    "    return np.array(dataX), np.array(dataY)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "131 (131,) 65 (65,)\n"
     ]
    }
   ],
   "source": [
    "train_size = int(X.shape[0]*0.67)\n",
    "test_size = X.shape[0] - train_size\n",
    "train = X[:train_size]\n",
    "test = X[train_size:]\n",
    "print(train_size, train.shape, test_size, test.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(129, 1) (129,) (63, 1) (63,)\n"
     ]
    }
   ],
   "source": [
    "look_back = 1\n",
    "trainX, trainY = create_dataset(train, look_back)\n",
    "testX, testY = create_dataset(test, look_back)\n",
    "\n",
    "print(trainX.shape, trainY.shape, testX.shape, testY.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[-0.93042627]\n",
      " [-0.96236488]\n",
      " [-0.98468786]\n",
      " [-0.99717216]\n",
      " [-0.99969304]\n",
      " [-0.99222533]\n",
      " [-0.97484362]\n",
      " [-0.9477216 ]\n",
      " [-0.91113026]\n",
      " [-0.86543521]]\n",
      "[-0.96236488 -0.98468786 -0.99717216 -0.99969304 -0.99222533 -0.97484362\n",
      " -0.9477216  -0.91113026 -0.86543521 -0.81109301]\n"
     ]
    }
   ],
   "source": [
    "print(trainX[:10])\n",
    "print(trainY[:10])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(10, 1, 1)"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.reshape(trainX[:10], (trainX[:10].shape[0], 1, trainX[:10].shape[1])).shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "# reshape input to be [samples, time steps, features]\n",
    "trainX = np.reshape(trainX, (trainX.shape[0], 1, trainX.shape[1]))\n",
    "testX = np.reshape(testX, (testX.shape[0], 1, testX.shape[1]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Keras model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "model = Sequential()\n",
    "model.add(LSTM(4, input_dim=look_back))\n",
    "model.add(Dense(1))\n",
    "model.compile(loss='mean_squared_error', optimizer='adam')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "129/129 [==============================] - 7s - loss: 0.4783      \b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "Epoch 2/10\n",
      "129/129 [==============================] - 0s - loss: 0.4761     \b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "Epoch 3/10\n",
      "129/129 [==============================] - 0s - loss: 0.4738     \b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "Epoch 4/10\n",
      "129/129 [==============================] - 0s - loss: 0.4715     \b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "Epoch 5/10\n",
      "129/129 [==============================] - 0s - loss: 0.4691     \b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "Epoch 6/10\n",
      "129/129 [==============================] - 0s - loss: 0.4667     \b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "Epoch 7/10\n",
      "129/129 [==============================] - 0s - loss: 0.4642     \b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "Epoch 8/10\n",
      "129/129 [==============================] - 0s - loss: 0.4619     \b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "Epoch 9/10\n",
      "129/129 [==============================] - 0s - loss: 0.4597     \b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "Epoch 10/10\n",
      "129/129 [==============================] - 0s - loss: 0.4572     \b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x7fc4d13ebb00>"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(trainX, trainY, nb_epoch=10, batch_size=50)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((129, 1), (63, 1))"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred_train = model.predict(trainX)\n",
    "pred_test = model.predict(testX)\n",
    "pred_train.shape, pred_test.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x7fc4cc0ea518>]"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYYAAAD8CAYAAABzTgP2AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n4+99i5QXINcAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fc4d11e6198>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(X)\n",
    "plt.plot(np.arange(pred_train.shape[0])+1, pred_train, color='r')\n",
    "plt.plot(np.arange(pred_test.shape[0])+train_size, pred_test, color='g')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "129/129 [==============================] - 0s - loss: 1.0712     \b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "Epoch 2/10\n",
      "129/129 [==============================] - 0s - loss: 0.9684     \b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "Epoch 3/10\n",
      "129/129 [==============================] - 0s - loss: 0.8725     \b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b - ETA: 0s - loss: 0.9015\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "Epoch 4/10\n",
      "129/129 [==============================] - 0s - loss: 0.7779     \b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "Epoch 5/10\n",
      "129/129 [==============================] - 0s - loss: 0.6846     \b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b - ETA: 0s - loss: 0.7559\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "Epoch 6/10\n",
      "129/129 [==============================] - 0s - loss: 0.5916     \b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "Epoch 7/10\n",
      "129/129 [==============================] - 0s - loss: 0.5043     \b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "Epoch 8/10\n",
      "129/129 [==============================] - 0s - loss: 0.4234     \b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "Epoch 9/10\n",
      "129/129 [==============================] - 0s - loss: 0.3488     \b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "Epoch 10/10\n",
      "129/129 [==============================] - 0s - loss: 0.2808     \b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b - ETA: 0s - loss: 0.2873\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x7fc4ac56a470>"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model = Sequential()\n",
    "model.add(SimpleRNN(4, input_dim=look_back))\n",
    "model.add(Dense(1))\n",
    "model.compile(loss='mean_squared_error', optimizer='adam')\n",
    "model.fit(trainX, trainY, nb_epoch=10, batch_size=5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((129, 1), (63, 1))"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred_train = model.predict(trainX)\n",
    "pred_test = model.predict(testX)\n",
    "pred_train.shape, pred_test.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x7fc490f1ebe0>]"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYYAAAD8CAYAAABzTgP2AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nwRhjTBdWGIwxxnRhhcEYY0wX/wfbHGYz+hv3uQAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fc4ac0f6dd8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(X)\n",
    "plt.plot(np.arange(pred_train.shape[0])+1, pred_train, color='r')\n",
    "plt.plot(np.arange(pred_test.shape[0])+train_size, pred_test, color='g')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(129, 1, 1)"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "trainX.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Multi task learning Keras"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((10, 3), (10, 1), (10, 1))"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X = np.random.randn(10,3)\n",
    "W1 = np.random.randn(3,1)\n",
    "W2 = np.random.randn(2,1)\n",
    "y1 = np.dot(X, W1)\n",
    "y2 = np.dot(X[:, :2], W2)\n",
    "\n",
    "X.shape, y1.shape, y2.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "inp1 = Input(name=\"inp1\", shape=(3,))\n",
    "inp2 = Input(name=\"inp2\", shape=(2,))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "dense1 = Dense(10)(inp1)\n",
    "merge_dens1_inp2 = merge([dense1, inp2], mode='concat')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "dense2 = Dense(10)(merge_dens1_inp2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "out1 = Dense(1, name=\"out1\")(dense1)\n",
    "out2 = Dense(1, name=\"out2\")(dense2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "model1 = Model(input=[inp1], output=[out1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model1.get_output_at(0) == out1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "model2 = Model(input=[inp1, inp2], output=[out2])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "model1.compile(optimizer='sgd', loss='mean_squared_error')\n",
    "model2.compile(optimizer='sgd', loss='mean_squared_error')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 0.3986451 ],\n",
       "       [-0.61922395],\n",
       "       [ 0.46553299],\n",
       "       [-0.0886135 ],\n",
       "       [ 0.72533548],\n",
       "       [ 0.02285887],\n",
       "       [ 0.30524015],\n",
       "       [ 0.19303803],\n",
       "       [-0.06842817],\n",
       "       [ 0.78346694]], dtype=float32)"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model2.predict([X, X[:, :2]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 0.76356345],\n",
       "       [-1.10389507],\n",
       "       [ 1.0578301 ],\n",
       "       [-0.39246041],\n",
       "       [ 1.21028972],\n",
       "       [ 0.77841467],\n",
       "       [ 0.66750604],\n",
       "       [-0.25272202],\n",
       "       [-0.48858267],\n",
       "       [ 0.57783008]], dtype=float32)"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model1.predict([X])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['out2']"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model2.output_names"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "model3 = Model(input=[inp1, inp2], output=[out1, out2])\n",
    "model3.compile(optimizer='sgd', loss='mean_squared_error')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[array([[ 0.76356345],\n",
       "        [-1.10389507],\n",
       "        [ 1.0578301 ],\n",
       "        [-0.39246041],\n",
       "        [ 1.21028972],\n",
       "        [ 0.77841467],\n",
       "        [ 0.66750604],\n",
       "        [-0.25272202],\n",
       "        [-0.48858267],\n",
       "        [ 0.57783008]], dtype=float32), array([[ 0.3986451 ],\n",
       "        [-0.61922395],\n",
       "        [ 0.46553299],\n",
       "        [-0.0886135 ],\n",
       "        [ 0.72533548],\n",
       "        [ 0.02285887],\n",
       "        [ 0.30524015],\n",
       "        [ 0.19303803],\n",
       "        [-0.06842817],\n",
       "        [ 0.78346694]], dtype=float32)]"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model3.predict([X, X[:, :2]])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Multi output LSTM example"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "data = [[1,2,3,4,5],\n",
    "        [2,3,2,2,2]\n",
    "       ]\n",
    "label_pos = [[1,2,2,2,1], None]\n",
    "label_ner = [None, [1,1,1,2,1]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "input_seq = Input(shape=(5,), name=\"input_seq\")\n",
    "shared_embedding = Embedding(6, 10)(input_seq)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "TensorShape([Dimension(None), Dimension(5), Dimension(10)])"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "shared_embedding.get_shape()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "shared_layer = LSTM(10, return_sequences=True)(shared_embedding)\n",
    "pos_layer = TimeDistributed(Dense(3, activation=\"softmax\"), name=\"pos_labels\")(shared_layer)\n",
    "ner_layer = TimeDistributed(Dense(3, activation=\"softmax\"), name=\"ner_labels\")(shared_layer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "model = Model(input=input_seq, output=[pos_layer, ner_layer])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "model.compile(loss='sparse_categorical_crossentropy', optimizer='sgd', sample_weight_mode=\"temporal\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((2, 5, 1), (2, 5, 1))"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "max_len = 5\n",
    "X = np.array(data)\n",
    "y_pos = np.expand_dims(np.array([np.zeros(max_len) if y is None else y for y in label_pos], dtype='int'), -1)\n",
    "y_ner = np.expand_dims(np.array([np.zeros(max_len) if y is None else y for y in label_ner], dtype='int'), -1)\n",
    "y_pos.shape, y_ner.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([[ 1.,  1.,  1.,  1.,  1.],\n",
       "        [ 0.,  0.,  0.,  0.,  0.]]), array([[ 0.,  0.,  0.,  0.,  0.],\n",
       "        [ 1.,  1.,  1.,  1.,  1.]]))"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pos_mask = np.array([np.ones(max_len)*(y is not None) for y in label_pos])\n",
    "ner_mask = np.array([np.ones(max_len)*(y is not None) for y in label_ner])\n",
    "pos_mask, ner_mask"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "2/2 [==============================] - 1s - loss: 2.1965 - pos_labels_loss: 1.1076 - ner_labels_loss: 1.0889\n",
      "Epoch 2/10\n",
      "2/2 [==============================] - 0s - loss: 2.1888 - pos_labels_loss: 1.1046 - ner_labels_loss: 1.0842\n",
      "Epoch 3/10\n",
      "2/2 [==============================] - 0s - loss: 2.1811 - pos_labels_loss: 1.1015 - ner_labels_loss: 1.0796\n",
      "Epoch 4/10\n",
      "2/2 [==============================] - 0s - loss: 2.1736 - pos_labels_loss: 1.0985 - ner_labels_loss: 1.0751\n",
      "Epoch 5/10\n",
      "2/2 [==============================] - 0s - loss: 2.1661 - pos_labels_loss: 1.0955 - ner_labels_loss: 1.0706\n",
      "Epoch 6/10\n",
      "2/2 [==============================] - 0s - loss: 2.1587 - pos_labels_loss: 1.0926 - ner_labels_loss: 1.0661\n",
      "Epoch 7/10\n",
      "2/2 [==============================] - 0s - loss: 2.1514 - pos_labels_loss: 1.0896 - ner_labels_loss: 1.0617\n",
      "Epoch 8/10\n",
      "2/2 [==============================] - 0s - loss: 2.1441 - pos_labels_loss: 1.0868 - ner_labels_loss: 1.0574\n",
      "Epoch 9/10\n",
      "2/2 [==============================] - 0s - loss: 2.1369 - pos_labels_loss: 1.0839 - ner_labels_loss: 1.0530\n",
      "Epoch 10/10\n",
      "2/2 [==============================] - 0s - loss: 2.1298 - pos_labels_loss: 1.0811 - ner_labels_loss: 1.0488\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x7fc490a09e48>"
      ]
     },
     "execution_count": 50,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(X, [y_pos, y_ner], sample_weight=[pos_mask, ner_mask])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor 'Mean_24:0' shape=(10,) dtype=float64>"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = mean_squared_error(np.random.randn(10,2), np.random.randn(10,2))\n",
    "t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[ 1.3982274   6.17379382  2.69410353  0.3092882   0.43249701  0.38894633\n",
      "  1.29343803  1.72605789  3.31974207  1.20314927]\n"
     ]
    }
   ],
   "source": [
    "sess = tf.Session()\n",
    "with sess.as_default():\n",
    "    print(t.eval())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor 'Mean_25:0' shape=(10, 2) dtype=float64>"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = mean_squared_error(np.random.randn(10,2,3), np.random.randn(10,2,3))\n",
    "t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[ 1.05625593  0.32636009]\n",
      " [ 2.79147239  2.42880816]\n",
      " [ 3.6945321   0.89615701]\n",
      " [ 5.3059893   2.32529554]\n",
      " [ 1.23866607  1.52450653]\n",
      " [ 2.41590791  1.81906616]\n",
      " [ 5.68505985  0.51474718]\n",
      " [ 7.88741765  0.90664582]\n",
      " [ 2.27765413  1.38585867]\n",
      " [ 3.98567263  3.73537818]]\n"
     ]
    }
   ],
   "source": [
    "sess = tf.Session()\n",
    "with sess.as_default():\n",
    "    print(t.eval())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "def custom_mse_loss(y_true, y_pred):\n",
    "    loss_masks = y_true[\"masks\"]\n",
    "    y_true_vals = y_true[\"vals\"]\n",
    "    mse_loss = mean_squared_error(y_true_vals, y_pred)\n",
    "    mse_loss_masked = loss_masks *  mse_loss\n",
    "    return mse_loss_masked"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "y_pred = np.random.randn(10,5,3)\n",
    "y_true_vals = np.random.randn(10,5,3)\n",
    "y_true_masks = np.random.randint(2, size=(10,1))\n",
    "y_true = {\"masks\": y_true_masks, \"vals\": y_true_vals}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor 'mul_109:0' shape=(10, 5) dtype=float64>"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = custom_mse_loss(y_true, y_pred)\n",
    "t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0],\n",
       "       [1],\n",
       "       [1],\n",
       "       [0],\n",
       "       [0],\n",
       "       [1],\n",
       "       [0],\n",
       "       [0],\n",
       "       [0],\n",
       "       [1]])"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_true_masks"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[ 0.          0.          0.          0.          0.        ]\n",
      " [ 0.59259878  0.37469407  1.83470867  1.35303259  2.49079611]\n",
      " [ 1.02586168  4.83988508  0.31900125  2.57466262  1.25451509]\n",
      " [ 0.          0.          0.          0.          0.        ]\n",
      " [ 0.          0.          0.          0.          0.        ]\n",
      " [ 1.22409462  1.64002248  2.09295172  0.85425327  4.29896931]\n",
      " [ 0.          0.          0.          0.          0.        ]\n",
      " [ 0.          0.          0.          0.          0.        ]\n",
      " [ 0.          0.          0.          0.          0.        ]\n",
      " [ 4.69053322  2.86792321  2.00081542  1.0392119   1.21555662]]\n"
     ]
    }
   ],
   "source": [
    "sess = tf.Session()\n",
    "with sess.as_default():\n",
    "    print(t.eval())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "TensorShape([Dimension(None), Dimension(None), Dimension(3)])"
      ]
     },
     "execution_count": 57,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pos_layer.get_shape()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 96,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def task_based_loss(y_true, y_pred):\n",
    "    print(\"Using layer: \", y_pred.get_shape(), y_pred.name)\n",
    "    pos_labels = K.cast(y_true[:, :, 0], 'int64')\n",
    "    ner_labels = K.cast(y_true[:, :, 1], 'int64')\n",
    "    #pos_labels = K.cast(y_true[0][:, :, 0], 'int64')\n",
    "    #ner_labels = K.cast(y_true[1][:, :, 0], 'int64')\n",
    "    pos_preds = y_pred[0]\n",
    "    ner_preds = y_pred[1]\n",
    "    pos_loss = sparse_categorical_crossentropy(pos_labels, pos_preds)\n",
    "    ner_loss = sparse_categorical_crossentropy(ner_labels, ner_preds)\n",
    "    #pos_mask = y_true[0][:, :, 1] # POS mask channel\n",
    "    #ner_mask = y_true[1][:, :, 1] # NER mask channel\n",
    "    #pos_loss = pos_loss * pos_mask\n",
    "    #ner_loss = ner_loss * ner_mask\n",
    "    return (pos_loss + ner_loss)/2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "metadata": {
    "collapsed": false,
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Using layer:  (?, ?, 3) Reshape_32:0\n",
      "Using layer:  (?, ?, 3) Reshape_34:0\n"
     ]
    }
   ],
   "source": [
    "model.compile(loss=task_based_loss, optimizer='sgd')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 100,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(2, 5)"
      ]
     },
     "execution_count": 100,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 101,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((2, 5), (2, 5), (2, 5), (2, 5))"
      ]
     },
     "execution_count": 101,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_pos.shape, pos_mask.shape, y_ner.shape, ner_mask.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "ename": "ValueError",
     "evalue": "Error when checking model target: expected pos_labels to have 3 dimensions, but got array with shape (2, 5)",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mValueError\u001b[0m                                Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-102-b50ac0cfe42d>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mfit\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mX\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0my_pos\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_ner\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0msample_weight\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mpos_mask\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mner_mask\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/home/napsternxg/anaconda3/lib/python3.5/site-packages/keras/engine/training.py\u001b[0m in \u001b[0;36mfit\u001b[0;34m(self, x, y, batch_size, nb_epoch, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch)\u001b[0m\n\u001b[1;32m   1115\u001b[0m             \u001b[0mclass_weight\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mclass_weight\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1116\u001b[0m             \u001b[0mcheck_batch_axis\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mFalse\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1117\u001b[0;31m             batch_size=batch_size)\n\u001b[0m\u001b[1;32m   1118\u001b[0m         \u001b[0;31m# prepare validation data\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1119\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mvalidation_data\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/home/napsternxg/anaconda3/lib/python3.5/site-packages/keras/engine/training.py\u001b[0m in \u001b[0;36m_standardize_user_data\u001b[0;34m(self, x, y, sample_weight, class_weight, check_batch_axis, batch_size)\u001b[0m\n\u001b[1;32m   1032\u001b[0m                                    \u001b[0moutput_shapes\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1033\u001b[0m                                    \u001b[0mcheck_batch_axis\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mFalse\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1034\u001b[0;31m                                    exception_prefix='model target')\n\u001b[0m\u001b[1;32m   1035\u001b[0m         sample_weights = standardize_sample_weights(sample_weight,\n\u001b[1;32m   1036\u001b[0m                                                     self.output_names)\n",
      "\u001b[0;32m/home/napsternxg/anaconda3/lib/python3.5/site-packages/keras/engine/training.py\u001b[0m in \u001b[0;36mstandardize_input_data\u001b[0;34m(data, names, shapes, check_batch_axis, exception_prefix)\u001b[0m\n\u001b[1;32m    110\u001b[0m                                  \u001b[0;34m' to have '\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0mstr\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mshapes\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mi\u001b[0m\u001b[0;34m]\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    111\u001b[0m                                  \u001b[0;34m' dimensions, but got array with shape '\u001b[0m \u001b[0;34m+\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 112\u001b[0;31m                                  str(array.shape))\n\u001b[0m\u001b[1;32m    113\u001b[0m             \u001b[0;32mfor\u001b[0m \u001b[0mj\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0mdim\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mref_dim\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32min\u001b[0m \u001b[0menumerate\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mzip\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0marray\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshape\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mshapes\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mi\u001b[0m\u001b[0;34m]\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    114\u001b[0m                 \u001b[0;32mif\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0mj\u001b[0m \u001b[0;32mand\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0mcheck_batch_axis\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mValueError\u001b[0m: Error when checking model target: expected pos_labels to have 3 dimensions, but got array with shape (2, 5)"
     ]
    }
   ],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "y_true = np.dstack([y_pos, y_ner, pos_mask, ner_mask])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(2, 5, 4)"
      ]
     },
     "execution_count": 80,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_true.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "ename": "ValueError",
     "evalue": "The model expects 2 input arrays, but only received one array. Found: array with shape (2, 5, 4)",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mValueError\u001b[0m                                Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-79-2bd41e7654de>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mfit\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mX\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_true\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/home/napsternxg/anaconda3/lib/python3.5/site-packages/keras/engine/training.py\u001b[0m in \u001b[0;36mfit\u001b[0;34m(self, x, y, batch_size, nb_epoch, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch)\u001b[0m\n\u001b[1;32m   1115\u001b[0m             \u001b[0mclass_weight\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mclass_weight\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1116\u001b[0m             \u001b[0mcheck_batch_axis\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mFalse\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1117\u001b[0;31m             batch_size=batch_size)\n\u001b[0m\u001b[1;32m   1118\u001b[0m         \u001b[0;31m# prepare validation data\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1119\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mvalidation_data\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/home/napsternxg/anaconda3/lib/python3.5/site-packages/keras/engine/training.py\u001b[0m in \u001b[0;36m_standardize_user_data\u001b[0;34m(self, x, y, sample_weight, class_weight, check_batch_axis, batch_size)\u001b[0m\n\u001b[1;32m   1032\u001b[0m                                    \u001b[0moutput_shapes\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1033\u001b[0m                                    \u001b[0mcheck_batch_axis\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mFalse\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1034\u001b[0;31m                                    exception_prefix='model target')\n\u001b[0m\u001b[1;32m   1035\u001b[0m         sample_weights = standardize_sample_weights(sample_weight,\n\u001b[1;32m   1036\u001b[0m                                                     self.output_names)\n",
      "\u001b[0;32m/home/napsternxg/anaconda3/lib/python3.5/site-packages/keras/engine/training.py\u001b[0m in \u001b[0;36mstandardize_input_data\u001b[0;34m(data, names, shapes, check_batch_axis, exception_prefix)\u001b[0m\n\u001b[1;32m     89\u001b[0m             raise ValueError('The model expects ' + str(len(names)) +\n\u001b[1;32m     90\u001b[0m                              \u001b[0;34m' input arrays, but only received one array. '\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 91\u001b[0;31m                              'Found: array with shape ' + str(data.shape))\n\u001b[0m\u001b[1;32m     92\u001b[0m         \u001b[0marrays\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mdata\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     93\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mValueError\u001b[0m: The model expects 2 input arrays, but only received one array. Found: array with shape (2, 5, 4)"
     ]
    }
   ],
   "source": [
    "model.fit(X, y_true)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((2, 5, 3), (2, 5, 3))"
      ]
     },
     "execution_count": 52,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_pred = model.predict(X)\n",
    "y_pred[0].shape, y_pred[1].shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([[2, 2, 2, 2, 2],\n",
       "        [2, 2, 2, 2, 2]]), array([[1, 1, 1, 1, 1],\n",
       "        [1, 1, 1, 1, 1]]))"
      ]
     },
     "execution_count": 55,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_pred[0].argmax(axis=-1), y_pred[1].argmax(axis=-1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [default]",
   "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": "3.5.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}