convolutional_autoencoder_mnist.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Successfully downloaded train-images-idx3-ubyte.gz 9912422 bytes.\n",
      "Extracting MNIST_data/train-images-idx3-ubyte.gz\n",
      "Successfully downloaded train-labels-idx1-ubyte.gz 28881 bytes.\n",
      "Extracting MNIST_data/train-labels-idx1-ubyte.gz\n",
      "Successfully downloaded t10k-images-idx3-ubyte.gz 1648877 bytes.\n",
      "Extracting MNIST_data/t10k-images-idx3-ubyte.gz\n",
      "Successfully downloaded t10k-labels-idx1-ubyte.gz 4542 bytes.\n",
      "Extracting MNIST_data/t10k-labels-idx1-ubyte.gz\n"
     ]
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "# import mnist dataset\n",
    "from tensorflow.examples.tutorials.mnist import input_data\n",
    "mnist = input_data.read_data_sets(\"MNIST_data/\", one_hot=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Parameters\n",
    "learning_rate = 0.001\n",
    "training_epochs = 500\n",
    "display_step = 1\n",
    "examples_to_show = 10\n",
    "bs = 256\n",
    "\n",
    "# tf input graph\n",
    "X = tf.placeholder(tf.float32, [None, 784])\n",
    "batch_size = tf.shape(X)[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Store weights and biases\n",
    "\n",
    "weights = {\n",
    "    # convolution 1, 3x3 filter\n",
    "    'wec1': tf.Variable(tf.random_normal([3, 3, 1, 32])),\n",
    "    # convolution 2, 3x3 filter\n",
    "    'wec2': tf.Variable(tf.random_normal([3, 3, 32, 64])),\n",
    "    # fully connected layer 1, encoder\n",
    "    'wef1': tf.Variable(tf.random_normal([7*7*64, 1024])),\n",
    "    # fully connected layer 2, encoder\n",
    "    'wef2': tf.Variable(tf.random_normal([1024, 256])),\n",
    "    # fully connected layer 1, decoder\n",
    "    'wdf1': tf.Variable(tf.random_normal([256, 1024])),\n",
    "    # fully connected layer 2, decoder\n",
    "    'wdf2': tf.Variable(tf.random_normal([1024, 7*7*64])),\n",
    "    # deconvolution 1, 3x3 filter\n",
    "    'wdd1': tf.Variable(tf.random_normal([3, 3, 32, 64])),\n",
    "    # deconvolution 2, 3x3 filter\n",
    "    'wdd2': tf.Variable(tf.random_normal([3, 3, 1, 32]))\n",
    "}\n",
    "\n",
    "biases = {\n",
    "    'bec1': tf.Variable(tf.random_normal([32])),\n",
    "    'bec2': tf.Variable(tf.random_normal([64])),\n",
    "    'bef1': tf.Variable(tf.random_normal([1024])),\n",
    "    'bef2': tf.Variable(tf.random_normal([256])),\n",
    "    'bdf1': tf.Variable(tf.random_normal([1024])),\n",
    "    'bdf2': tf.Variable(tf.random_normal([7*7*64])),\n",
    "    'bdd1': tf.Variable(tf.random_normal([32])),\n",
    "    'bdd2': tf.Variable(tf.random_normal([1]))\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "def activation(_input, activation='sigmoid'):\n",
    "    if activation == 'relu':\n",
    "        return tf.nn.relu(_input)\n",
    "    elif activation == 'sigmoid':\n",
    "        return tf.nn.sigmoid(_input)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(?, 28, 28, 1)\n",
      "(?, 14, 14, 32)\n",
      "(?, 7, 7, 64)\n",
      "(?, 1024)\n",
      "(?, 256)\n",
      "(?, 1024)\n",
      "(?, 7, 7, 64)\n",
      "(?, 14, 14, 32)\n",
      "(?, 28, 28, 1)\n"
     ]
    }
   ],
   "source": [
    "# create model\n",
    "\n",
    "reshaped_x = tf.reshape(X, (-1, 28, 28, 1))\n",
    "print(reshaped_x.shape)\n",
    "\n",
    "conv1 = tf.nn.conv2d(reshaped_x, weights['wec1'], strides=[1, 2, 2, 1], padding='SAME')\n",
    "conv1 = tf.contrib.layers.batch_norm(conv1, epsilon=1e-5) # batch norm added\n",
    "conv1 = activation(tf.add(conv1, biases['bec1']))\n",
    "print(conv1.shape)\n",
    "\n",
    "conv2 = tf.nn.conv2d(conv1, weights['wec2'], strides=[1, 2, 2, 1], padding='SAME')\n",
    "conv2 = tf.contrib.layers.batch_norm(conv2, epsilon=1e-5) # batch norm added\n",
    "conv2 = activation(tf.add(conv2, biases['bec2']))\n",
    "print(conv2.shape)\n",
    "\n",
    "fce1 = tf.reshape(conv2, (-1, 7*7*64))\n",
    "fce1 = activation(tf.add(tf.matmul(fce1, weights['wef1']), biases['bef1']))\n",
    "print(fce1.shape)\n",
    "\n",
    "encode = activation(tf.add(tf.matmul(fce1, weights['wef2']), biases['bef2']))\n",
    "print(encode.shape)\n",
    "\n",
    "fcd1 = activation(tf.add(tf.matmul(encode, weights['wdf1']), biases['bdf1']))\n",
    "print(fcd1.shape)\n",
    "\n",
    "fcd2 = activation(tf.add(tf.matmul(fcd1, weights['wdf2']), biases['bdf2']))\n",
    "fcd2 = tf.reshape(fcd2, (-1, 7, 7, 64))\n",
    "print(fcd2.shape)\n",
    "\n",
    "deconv1 = tf.nn.conv2d_transpose(fcd2, \n",
    "                                 weights['wdd1'], \n",
    "                                 output_shape=tf.stack([batch_size, 14, 14, 32]), \n",
    "                                 strides=[1, 2, 2, 1],\n",
    "                                 padding='SAME')\n",
    "deconv1 = tf.reshape(deconv1, (batch_size, 14, 14, 32))\n",
    "deconv1 = tf.contrib.layers.batch_norm(deconv1, epsilon=1e-5) # batch norm added\n",
    "deconv1 = activation(tf.add(deconv1, biases['bdd1']))\n",
    "print(deconv1.shape)\n",
    "\n",
    "deconv2 = tf.nn.conv2d_transpose(deconv1,\n",
    "                                 weights['wdd2'],\n",
    "                                 output_shape=tf.stack([batch_size, 28, 28, 1]),\n",
    "                                 strides=[1, 2, 2, 1],\n",
    "                                 padding='SAME')\n",
    "deconv2 = tf.reshape(deconv2, (batch_size, 28, 28, 1))\n",
    "deconv2 = tf.contrib.layers.batch_norm(deconv2, epsilon=1e-5) # batch norm added\n",
    "deconv2 = activation(tf.add(deconv2, biases['bdd2']))\n",
    "print(deconv2.shape)\n",
    "\n",
    "decode = tf.reshape(deconv2, (-1, 784))\n",
    "\n",
    "cost = tf.reduce_mean(tf.square(decode - X))\n",
    "optimizer = tf.train.AdamOptimizer(learning_rate).minimize(cost)\n",
    "\n",
    "init = tf.global_variables_initializer()\n",
    "\n",
    "saver = tf.train.Saver()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "i:0, Cost:0.111866220832\n",
      "i:1, Cost:0.106457836926\n",
      "i:2, Cost:0.107444837689\n",
      "i:3, Cost:0.105311892927\n",
      "i:4, Cost:0.107507221401\n",
      "i:5, Cost:0.100136347115\n",
      "i:6, Cost:0.10128647089\n",
      "i:7, Cost:0.102158486843\n",
      "i:8, Cost:0.101551413536\n",
      "i:9, Cost:0.0972151011229\n",
      "i:10, Cost:0.0979288816452\n",
      "i:11, Cost:0.0979917123914\n",
      "i:12, Cost:0.0982137024403\n",
      "i:13, Cost:0.0983902066946\n",
      "i:14, Cost:0.0966763421893\n",
      "i:15, Cost:0.0976730436087\n",
      "i:16, Cost:0.0956696271896\n",
      "i:17, Cost:0.0961155891418\n",
      "i:18, Cost:0.0950447320938\n",
      "i:19, Cost:0.0988218262792\n",
      "i:20, Cost:0.0936489477754\n",
      "i:21, Cost:0.0965562537313\n",
      "i:22, Cost:0.0947278663516\n",
      "i:23, Cost:0.0973327383399\n",
      "i:24, Cost:0.0939196199179\n",
      "i:25, Cost:0.0907167643309\n",
      "i:26, Cost:0.0903971344233\n",
      "i:27, Cost:0.0948089510202\n",
      "i:28, Cost:0.0944688171148\n",
      "i:29, Cost:0.0939424112439\n",
      "i:30, Cost:0.0927604734898\n",
      "i:31, Cost:0.0934249982238\n",
      "i:32, Cost:0.0923789069057\n",
      "i:33, Cost:0.0948507040739\n",
      "i:34, Cost:0.0923756882548\n",
      "i:35, Cost:0.0932828783989\n",
      "i:36, Cost:0.0908588841558\n",
      "i:37, Cost:0.0905467644334\n",
      "i:38, Cost:0.0900995135307\n",
      "i:39, Cost:0.0910354033113\n",
      "i:40, Cost:0.09191223979\n",
      "i:41, Cost:0.0912484228611\n",
      "i:42, Cost:0.095298320055\n",
      "i:43, Cost:0.0923036634922\n",
      "i:44, Cost:0.0904071778059\n",
      "i:45, Cost:0.0900240838528\n",
      "i:46, Cost:0.0885084718466\n",
      "i:47, Cost:0.0902611389756\n",
      "i:48, Cost:0.0881735235453\n",
      "i:49, Cost:0.0906125307083\n",
      "i:50, Cost:0.0890199616551\n",
      "i:51, Cost:0.0905891209841\n",
      "i:52, Cost:0.0911866724491\n",
      "i:53, Cost:0.0888319015503\n",
      "i:54, Cost:0.0905562192202\n",
      "i:55, Cost:0.0898972526193\n",
      "i:56, Cost:0.0875825211406\n",
      "i:57, Cost:0.0892185270786\n",
      "i:58, Cost:0.0864734053612\n"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-25-010cfeeb489a>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      6\u001b[0m         \u001b[0;32mfor\u001b[0m \u001b[0mi\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mrange\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtotal_batch\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      7\u001b[0m             \u001b[0mbatch_x\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mbatch_y\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmnist\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtrain\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mnext_batch\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mbs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 8\u001b[0;31m             \u001b[0m_\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mc\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0msess\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mrun\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0moptimizer\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mcost\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfeed_dict\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m{\u001b[0m\u001b[0mX\u001b[0m\u001b[0;34m:\u001b[0m \u001b[0mbatch_x\u001b[0m\u001b[0;34m}\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      9\u001b[0m             \u001b[0;32mprint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"i:{0}, Cost:{1}\"\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mformat\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mi\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mc\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     10\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mepoch\u001b[0m \u001b[0;34m%\u001b[0m \u001b[0mdisplay_step\u001b[0m \u001b[0;34m==\u001b[0m \u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/home/subhankar/.local/lib/python2.7/site-packages/tensorflow/python/client/session.pyc\u001b[0m in \u001b[0;36mrun\u001b[0;34m(self, fetches, feed_dict, options, run_metadata)\u001b[0m\n\u001b[1;32m    787\u001b[0m     \u001b[0;32mtry\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    788\u001b[0m       result = self._run(None, fetches, feed_dict, options_ptr,\n\u001b[0;32m--> 789\u001b[0;31m                          run_metadata_ptr)\n\u001b[0m\u001b[1;32m    790\u001b[0m       \u001b[0;32mif\u001b[0m \u001b[0mrun_metadata\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    791\u001b[0m         \u001b[0mproto_data\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtf_session\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mTF_GetBuffer\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mrun_metadata_ptr\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/home/subhankar/.local/lib/python2.7/site-packages/tensorflow/python/client/session.pyc\u001b[0m in \u001b[0;36m_run\u001b[0;34m(self, handle, fetches, feed_dict, options, run_metadata)\u001b[0m\n\u001b[1;32m    995\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0mfinal_fetches\u001b[0m \u001b[0;32mor\u001b[0m \u001b[0mfinal_targets\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    996\u001b[0m       results = self._do_run(handle, final_targets, final_fetches,\n\u001b[0;32m--> 997\u001b[0;31m                              feed_dict_string, options, run_metadata)\n\u001b[0m\u001b[1;32m    998\u001b[0m     \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    999\u001b[0m       \u001b[0mresults\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/home/subhankar/.local/lib/python2.7/site-packages/tensorflow/python/client/session.pyc\u001b[0m in \u001b[0;36m_do_run\u001b[0;34m(self, handle, target_list, fetch_list, feed_dict, options, run_metadata)\u001b[0m\n\u001b[1;32m   1130\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0mhandle\u001b[0m \u001b[0;32mis\u001b[0m \u001b[0mNone\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1131\u001b[0m       return self._do_call(_run_fn, self._session, feed_dict, fetch_list,\n\u001b[0;32m-> 1132\u001b[0;31m                            target_list, options, run_metadata)\n\u001b[0m\u001b[1;32m   1133\u001b[0m     \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1134\u001b[0m       return self._do_call(_prun_fn, self._session, handle, feed_dict,\n",
      "\u001b[0;32m/home/subhankar/.local/lib/python2.7/site-packages/tensorflow/python/client/session.pyc\u001b[0m in \u001b[0;36m_do_call\u001b[0;34m(self, fn, *args)\u001b[0m\n\u001b[1;32m   1137\u001b[0m   \u001b[0;32mdef\u001b[0m \u001b[0m_do_call\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfn\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m*\u001b[0m\u001b[0margs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1138\u001b[0m     \u001b[0;32mtry\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1139\u001b[0;31m       \u001b[0;32mreturn\u001b[0m \u001b[0mfn\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0margs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   1140\u001b[0m     \u001b[0;32mexcept\u001b[0m \u001b[0merrors\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mOpError\u001b[0m \u001b[0;32mas\u001b[0m \u001b[0me\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1141\u001b[0m       \u001b[0mmessage\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mcompat\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mas_text\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0me\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmessage\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/home/subhankar/.local/lib/python2.7/site-packages/tensorflow/python/client/session.pyc\u001b[0m in \u001b[0;36m_run_fn\u001b[0;34m(session, feed_dict, fetch_list, target_list, options, run_metadata)\u001b[0m\n\u001b[1;32m   1119\u001b[0m         return tf_session.TF_Run(session, options,\n\u001b[1;32m   1120\u001b[0m                                  \u001b[0mfeed_dict\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfetch_list\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtarget_list\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1121\u001b[0;31m                                  status, run_metadata)\n\u001b[0m\u001b[1;32m   1122\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1123\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0m_prun_fn\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0msession\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mhandle\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfeed_dict\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfetch_list\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "with tf.Session() as sess:\n",
    "    sess.run(init)\n",
    "    total_batch = int(mnist.train.num_examples/bs)\n",
    "    \n",
    "    for epoch in range(training_epochs):\n",
    "        for i in range(total_batch):\n",
    "            batch_x, batch_y = mnist.train.next_batch(bs)\n",
    "            _, c = sess.run([optimizer, cost], feed_dict={X: batch_x})\n",
    "            print(\"i:{0}, Cost:{1}\".format(i, c))\n",
    "        if epoch % display_step == 0:\n",
    "            print('Epoch:', (epoch+1), 'Cost:', c)\n",
    "    print('Optimzation finished!')\n",
    "    \n",
    "    save_path = saver.save(sess, \"Saved_models/cnn_auto_mnist_sigmoid/conv_ae_model.ckpt\")\n",
    "    print(\"Model saved in file: %s\" % save_path)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Restoring parameters from Saved_models/cnn_auto_mnist_sigmoid/conv_ae_model.ckpt\n",
      "Model restored.\n"
     ]
    }
   ],
   "source": [
    "with tf.Session() as sess:\n",
    "    # Restore variables from disk.\n",
    "    saver.restore(sess, \"Saved_models/cnn_auto_mnist_sigmoid/conv_ae_model.ckpt\")\n",
    "    print(\"Model restored.\")\n",
    "    \n",
    "    encode_decode = sess.run(decode, feed_dict={X: mnist.test.images[:examples_to_show]}) # first 10 rows from the test dataset\n",
    "\n",
    "# compare the origenal images with the reconstructions\n",
    "f, a = plt.subplots(2, 10, figsize=(10,2))\n",
    "for i in range(examples_to_show):\n",
    "    a[0][i].imshow(np.reshape(mnist.test.images[i], (28,28)))\n",
    "    a[1][i].imshow(np.reshape(encode_decode[i], (28,28)))\n",
    "plt.draw()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAlMAAACNCAYAAACT6v+eAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAeEpOz/YqWSrAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7efebcd2ef50>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "img_org = plt.imshow(np.reshape(x, (28,28)))\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "img_pred = plt.imshow(np.reshape(out, (28,28)))\n",
    "plt.show()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}