Convolutional Autoencoder For Denoising.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Denoising\n",
    "\n",
    "Autoencoders can be used to denoise images quite successfully just by training the network on noisy images. We can create the noisy images ourselves by adding Gaussian noise to the training images, then clipping the values to be between 0 and 1. We'll use noisy images as input and the original, clean images as targets. Here's an example of the noisy images I generated and the denoised images.\n",
    "\n",
    "![Denoising autoencoder](assets/denoising.png)\n",
    "\n",
    "\n",
    "Since this is a harder problem for the network, we'll want to use deeper convolutional layers here, more feature maps. I suggest something like 32-32-16 for the depths of the convolutional layers in the encoder, and the same depths going backward through the decoder. Otherwise the architecture is the same as the convolutional autoencoder.\n",
    "\n",
    "> **Exercise:** Build the network for the denoising autoencoder. It's the same as before, but with deeper layers. I suggest 32-32-16 for the depths, but you can play with these numbers, or add more layers."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Extracting MNIST_data/train-images-idx3-ubyte.gz\n",
      "Extracting MNIST_data/train-labels-idx1-ubyte.gz\n",
      "Extracting MNIST_data/t10k-images-idx3-ubyte.gz\n",
      "Extracting MNIST_data/t10k-labels-idx1-ubyte.gz\n"
     ]
    }
   ],
   "source": [
    "from tensorflow.examples.tutorials.mnist import input_data\n",
    "mnist = input_data.read_data_sets('MNIST_data', validation_size=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7f381eb75b38>"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nwy8VZfilogy/VJThl4r6f9/eEE4pKxLvAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f3822b03a58>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "img = mnist.train.images[3]\n",
    "plt.imshow(img.reshape((28, 28)), cmap='Greys_r')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "learning_rate = 0.001\n",
    "inputs_ = tf.placeholder(tf.float32, (None, 28, 28, 1), name='inputs')\n",
    "targets_ = tf.placeholder(tf.float32, (None, 28, 28, 1), name='targets')\n",
    "kernel_dimension = 5\n",
    "pool_dimension = 2\n",
    "pool_stride = 2\n",
    "\n",
    "### Encoder\n",
    "conv1 = tf.layers.conv2d(inputs_,32 ,(kernel_dimension,kernel_dimension),padding='same',activation=tf.nn.relu)\n",
    "# Now 28x28x32\n",
    "maxpool1 = tf.layers.max_pooling2d(conv1,(pool_dimension,pool_dimension),(pool_stride,pool_stride),padding='same')\n",
    "# Now 14x14x32\n",
    "conv2 = tf.layers.conv2d(maxpool1,32,(kernel_dimension,kernel_dimension),padding='same',activation=tf.nn.relu)\n",
    "# Now 14x14x32\n",
    "maxpool2 =  tf.layers.max_pooling2d(conv2,(pool_dimension,pool_dimension),(pool_stride,pool_stride),padding='same')\n",
    "# Now 7x7x32\n",
    "conv3 = tf.layers.conv2d(maxpool2,16,(kernel_dimension,kernel_dimension),padding='same',activation=tf.nn.relu)\n",
    "# Now 7x7x16\n",
    "encoded = tf.layers.max_pooling2d(conv3,(pool_dimension,pool_dimension),(pool_stride,pool_stride),padding='same')\n",
    "# Now 4x4x16\n",
    "\n",
    "### Decoder\n",
    "upsample1 = tf.image.resize_nearest_neighbor(encoded,(7,7))\n",
    "# Now 7x7x16\n",
    "conv4 = tf.layers.conv2d(upsample1,16,(kernel_dimension,kernel_dimension),padding='same',activation=tf.nn.relu)\n",
    "# Now 7x7x16\n",
    "upsample2 = tf.image.resize_nearest_neighbor(conv4,(14,14))\n",
    "# Now 14x14x16\n",
    "conv5 = tf.layers.conv2d(upsample2,32,(kernel_dimension,kernel_dimension),padding='same',activation=tf.nn.relu)\n",
    "# Now 14x14x32\n",
    "upsample3 = tf.image.resize_nearest_neighbor(conv5,(28,28))\n",
    "# Now 28x28x32\n",
    "conv6 = tf.layers.conv2d(upsample3,32,(kernel_dimension,kernel_dimension),padding='same',activation=tf.nn.relu)\n",
    "# Now 28x28x32\n",
    "\n",
    "logits = tf.layers.conv2d(conv6,1,(kernel_dimension,kernel_dimension),padding='same',activation=None)\n",
    "#Now 28x28x1\n",
    "\n",
    "# Pass logits through sigmoid to get reconstructed image\n",
    "decoded = tf.nn.sigmoid(logits)\n",
    "\n",
    "# Pass logits through sigmoid and calculate the cross-entropy loss\n",
    "loss = tf.nn.sigmoid_cross_entropy_with_logits(logits=logits,labels=targets_)\n",
    "\n",
    "# Get cost and define the optimizer\n",
    "cost = tf.reduce_mean(loss)\n",
    "opt = tf.train.AdamOptimizer(learning_rate).minimize(cost)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sess = tf.Session()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 1/100... Training loss: 0.6892\n",
      "Epoch: 1/100... Training loss: 0.2068\n",
      "Epoch: 1/100... Training loss: 0.1672\n",
      "Epoch: 2/100... Training loss: 0.1418\n",
      "Epoch: 2/100... Training loss: 0.1363\n",
      "Epoch: 2/100... Training loss: 0.1286\n",
      "Epoch: 3/100... Training loss: 0.1220\n",
      "Epoch: 3/100... Training loss: 0.1210\n",
      "Epoch: 3/100... Training loss: 0.1201\n",
      "Epoch: 4/100... Training loss: 0.1207\n",
      "Epoch: 4/100... Training loss: 0.1172\n",
      "Epoch: 4/100... Training loss: 0.1131\n",
      "Epoch: 5/100... Training loss: 0.1157\n",
      "Epoch: 5/100... Training loss: 0.1147\n",
      "Epoch: 5/100... Training loss: 0.1121\n",
      "Epoch: 6/100... Training loss: 0.1089\n",
      "Epoch: 6/100... Training loss: 0.1079\n",
      "Epoch: 6/100... Training loss: 0.1106\n",
      "Epoch: 7/100... Training loss: 0.1066\n",
      "Epoch: 7/100... Training loss: 0.1073\n",
      "Epoch: 7/100... Training loss: 0.1053\n",
      "Epoch: 8/100... Training loss: 0.1066\n",
      "Epoch: 8/100... Training loss: 0.1036\n",
      "Epoch: 8/100... Training loss: 0.1057\n",
      "Epoch: 9/100... Training loss: 0.1047\n",
      "Epoch: 9/100... Training loss: 0.1036\n",
      "Epoch: 9/100... Training loss: 0.1034\n",
      "Epoch: 10/100... Training loss: 0.1030\n",
      "Epoch: 10/100... Training loss: 0.1030\n",
      "Epoch: 10/100... Training loss: 0.1019\n",
      "Epoch: 11/100... Training loss: 0.0991\n",
      "Epoch: 11/100... Training loss: 0.1022\n",
      "Epoch: 11/100... Training loss: 0.1002\n",
      "Epoch: 12/100... Training loss: 0.1012\n",
      "Epoch: 12/100... Training loss: 0.0973\n",
      "Epoch: 12/100... Training loss: 0.1012\n",
      "Epoch: 13/100... Training loss: 0.1037\n",
      "Epoch: 13/100... Training loss: 0.1046\n",
      "Epoch: 13/100... Training loss: 0.1044\n",
      "Epoch: 14/100... Training loss: 0.1001\n",
      "Epoch: 14/100... Training loss: 0.0993\n",
      "Epoch: 14/100... Training loss: 0.1039\n",
      "Epoch: 15/100... Training loss: 0.1038\n",
      "Epoch: 15/100... Training loss: 0.0994\n",
      "Epoch: 15/100... Training loss: 0.0981\n",
      "Epoch: 16/100... Training loss: 0.0987\n",
      "Epoch: 16/100... Training loss: 0.1012\n",
      "Epoch: 16/100... Training loss: 0.0949\n",
      "Epoch: 17/100... Training loss: 0.1004\n",
      "Epoch: 17/100... Training loss: 0.1025\n",
      "Epoch: 17/100... Training loss: 0.0991\n",
      "Epoch: 18/100... Training loss: 0.0986\n",
      "Epoch: 18/100... Training loss: 0.0972\n",
      "Epoch: 18/100... Training loss: 0.1016\n",
      "Epoch: 19/100... Training loss: 0.0976\n",
      "Epoch: 19/100... Training loss: 0.0968\n",
      "Epoch: 19/100... Training loss: 0.0980\n",
      "Epoch: 20/100... Training loss: 0.1000\n",
      "Epoch: 20/100... Training loss: 0.0982\n",
      "Epoch: 20/100... Training loss: 0.0944\n",
      "Epoch: 21/100... Training loss: 0.0981\n",
      "Epoch: 21/100... Training loss: 0.0954\n",
      "Epoch: 21/100... Training loss: 0.0951\n",
      "Epoch: 22/100... Training loss: 0.0991\n",
      "Epoch: 22/100... Training loss: 0.0991\n",
      "Epoch: 22/100... Training loss: 0.0975\n",
      "Epoch: 23/100... Training loss: 0.0987\n",
      "Epoch: 23/100... Training loss: 0.0969\n",
      "Epoch: 23/100... Training loss: 0.0936\n",
      "Epoch: 24/100... Training loss: 0.0964\n",
      "Epoch: 24/100... Training loss: 0.0952\n",
      "Epoch: 24/100... Training loss: 0.0962\n",
      "Epoch: 25/100... Training loss: 0.0952\n",
      "Epoch: 25/100... Training loss: 0.0971\n",
      "Epoch: 25/100... Training loss: 0.0915\n",
      "Epoch: 26/100... Training loss: 0.0987\n",
      "Epoch: 26/100... Training loss: 0.0992\n",
      "Epoch: 26/100... Training loss: 0.0945\n",
      "Epoch: 27/100... Training loss: 0.0982\n",
      "Epoch: 27/100... Training loss: 0.0970\n",
      "Epoch: 27/100... Training loss: 0.0966\n",
      "Epoch: 28/100... Training loss: 0.0976\n",
      "Epoch: 28/100... Training loss: 0.0946\n",
      "Epoch: 28/100... Training loss: 0.0941\n",
      "Epoch: 29/100... Training loss: 0.0979\n",
      "Epoch: 29/100... Training loss: 0.0956\n",
      "Epoch: 29/100... Training loss: 0.0956\n",
      "Epoch: 30/100... Training loss: 0.0947\n",
      "Epoch: 30/100... Training loss: 0.0978\n",
      "Epoch: 30/100... Training loss: 0.0940\n",
      "Epoch: 31/100... Training loss: 0.0973\n",
      "Epoch: 31/100... Training loss: 0.0933\n",
      "Epoch: 31/100... Training loss: 0.0976\n",
      "Epoch: 32/100... Training loss: 0.0956\n",
      "Epoch: 32/100... Training loss: 0.0952\n",
      "Epoch: 32/100... Training loss: 0.0927\n",
      "Epoch: 33/100... Training loss: 0.0980\n",
      "Epoch: 33/100... Training loss: 0.0951\n",
      "Epoch: 33/100... Training loss: 0.0964\n",
      "Epoch: 34/100... Training loss: 0.0954\n",
      "Epoch: 34/100... Training loss: 0.0959\n",
      "Epoch: 34/100... Training loss: 0.0981\n",
      "Epoch: 35/100... Training loss: 0.0955\n",
      "Epoch: 35/100... Training loss: 0.0953\n",
      "Epoch: 35/100... Training loss: 0.0947\n",
      "Epoch: 36/100... Training loss: 0.0951\n",
      "Epoch: 36/100... Training loss: 0.0941\n",
      "Epoch: 36/100... Training loss: 0.0958\n",
      "Epoch: 37/100... Training loss: 0.0943\n",
      "Epoch: 37/100... Training loss: 0.0925\n",
      "Epoch: 37/100... Training loss: 0.0892\n",
      "Epoch: 38/100... Training loss: 0.0943\n",
      "Epoch: 38/100... Training loss: 0.0935\n",
      "Epoch: 38/100... Training loss: 0.0924\n",
      "Epoch: 39/100... Training loss: 0.0959\n",
      "Epoch: 39/100... Training loss: 0.0925\n",
      "Epoch: 39/100... Training loss: 0.0906\n",
      "Epoch: 40/100... Training loss: 0.0993\n",
      "Epoch: 40/100... Training loss: 0.0947\n",
      "Epoch: 40/100... Training loss: 0.0941\n",
      "Epoch: 41/100... Training loss: 0.0923\n",
      "Epoch: 41/100... Training loss: 0.0921\n",
      "Epoch: 41/100... Training loss: 0.0903\n",
      "Epoch: 42/100... Training loss: 0.0931\n",
      "Epoch: 42/100... Training loss: 0.0927\n",
      "Epoch: 42/100... Training loss: 0.0962\n",
      "Epoch: 43/100... Training loss: 0.0910\n",
      "Epoch: 43/100... Training loss: 0.0958\n",
      "Epoch: 43/100... Training loss: 0.0927\n",
      "Epoch: 44/100... Training loss: 0.0966\n",
      "Epoch: 44/100... Training loss: 0.0912\n",
      "Epoch: 44/100... Training loss: 0.0926\n",
      "Epoch: 45/100... Training loss: 0.0951\n",
      "Epoch: 45/100... Training loss: 0.0930\n",
      "Epoch: 45/100... Training loss: 0.0957\n",
      "Epoch: 46/100... Training loss: 0.0944\n",
      "Epoch: 46/100... Training loss: 0.0938\n",
      "Epoch: 46/100... Training loss: 0.0925\n",
      "Epoch: 47/100... Training loss: 0.0939\n",
      "Epoch: 47/100... Training loss: 0.0937\n",
      "Epoch: 47/100... Training loss: 0.0917\n",
      "Epoch: 48/100... Training loss: 0.0911\n",
      "Epoch: 48/100... Training loss: 0.0937\n",
      "Epoch: 48/100... Training loss: 0.0921\n",
      "Epoch: 49/100... Training loss: 0.0926\n",
      "Epoch: 49/100... Training loss: 0.0938\n",
      "Epoch: 49/100... Training loss: 0.0917\n",
      "Epoch: 50/100... Training loss: 0.0911\n",
      "Epoch: 50/100... Training loss: 0.0936\n",
      "Epoch: 50/100... Training loss: 0.0983\n",
      "Epoch: 51/100... Training loss: 0.0929\n",
      "Epoch: 51/100... Training loss: 0.0927\n",
      "Epoch: 51/100... Training loss: 0.0959\n",
      "Epoch: 52/100... Training loss: 0.0914\n",
      "Epoch: 52/100... Training loss: 0.0943\n",
      "Epoch: 52/100... Training loss: 0.0912\n",
      "Epoch: 53/100... Training loss: 0.0932\n",
      "Epoch: 53/100... Training loss: 0.0948\n",
      "Epoch: 53/100... Training loss: 0.0902\n",
      "Epoch: 54/100... Training loss: 0.0939\n",
      "Epoch: 54/100... Training loss: 0.0934\n",
      "Epoch: 54/100... Training loss: 0.0930\n",
      "Epoch: 55/100... Training loss: 0.0920\n",
      "Epoch: 55/100... Training loss: 0.0931\n",
      "Epoch: 55/100... Training loss: 0.0908\n",
      "Epoch: 56/100... Training loss: 0.0927\n",
      "Epoch: 56/100... Training loss: 0.0902\n",
      "Epoch: 56/100... Training loss: 0.0921\n",
      "Epoch: 57/100... Training loss: 0.0934\n",
      "Epoch: 57/100... Training loss: 0.0949\n",
      "Epoch: 57/100... Training loss: 0.0952\n",
      "Epoch: 58/100... Training loss: 0.0897\n",
      "Epoch: 58/100... Training loss: 0.0918\n",
      "Epoch: 58/100... Training loss: 0.0903\n",
      "Epoch: 59/100... Training loss: 0.0952\n",
      "Epoch: 59/100... Training loss: 0.0917\n",
      "Epoch: 59/100... Training loss: 0.0955\n",
      "Epoch: 60/100... Training loss: 0.0916\n",
      "Epoch: 60/100... Training loss: 0.0938\n",
      "Epoch: 60/100... Training loss: 0.0925\n",
      "Epoch: 61/100... Training loss: 0.0913\n",
      "Epoch: 61/100... Training loss: 0.0949\n",
      "Epoch: 61/100... Training loss: 0.0923\n",
      "Epoch: 62/100... Training loss: 0.0931\n",
      "Epoch: 62/100... Training loss: 0.0940\n",
      "Epoch: 62/100... Training loss: 0.0932\n",
      "Epoch: 63/100... Training loss: 0.0923\n",
      "Epoch: 63/100... Training loss: 0.0911\n",
      "Epoch: 63/100... Training loss: 0.0912\n",
      "Epoch: 64/100... Training loss: 0.0907\n",
      "Epoch: 64/100... Training loss: 0.0930\n",
      "Epoch: 64/100... Training loss: 0.0931\n",
      "Epoch: 65/100... Training loss: 0.0903\n",
      "Epoch: 65/100... Training loss: 0.0938\n",
      "Epoch: 65/100... Training loss: 0.0924\n",
      "Epoch: 66/100... Training loss: 0.0925\n",
      "Epoch: 66/100... Training loss: 0.0892\n",
      "Epoch: 66/100... Training loss: 0.0952\n",
      "Epoch: 67/100... Training loss: 0.0930\n",
      "Epoch: 67/100... Training loss: 0.0937\n",
      "Epoch: 67/100... Training loss: 0.0932\n",
      "Epoch: 68/100... Training loss: 0.0939\n",
      "Epoch: 68/100... Training loss: 0.0933\n",
      "Epoch: 68/100... Training loss: 0.0934\n",
      "Epoch: 69/100... Training loss: 0.0908\n",
      "Epoch: 69/100... Training loss: 0.0907\n",
      "Epoch: 69/100... Training loss: 0.0918\n",
      "Epoch: 70/100... Training loss: 0.0933\n",
      "Epoch: 70/100... Training loss: 0.0945\n",
      "Epoch: 70/100... Training loss: 0.0912\n",
      "Epoch: 71/100... Training loss: 0.0932\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 71/100... Training loss: 0.0934\n",
      "Epoch: 71/100... Training loss: 0.0931\n",
      "Epoch: 72/100... Training loss: 0.0933\n",
      "Epoch: 72/100... Training loss: 0.0934\n",
      "Epoch: 72/100... Training loss: 0.0925\n",
      "Epoch: 73/100... Training loss: 0.0913\n",
      "Epoch: 73/100... Training loss: 0.0910\n",
      "Epoch: 73/100... Training loss: 0.0909\n",
      "Epoch: 74/100... Training loss: 0.0922\n",
      "Epoch: 74/100... Training loss: 0.0942\n",
      "Epoch: 74/100... Training loss: 0.0932\n",
      "Epoch: 75/100... Training loss: 0.0923\n",
      "Epoch: 75/100... Training loss: 0.0909\n",
      "Epoch: 75/100... Training loss: 0.0909\n",
      "Epoch: 76/100... Training loss: 0.0923\n",
      "Epoch: 76/100... Training loss: 0.0917\n",
      "Epoch: 76/100... Training loss: 0.0952\n",
      "Epoch: 77/100... Training loss: 0.0929\n",
      "Epoch: 77/100... Training loss: 0.0917\n",
      "Epoch: 77/100... Training loss: 0.0921\n",
      "Epoch: 78/100... Training loss: 0.0906\n",
      "Epoch: 78/100... Training loss: 0.0901\n",
      "Epoch: 78/100... Training loss: 0.0927\n",
      "Epoch: 79/100... Training loss: 0.0919\n",
      "Epoch: 79/100... Training loss: 0.0908\n",
      "Epoch: 79/100... Training loss: 0.0903\n",
      "Epoch: 80/100... Training loss: 0.0916\n",
      "Epoch: 80/100... Training loss: 0.0959\n",
      "Epoch: 80/100... Training loss: 0.0932\n",
      "Epoch: 81/100... Training loss: 0.0898\n",
      "Epoch: 81/100... Training loss: 0.0940\n",
      "Epoch: 81/100... Training loss: 0.0936\n",
      "Epoch: 82/100... Training loss: 0.0946\n",
      "Epoch: 82/100... Training loss: 0.0937\n",
      "Epoch: 82/100... Training loss: 0.0931\n",
      "Epoch: 83/100... Training loss: 0.0939\n",
      "Epoch: 83/100... Training loss: 0.0920\n",
      "Epoch: 83/100... Training loss: 0.0944\n",
      "Epoch: 84/100... Training loss: 0.0901\n",
      "Epoch: 84/100... Training loss: 0.0928\n",
      "Epoch: 84/100... Training loss: 0.0902\n",
      "Epoch: 85/100... Training loss: 0.0903\n",
      "Epoch: 85/100... Training loss: 0.0935\n",
      "Epoch: 85/100... Training loss: 0.0912\n",
      "Epoch: 86/100... Training loss: 0.0909\n",
      "Epoch: 86/100... Training loss: 0.0914\n",
      "Epoch: 86/100... Training loss: 0.0934\n",
      "Epoch: 87/100... Training loss: 0.0923\n",
      "Epoch: 87/100... Training loss: 0.0890\n",
      "Epoch: 87/100... Training loss: 0.0930\n",
      "Epoch: 88/100... Training loss: 0.0930\n",
      "Epoch: 88/100... Training loss: 0.0900\n",
      "Epoch: 88/100... Training loss: 0.0934\n",
      "Epoch: 89/100... Training loss: 0.0898\n",
      "Epoch: 89/100... Training loss: 0.0914\n",
      "Epoch: 89/100... Training loss: 0.0940\n",
      "Epoch: 90/100... Training loss: 0.0945\n",
      "Epoch: 90/100... Training loss: 0.0897\n",
      "Epoch: 90/100... Training loss: 0.0901\n",
      "Epoch: 91/100... Training loss: 0.0903\n",
      "Epoch: 91/100... Training loss: 0.0911\n",
      "Epoch: 91/100... Training loss: 0.0916\n",
      "Epoch: 92/100... Training loss: 0.0927\n",
      "Epoch: 92/100... Training loss: 0.0937\n",
      "Epoch: 92/100... Training loss: 0.0905\n",
      "Epoch: 93/100... Training loss: 0.0915\n",
      "Epoch: 93/100... Training loss: 0.0911\n",
      "Epoch: 93/100... Training loss: 0.0882\n",
      "Epoch: 94/100... Training loss: 0.0923\n",
      "Epoch: 94/100... Training loss: 0.0911\n",
      "Epoch: 94/100... Training loss: 0.0921\n",
      "Epoch: 95/100... Training loss: 0.0934\n",
      "Epoch: 95/100... Training loss: 0.0928\n",
      "Epoch: 95/100... Training loss: 0.0907\n",
      "Epoch: 96/100... Training loss: 0.0922\n",
      "Epoch: 96/100... Training loss: 0.0924\n",
      "Epoch: 96/100... Training loss: 0.0906\n",
      "Epoch: 97/100... Training loss: 0.0886\n",
      "Epoch: 97/100... Training loss: 0.0936\n",
      "Epoch: 97/100... Training loss: 0.0930\n",
      "Epoch: 98/100... Training loss: 0.0909\n",
      "Epoch: 98/100... Training loss: 0.0924\n",
      "Epoch: 98/100... Training loss: 0.0920\n",
      "Epoch: 99/100... Training loss: 0.0931\n",
      "Epoch: 99/100... Training loss: 0.0918\n",
      "Epoch: 99/100... Training loss: 0.0933\n",
      "Epoch: 100/100... Training loss: 0.0926\n",
      "Epoch: 100/100... Training loss: 0.0912\n",
      "Epoch: 100/100... Training loss: 0.0926\n"
     ]
    }
   ],
   "source": [
    "epochs = 100\n",
    "batch_size = 200\n",
    "step_counter = 0\n",
    "# Set's how much noise we're adding to the MNIST images\n",
    "noise_factor = 0.5\n",
    "sess.run(tf.global_variables_initializer())\n",
    "for e in range(epochs):\n",
    "    for ii in range(mnist.train.num_examples//batch_size):\n",
    "        batch = mnist.train.next_batch(batch_size)\n",
    "        # Get images from the batch\n",
    "        imgs = batch[0].reshape((-1, 28, 28, 1))\n",
    "        \n",
    "        # Add random noise to the input images\n",
    "        noisy_imgs = imgs + noise_factor * np.random.randn(*imgs.shape)\n",
    "        # Clip the images to be between 0 and 1\n",
    "        noisy_imgs = np.clip(noisy_imgs, 0., 1.)\n",
    "        \n",
    "        # Noisy images as inputs, original images as targets\n",
    "        batch_cost, _ = sess.run([cost, opt], feed_dict={inputs_: noisy_imgs,\n",
    "                                                         targets_: imgs})\n",
    "\n",
    "        if step_counter%100 == 0:\n",
    "            print(\"Epoch: {}/{}...\".format(e+1, epochs),\n",
    "              \"Training loss: {:.4f}\".format(batch_cost))\n",
    "        step_counter+=1"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Checking out the performance\n",
    "\n",
    "Here I'm adding noise to the test images and passing them through the autoencoder. It does a suprisingly great job of removing the noise, even though it's sometimes difficult to tell what the original number is."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAABawAAAEsCAYAAAAvofT2AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nyhTqOLwAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f381ead69b0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, axes = plt.subplots(nrows=2, ncols=10, sharex=True, sharey=True, figsize=(20,4))\n",
    "in_imgs = mnist.test.images[:10]\n",
    "noisy_imgs = in_imgs + noise_factor * np.random.randn(*in_imgs.shape)\n",
    "noisy_imgs = np.clip(noisy_imgs, 0., 1.)\n",
    "\n",
    "reconstructed = sess.run(decoded, feed_dict={inputs_: noisy_imgs.reshape((10, 28, 28, 1))})\n",
    "\n",
    "for images, row in zip([noisy_imgs, reconstructed], axes):\n",
    "    for img, ax in zip(images, row):\n",
    "        ax.imshow(img.reshape((28, 28)), cmap='Greys_r')\n",
    "        ax.get_xaxis().set_visible(False)\n",
    "        ax.get_yaxis().set_visible(False)\n",
    "\n",
    "fig.tight_layout(pad=0.1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}