{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "from keras.models import Model\n",
    "from keras.layers import Input\n",
    "from keras.layers.pooling import GlobalMaxPooling2D\n",
    "from keras import backend as K"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def format_decimal(arr, places=6):\n",
    "    return [round(x * 10**places) / 10**places for x in arr]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### GlobalMaxPooling2D"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**[pooling.GlobalMaxPooling2D.0] input 6x6x3, dim_ordering=tf**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tf\n",
      "\n",
      "in shape: (6, 6, 3)\n",
      "in: [0.38825, 0.616859, -0.903684, 0.806884, -0.727288, -0.673966, -0.577175, -0.387585, -0.257311, -0.509074, -0.260107, -0.175546, -0.36369, 0.000508, 0.660501, 0.597718, -0.516095, -0.635432, 0.051806, 0.827972, 0.776157, -0.924706, 0.475891, -0.245901, 0.938555, -0.153749, -0.78604, -0.743216, 0.623975, -0.88669, -0.405761, -0.047077, 0.93782, 0.817619, -0.698627, -0.779825, 0.082966, -0.198865, -0.803914, -0.013523, 0.656264, 0.20553, -0.99514, -0.845919, -0.780934, -0.54633, -0.584082, -0.381479, 0.07203, -0.932858, -0.388963, -0.640372, -0.762033, 0.520289, -0.850815, 0.83055, 0.657538, -0.995219, 0.924136, -0.09163, -0.83776, 0.570101, 0.811735, -0.248129, -0.669459, -0.622836, 0.488153, 0.819239, 0.684766, -0.329173, 0.541295, 0.675745, 0.552436, -0.314967, 0.588244, -0.764561, -0.405427, 0.165388, 0.036045, -0.7066, 0.987815, 0.424845, -0.898345, -0.084084, 0.424759, -0.463452, -0.464603, -0.889256, 0.047486, 0.593987, -0.041217, -0.674434, -0.437693, -0.698919, 0.415884, 0.503516, 0.804258, 0.474695, -0.147267, -0.807198, -0.922972, 0.369532, -0.214226, 0.34421, -0.267339, -0.858829, -0.005027, 0.268902]\n",
      "out shape: (3,)\n",
      "out: [0.938555, 0.924136, 0.987815]\n"
     ]
    }
   ],
   "source": [
    "data_in_shape = (6, 6, 3)\n",
    "L = GlobalMaxPooling2D(dim_ordering='tf')\n",
    "\n",
    "layer_0 = Input(shape=data_in_shape)\n",
    "layer_1 = L(layer_0)\n",
    "model = Model(input=layer_0, output=layer_1)\n",
    "\n",
    "# set weights to random (use seed for reproducibility)\n",
    "np.random.seed(270)\n",
    "data_in = 2 * np.random.random(data_in_shape) - 1\n",
    "print('')\n",
    "print('in shape:', data_in_shape)\n",
    "print('in:', format_decimal(data_in.ravel().tolist()))\n",
    "result = model.predict(np.array([data_in]))\n",
    "print('out shape:', result[0].shape)\n",
    "print('out:', format_decimal(result[0].ravel().tolist()))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**[pooling.GlobalMaxPooling2D.1] input 3x6x6, dim_ordering=th**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "th\n",
      "\n",
      "in shape: (3, 6, 6)\n",
      "in: [0.277769, -0.324717, -0.828538, 0.27327, -0.103567, 0.352973, -0.424245, 0.084882, -0.667369, -0.013292, -0.730917, 0.020447, -0.56258, -0.930644, -0.372492, 0.595927, 0.519345, 0.659999, -0.520708, -0.060843, 0.140459, 0.198302, 0.525008, 0.508564, 0.513985, 0.548433, 0.271675, -0.565162, -0.896829, 0.987411, 0.60257, 0.39488, 0.475653, 0.972433, 0.770687, -0.124055, 0.063721, -0.033112, 0.644369, -0.013292, 0.534668, -0.326544, 0.600593, -0.767345, -0.525822, -0.635478, 0.78529, 0.647604, 0.495353, -0.380054, -0.630392, 0.527641, -0.343892, 0.405197, -0.065455, -0.574106, -0.493058, 0.61698, -0.069963, 0.061398, 0.283554, 0.200134, 0.271897, 0.354207, -0.121057, -0.577198, -0.193319, -0.167341, 0.497023, 0.674803, 0.365241, -0.254262, -0.267867, 0.90223, -0.960562, 0.833171, -0.360657, 0.290601, 0.737442, -0.656173, -0.311816, -0.544177, 0.681235, 0.530767, 0.674875, -0.511685, -0.21598, -0.544874, -0.748065, 0.353896, 0.636164, -0.775521, 0.5865, -0.297541, -0.861248, 0.22867, -0.888671, -0.295104, -0.709528, -0.899564, -0.16393, 0.021623, 0.23474, 0.825311, 0.721459, -0.493389, -0.979736, 0.698471]\n",
      "out shape: (3,)\n",
      "out: [0.987411, 0.78529, 0.90223]\n"
     ]
    }
   ],
   "source": [
    "data_in_shape = (3, 6, 6)\n",
    "L = GlobalMaxPooling2D(dim_ordering='th')\n",
    "\n",
    "layer_0 = Input(shape=data_in_shape)\n",
    "layer_1 = L(layer_0)\n",
    "model = Model(input=layer_0, output=layer_1)\n",
    "\n",
    "# set weights to random (use seed for reproducibility)\n",
    "np.random.seed(271)\n",
    "data_in = 2 * np.random.random(data_in_shape) - 1\n",
    "print('')\n",
    "print('in shape:', data_in_shape)\n",
    "print('in:', format_decimal(data_in.ravel().tolist()))\n",
    "result = model.predict(np.array([data_in]))\n",
    "print('out shape:', result[0].shape)\n",
    "print('out:', format_decimal(result[0].ravel().tolist()))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**[pooling.GlobalMaxPooling2D.2] input 5x3x2, dim_ordering=tf**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tf\n",
      "\n",
      "in shape: (5, 3, 2)\n",
      "in: [-0.504559, 0.757141, 0.875796, -0.387981, -0.0693, -0.652691, -0.555212, -0.434044, 0.235791, -0.605425, 0.718481, -0.209725, 0.39433, -0.062265, -0.773059, 0.044988, 0.788582, -0.418206, 0.614434, -0.961147, 0.540911, 0.289499, 0.209071, -0.023188, 0.527788, -0.515541, 0.942285, 0.522851, -0.896798, -0.850198]\n",
      "out shape: (2,)\n",
      "out: [0.942285, 0.757141]\n"
     ]
    }
   ],
   "source": [
    "data_in_shape = (5, 3, 2)\n",
    "L = GlobalMaxPooling2D(dim_ordering='tf')\n",
    "\n",
    "layer_0 = Input(shape=data_in_shape)\n",
    "layer_1 = L(layer_0)\n",
    "model = Model(input=layer_0, output=layer_1)\n",
    "\n",
    "# set weights to random (use seed for reproducibility)\n",
    "np.random.seed(272)\n",
    "data_in = 2 * np.random.random(data_in_shape) - 1\n",
    "print('')\n",
    "print('in shape:', data_in_shape)\n",
    "print('in:', format_decimal(data_in.ravel().tolist()))\n",
    "result = model.predict(np.array([data_in]))\n",
    "print('out shape:', result[0].shape)\n",
    "print('out:', format_decimal(result[0].ravel().tolist()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "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": 0
}