auto_write.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Application of Generative Adversarial Networks (GAN)\n",
    "## Building a neural network that learns to generate handwritten text.\n",
    "\n",
    "### Prerequisites\n",
    "Tensorflow, Matplotlib, Numpy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "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": [
    "import tensorflow as tf\n",
    "import numpy as np\n",
    "import datetime\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "\n",
    "from tensorflow.examples.tutorials.mnist import input_data\n",
    "mnist = input_data.read_data_sets(\"MNIST_data/\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(1, 784)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7f07cf5f0128>"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAADQZJREFUeJzt3W+oXPWdx/HPZ7Otgqkh2dwNwT97Y5AlMbDpOsaFylrR\nVCuFJKASCSFC3NsHEbbYByvug/WJGJZtS8G1km5DU+narrTBoNJtNmw0gaU4StaYZl3dcEsTYnKj\nhVhUqua7D+5RbuOd30zm35nb7/sFw50533PmfBnyyZk5v5nzc0QIQD5/VHcDAOpB+IGkCD+QFOEH\nkiL8QFKEH0iK8ANJEX4gKcIPJPXHw9zZ4sWLY3x8fJi7BFKZnJzUmTNn3Mm6PYXf9m2Svi1pnqR/\niYjtpfXHx8fVbDZ72SWAgkaj0fG6Xb/ttz1P0j9L+rKklZLutr2y2+cDMFy9fOZfI+mNiDgWEb+T\n9CNJ6/rTFoBB6yX8l0n69YzHx6tlv8f2hO2m7ebU1FQPuwPQTwM/2x8ROyKiERGNsbGxQe8OQId6\nCf8JSVfMeHx5tQzAHNBL+F+UdLXtZbY/K2mjpD39aQvAoHU91BcRH9q+T9K/a3qob2dEHOlbZwAG\nqqdx/oh4TtJzfeoFwBDx9V4gKcIPJEX4gaQIP5AU4QeSIvxAUoQfSIrwA0kRfiApwg8kRfiBpAg/\nkBThB5Ii/EBShB9IivADSRF+ICnCDyRF+IGkCD+QFOEHkiL8QFKEH0iK8ANJEX4gKcIPJEX4gaQI\nP5AU4QeS6mmWXtuTkt6R9JGkDyOi0Y+mAAxeT+Gv3BQRZ/rwPACGiLf9QFK9hj8k/dz2S7Yn+tEQ\ngOHo9W3/DRFxwvafStpr+38i4oWZK1T/KUxI0pVXXtnj7gD0S09H/og4Uf09LWm3pDWzrLMjIhoR\n0RgbG+tldwD6qOvw277E9uc+vi/pS5Je7VdjAAarl7f9SyTttv3x8/xrRPysL10BGLiuwx8RxyT9\nRR97ATBEDPUBSRF+ICnCDyRF+IGkCD+QFOEHkiL8QFKEH0iK8ANJEX4gKcIPJEX4gaQIP5AU4QeS\n6sfVe4GRdO7cuZa1iBhiJ582b968WvcvceQH0iL8QFKEH0iK8ANJEX4gKcIPJEX4gaQY50dt3n//\n/WJ9//79xfrBgweL9aNHj7as7d69u7htrxYsWFCsb9u2rWVtxYoVxW03bdrUVU/n48gPJEX4gaQI\nP5AU4QeSIvxAUoQfSIrwA0m1Hee3vVPSVySdjohV1bJFkn4saVzSpKS7IuI3g2sTdTl79myxXhpL\nl6RnnnmmZe3ZZ58tbnvo0KFivRf33HNPsd7u9/arV68u1m+++eYLbekTy5cv73rbC9HJkf/7km47\nb9kDkvZFxNWS9lWPAcwhbcMfES9Ievu8xesk7aru75K0vs99ARiwbj/zL4mIk9X9NyUt6VM/AIak\n5xN+MX0xtJYXRLM9Ybtpuzk1NdXr7gD0SbfhP2V7qSRVf0+3WjEidkREIyIaY2NjXe4OQL91G/49\nkrZU97dIero/7QAYlrbht/2kpP+S9Oe2j9veKmm7pLW2X5d0S/UYwBzSdpw/Iu5uUep+IPMPTLtr\nwB84cKBYP3bsWE/7L/3++/rrry9uu2/fvmJ948aNxXq77wGUxqzvvffe4rYTExPFei+/a58/f36x\nbrvr554r+IYfkBThB5Ii/EBShB9IivADSRF+ICku3d0Hjz/+eLFeukyzJF1++eXF+iOPPFKsv/vu\nu8V6Sbshr5UrVxbrO3fuLNaH9fNUXDiO/EBShB9IivADSRF+ICnCDyRF+IGkCD+QFOP8ffDYY4/1\ntH277wH0a0rm2bT7ye/zzz8/sH2jXhz5gaQIP5AU4QeSIvxAUoQfSIrwA0kRfiApxvn74NFHHy3W\n165dW6w//PDDxfqdd95ZrF911VXFOjAbjvxAUoQfSIrwA0kRfiApwg8kRfiBpAg/kFTbcX7bOyV9\nRdLpiFhVLXtI0t9ImqpWezAinhtUk6PuxhtvLNb37NlTrG/YsKFYv+6664r1zZs3t6xt3769uO3F\nF19crOMPVydH/u9Lum2W5d+KiNXVLW3wgbmqbfgj4gVJbw+hFwBD1Mtn/vtsv2J7p+2FfesIwFB0\nG/7vSFouabWkk5K+0WpF2xO2m7abU1NTrVYDMGRdhT8iTkXERxFxTtJ3Ja0prLsjIhoR0RgbG+u2\nTwB91lX4bS+d8XCDpFf70w6AYelkqO9JSV+UtNj2cUn/IOmLtldLCkmTkr46wB4BDIAjYmg7azQa\n0Ww2h7a/uWLv3r3F+q233tr1c990003F+lNPPVWsL1q0qOt9Y/gajYaazaY7WZdv+AFJEX4gKcIP\nJEX4gaQIP5AU4QeS4tLdI+CWW24p1t96661iff369S1r7abYvuOOO4r1dsOQ8+bNK9YxujjyA0kR\nfiApwg8kRfiBpAg/kBThB5Ii/EBSjPOPALv8C8yFC8uXSCyN5Y+Pjxe33b9/f7E+OTlZrC9fvrxY\nx+jiyA8kRfiBpAg/kBThB5Ii/EBShB9IivADSRF+ICnCDyRF+IGkCD+QFOEHkiL8QFKEH0iK8ANJ\ntf09v+0rJP1A0hJJIWlHRHzb9iJJP5Y0LmlS0l0R8ZvBtTp3ffDBB8X6kSNHivXXXnutWN+6dWvL\n2nvvvVfcdtWqVcX62NhYsY65q5Mj/4eSvh4RKyX9laRttldKekDSvoi4WtK+6jGAOaJt+CPiZES8\nXN1/R9JRSZdJWidpV7XaLkmtp40BMHIu6DO/7XFJn5f0C0lLIuJkVXpT0x8LAMwRHYff9nxJP5H0\ntYg4O7MWEaHp8wGzbTdhu2m7OTU11VOzAPqno/Db/oymg//DiPhptfiU7aVVfamk07NtGxE7IqIR\nEQ1OHgGjo234PX1p2e9JOhoR35xR2iNpS3V/i6Sn+98egEHp5NLdX5C0WdJh24eqZQ9K2i7p32xv\nlfQrSXcNpsW57/Dhw8X6tddeW6y3u7R3yYoVK4r1AwcOFOuXXnpp1/vGaGsb/og4KKnVv76b+9sO\ngGHhG35AUoQfSIrwA0kRfiApwg8kRfiBpJiiewiuueaaYn3Tpk3F+rJly7re9/3331+sL1iwoOvn\nxtzGkR9IivADSRF+ICnCDyRF+IGkCD+QFOEHkmKcfwguuuiiYv2JJ54o1nv5PT/QCkd+ICnCDyRF\n+IGkCD+QFOEHkiL8QFKEH0iKcf4RwDg+6sCRH0iK8ANJEX4gKcIPJEX4gaQIP5AU4QeSaht+21fY\n/k/bv7R9xPbfVssfsn3C9qHqdvvg2wXQL518yedDSV+PiJdtf07SS7b3VrVvRcQ/Da49AIPSNvwR\ncVLSyer+O7aPSrps0I0BGKwL+sxve1zS5yX9olp0n+1XbO+0vbDFNhO2m7abU1NTPTULoH86Dr/t\n+ZJ+IulrEXFW0nckLZe0WtPvDL4x23YRsSMiGhHRGBsb60PLAPqho/Db/oymg//DiPipJEXEqYj4\nKCLOSfqupDWDaxNAv3Vytt+SvifpaER8c8bypTNW2yDp1f63B2BQOjnb/wVJmyUdtn2oWvagpLtt\nr5YUkiYlfXUgHQIYiE7O9h+UNNsPzp/rfzsAhoVv+AFJEX4gKcIPJEX4gaQIP5AU4QeSIvxAUoQf\nSIrwA0kRfiApwg8kRfiBpAg/kBThB5JyRAxvZ/aUpF/NWLRY0pmhNXBhRrW3Ue1Lordu9bO3P4uI\njq6XN9Twf2rndjMiGrU1UDCqvY1qXxK9dauu3njbDyRF+IGk6g7/jpr3XzKqvY1qXxK9dauW3mr9\nzA+gPnUf+QHUpJbw277N9mu237D9QB09tGJ70vbhaubhZs297LR92varM5Ytsr3X9uvV31mnSaup\nt5GYubkws3Str92ozXg99Lf9tudJ+l9JayUdl/SipLsj4pdDbaQF25OSGhFR+5iw7b+W9FtJP4iI\nVdWyf5T0dkRsr/7jXBgRfzcivT0k6bd1z9xcTSizdObM0pLWS7pHNb52hb7uUg2vWx1H/jWS3oiI\nYxHxO0k/krSuhj5GXkS8IOnt8xavk7Srur9L0/94hq5FbyMhIk5GxMvV/XckfTyzdK2vXaGvWtQR\n/ssk/XrG4+MarSm/Q9LPbb9ke6LuZmaxpJo2XZLelLSkzmZm0Xbm5mE6b2bpkXntupnxut844fdp\nN0TEX0r6sqRt1dvbkRTTn9lGabimo5mbh2WWmaU/Uedr1+2M1/1WR/hPSLpixuPLq2UjISJOVH9P\nS9qt0Zt9+NTHk6RWf0/X3M8nRmnm5tlmltYIvHajNON1HeF/UdLVtpfZ/qykjZL21NDHp9i+pDoR\nI9uXSPqSRm/24T2StlT3t0h6usZefs+ozNzcamZp1fzajdyM1xEx9Juk2zV9xv//JP19HT206Osq\nSf9d3Y7U3ZukJzX9NvADTZ8b2SrpTyTtk/S6pP+QtGiEentC0mFJr2g6aEtr6u0GTb+lf0XSoep2\ne92vXaGvWl43vuEHJMUJPyApwg8kRfiBpAg/kBThB5Ii/EBShB9IivADSf0//Z/0d8kENG0AAAAA\nSUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f07da361ef0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "sample_image = mnist.train.next_batch(1)[0]\n",
    "print(sample_image.shape)\n",
    "\n",
    "sample_image = sample_image.reshape([28, 28])\n",
    "plt.imshow(sample_image, cmap='Greys')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "## The Discriminator\n",
    "Now, starting with a simple binary classifier and then moving on."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def discriminator(images, reuse_variables=None):\n",
    "    with tf.variable_scope(tf.get_variable_scope(), reuse=reuse_variables) as scope:\n",
    "        # First convolutional and pool layers\n",
    "        # This finds 32 different 5 x 5 pixel features\n",
    "        d_w1 = tf.get_variable('d_w1', [5, 5, 1, 32], initializer=tf.truncated_normal_initializer(stddev=0.02))\n",
    "        d_b1 = tf.get_variable('d_b1', [32], initializer=tf.constant_initializer(0))\n",
    "        d1 = tf.nn.conv2d(input=images, filter=d_w1, strides=[1, 1, 1, 1], padding='SAME')\n",
    "        d1 = d1 + d_b1\n",
    "        d1 = tf.nn.relu(d1)\n",
    "        d1 = tf.nn.avg_pool(d1, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')\n",
    "\n",
    "        # Second convolutional and pool layers\n",
    "        # This finds 64 different 5 x 5 pixel features\n",
    "        d_w2 = tf.get_variable('d_w2', [5, 5, 32, 64], initializer=tf.truncated_normal_initializer(stddev=0.02))\n",
    "        d_b2 = tf.get_variable('d_b2', [64], initializer=tf.constant_initializer(0))\n",
    "        d2 = tf.nn.conv2d(input=d1, filter=d_w2, strides=[1, 1, 1, 1], padding='SAME')\n",
    "        d2 = d2 + d_b2\n",
    "        d2 = tf.nn.relu(d2)\n",
    "        d2 = tf.nn.avg_pool(d2, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')\n",
    "\n",
    "        # First fully connected layer\n",
    "        d_w3 = tf.get_variable('d_w3', [7 * 7 * 64, 1024], initializer=tf.truncated_normal_initializer(stddev=0.02))\n",
    "        d_b3 = tf.get_variable('d_b3', [1024], initializer=tf.constant_initializer(0))\n",
    "        d3 = tf.reshape(d2, [-1, 7 * 7 * 64])\n",
    "        d3 = tf.matmul(d3, d_w3)\n",
    "        d3 = d3 + d_b3\n",
    "        d3 = tf.nn.relu(d3)\n",
    "\n",
    "        # Second fully connected layer\n",
    "        d_w4 = tf.get_variable('d_w4', [1024, 1], initializer=tf.truncated_normal_initializer(stddev=0.02))\n",
    "        d_b4 = tf.get_variable('d_b4', [1], initializer=tf.constant_initializer(0))\n",
    "        d4 = tf.matmul(d3, d_w4) + d_b4\n",
    "\n",
    "        # d4 contains unscaled values\n",
    "        return d4"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# The Generator\n",
    "In the generator here, we are using three convolutional layers along with a interpolation until we get images of 28x28 pixel."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def generator(z, batch_size, z_dim):\n",
    "    g_w1 = tf.get_variable('g_w1', [z_dim, 3136], dtype=tf.float32, initializer=tf.truncated_normal_initializer(stddev=0.02))\n",
    "    g_b1 = tf.get_variable('g_b1', [3136], initializer=tf.truncated_normal_initializer(stddev=0.02))\n",
    "    g1 = tf.matmul(z, g_w1) + g_b1\n",
    "    g1 = tf.reshape(g1, [-1, 56, 56, 1])\n",
    "    g1 = tf.contrib.layers.batch_norm(g1, epsilon=1e-5, scope='bn1')\n",
    "    g1 = tf.nn.relu(g1)\n",
    "\n",
    "    # Generate 50 features\n",
    "    g_w2 = tf.get_variable('g_w2', [3, 3, 1, z_dim/2], dtype=tf.float32, initializer=tf.truncated_normal_initializer(stddev=0.02))\n",
    "    g_b2 = tf.get_variable('g_b2', [z_dim/2], initializer=tf.truncated_normal_initializer(stddev=0.02))\n",
    "    g2 = tf.nn.conv2d(g1, g_w2, strides=[1, 2, 2, 1], padding='SAME')\n",
    "    g2 = g2 + g_b2\n",
    "    g2 = tf.contrib.layers.batch_norm(g2, epsilon=1e-5, scope='bn2')\n",
    "    g2 = tf.nn.relu(g2)\n",
    "    g2 = tf.image.resize_images(g2, [56, 56])\n",
    "\n",
    "    # Generate 25 features\n",
    "    g_w3 = tf.get_variable('g_w3', [3, 3, z_dim/2, z_dim/4], dtype=tf.float32, initializer=tf.truncated_normal_initializer(stddev=0.02))\n",
    "    g_b3 = tf.get_variable('g_b3', [z_dim/4], initializer=tf.truncated_normal_initializer(stddev=0.02))\n",
    "    g3 = tf.nn.conv2d(g2, g_w3, strides=[1, 2, 2, 1], padding='SAME')\n",
    "    g3 = g3 + g_b3\n",
    "    g3 = tf.contrib.layers.batch_norm(g3, epsilon=1e-5, scope='bn3')\n",
    "    g3 = tf.nn.relu(g3)\n",
    "    g3 = tf.image.resize_images(g3, [56, 56])\n",
    "\n",
    "    # Final convolution with one output channel\n",
    "    g_w4 = tf.get_variable('g_w4', [1, 1, z_dim/4, 1], dtype=tf.float32, initializer=tf.truncated_normal_initializer(stddev=0.02))\n",
    "    g_b4 = tf.get_variable('g_b4', [1], initializer=tf.truncated_normal_initializer(stddev=0.02))\n",
    "    g4 = tf.nn.conv2d(g3, g_w4, strides=[1, 2, 2, 1], padding='SAME')\n",
    "    g4 = g4 + g_b4\n",
    "    g4 = tf.sigmoid(g4)\n",
    "    \n",
    "    # Dimensions of g4: batch_size x 28 x 28 x 1\n",
    "    return g4"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Generating a sample image from untrained generator\n",
    "\n",
    "Now both the generator and discriminator functions are done. Lets see what the untarined classifier produces as a result now."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "z_dimensions = 100\n",
    "z_placeholder = tf.placeholder(tf.float32, [None, z_dimensions])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "generated_image_output = generator(z_placeholder, 1, z_dimensions)\n",
    "z_batch = np.random.normal(0, 1, [1, z_dimensions])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nUWp+kUSp+UUSpeYXSZSaXyRRan6RRP0/uiR8wVcrDiIAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f07cf5b8c50>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "with tf.Session() as sess:\n",
    "    sess.run(tf.global_variables_initializer())\n",
    "    generated_image = sess.run(generated_image_output,\n",
    "                                feed_dict={z_placeholder: z_batch})\n",
    "    generated_image = generated_image.reshape([28, 28])\n",
    "    plt.imshow(generated_image, cmap='Greys')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Noise, noise, noise! Obviously, what else to expect from an untrained generator. Let's begin training our generator network to convert random numbers into recognizable digits."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Training the GAN\n",
    "\n",
    "Here comes my favourite part about GAN's. GAN's have two loss functions: one that encorages the generator to make better and better images and the other that encourages the discriminator to distinguish generated images form the real images.\n",
    "\n",
    "Thus both the generator and the discriminator are being traned simultaneously. Now as the disciminator gets better and better at distinguishing the real images form the generated images, the generator is able to better tune its weights and biases to create more and more convincing images of handwritten text."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "tf.reset_default_graph()\n",
    "batch_size = 50\n",
    "\n",
    "z_placeholder = tf.placeholder(tf.float32, [None, z_dimensions], name='z_placeholder') \n",
    "# z_placeholder is for feeding input noise to the generator\n",
    "\n",
    "x_placeholder = tf.placeholder(tf.float32, shape = [None,28,28,1], name='x_placeholder') \n",
    "# x_placeholder is for feeding input images to the discriminator\n",
    "\n",
    "Gz = generator(z_placeholder, batch_size, z_dimensions) \n",
    "# Gz holds the generated images\n",
    "\n",
    "Dx = discriminator(x_placeholder) \n",
    "# Dx will hold discriminator prediction probabilities\n",
    "# for the real MNIST images\n",
    "\n",
    "Dg = discriminator(Gz, reuse_variables=True)\n",
    "# Dg will hold discriminator prediction probabilities for generated images"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Calculating the cross-entropy losses we have."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "d_loss_real = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits = Dx, labels = tf.ones_like(Dx)))\n",
    "d_loss_fake = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits = Dg, labels = tf.zeros_like(Dg)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "g_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits = Dg, labels = tf.ones_like(Dg))) "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Defining the optimizers. First only for the generators holding the discriminators weights fixed and then the other way round."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['d_w1:0', 'd_b1:0', 'd_w2:0', 'd_b2:0', 'd_w3:0', 'd_b3:0', 'd_w4:0', 'd_b4:0']\n",
      "['g_w1:0', 'g_b1:0', 'g_w2:0', 'g_b2:0', 'g_w3:0', 'g_b3:0', 'g_w4:0', 'g_b4:0']\n"
     ]
    }
   ],
   "source": [
    "tvars = tf.trainable_variables()\n",
    "\n",
    "d_vars = [var for var in tvars if 'd_' in var.name]\n",
    "g_vars = [var for var in tvars if 'g_' in var.name]\n",
    "\n",
    "print([v.name for v in d_vars])\n",
    "print([v.name for v in g_vars])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Using the [Adam](https://www.tensorflow.org/api_docs/python/tf/train/AdamOptimizer) Optimizer for training the generator and the discriminator. The discriminator is trained on two different operations: one one the real images and one on the fake images."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Train the discriminator\n",
    "d_trainer_fake = tf.train.AdamOptimizer(0.0003).minimize(d_loss_fake, var_list=d_vars)\n",
    "d_trainer_real = tf.train.AdamOptimizer(0.0003).minimize(d_loss_real, var_list=d_vars)\n",
    "\n",
    "# Train the generator\n",
    "g_trainer = tf.train.AdamOptimizer(0.0001).minimize(g_loss, var_list=g_vars)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As GAN's take a long time to train, we can view their progress on Tensorboard."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# From this point forward, reuse variables\n",
    "tf.get_variable_scope().reuse_variables()\n",
    "\n",
    "tf.summary.scalar('Generator_loss', g_loss)\n",
    "tf.summary.scalar('Discriminator_loss_real', d_loss_real)\n",
    "tf.summary.scalar('Discriminator_loss_fake', d_loss_fake)\n",
    "\n",
    "images_for_tensorboard = generator(z_placeholder, batch_size, z_dimensions)\n",
    "tf.summary.image('Generated_images', images_for_tensorboard, 5)\n",
    "merged = tf.summary.merge_all()\n",
    "logdir = \"tensorboard/\" + datetime.datetime.now().strftime(\"%Y%m%d-%H%M%S\") + \"/\"\n",
    "writer = tf.summary.FileWriter(logdir, sess.graph)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Now, we have two options here. Either to train the model for about three hours on a fast GPU, or for around a day on a desktop CPU. And the other option is we can load the pretrained model, and can experiment with the output of a trained GAN.\n",
    "Here, I am going with the second one as time is precious !"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Restoring parameters from pretrained-model/pretrained_gan.ckpt\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f07ba165978>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nKMIPBEX4gaAIPxAU4QeC+n+W4DZ0+cST3QAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f07ba76cd68>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f07ba1a3390>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nA0ERfiAowg8ERfiBoAg/ENT/A71jcDbhWun1AAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f07ba22fa20>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nHwjq/wFXQzboXk0ePgAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f07ba1a8978>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n3XQAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f07ba2d4518>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nDwT1Xwo0LcDoUqCiAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f07b9944c88>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAADkNJREFUeJzt3X+IXPW5x/HPk80mgW1E7e5NFhvvxqqFINxUhiA0Si+9\nLVYCsSgS/6gxxqRIlFtJQPEiN39FkduW/CGF7W1IvPbmB7RiEKmN8UooXosTyfVH1t9szY81mZhi\nUwLGbJ77x56UNe58ZzJzZs7MPu8XLDtznnPmPJzks2dmvjPna+4uAPHMKLoBAMUg/EBQhB8IivAD\nQRF+ICjCDwRF+IGgCD8QFOEHgprZzp319/f70NBQO3cJhDI6OqoTJ05YPes2FX4zu1nSZkk9kv7T\n3R9PrT80NKRyudzMLgEklEqlutdt+Gm/mfVIelLSDyUtknSnmS1q9PEAtFczr/mXSPrA3T9y9zOS\ndkhank9bAFqtmfBfIenQpPuHs2VfYmZrzaxsZuVKpdLE7gDkqeXv9rv7sLuX3L00MDDQ6t0BqFMz\n4T8iacGk+9/IlgHoAs2E/zVJ15jZQjObJWmFpN35tAWg1Roe6nP3s2Z2v6QXNDHUt8Xd386tMwAt\n1dQ4v7s/L+n5nHoB0EZ8vBcIivADQRF+ICjCDwRF+IGgCD8QFOEHgiL8QFCEHwiK8ANBEX4gKMIP\nBEX4gaAIPxAU4QeCIvxAUIQfCIrwA0ERfiAowg8ERfiBoAg/EBThB4Ii/EBQhB8IivADQRF+ICjC\nDwRF+IGgmpql18xGJZ2SNC7prLuX8mgKnWN8fDxZ37FjR7K+f//+qrUXX3wxue27776brN92223J\n+tNPP121NmMG572mwp/5Z3c/kcPjAGgj/vwBQTUbfpf0BzPbb2Zr82gIQHs0+7R/qbsfMbN/kLTH\nzN5x932TV8j+KKyVpCuvvLLJ3QHIS1Nnfnc/kv0+LukZSUumWGfY3UvuXhoYGGhmdwBy1HD4zazP\nzOaevy3pB5LeyqsxAK3VzNP+eZKeMbPzj/Pf7v77XLoC0HINh9/dP5L0Tzn2ggLs27cvWb/rrruS\n9Y8//jhZd/eL7qle27dvT9ZTn1HYuXNn3u10HYb6gKAIPxAU4QeCIvxAUIQfCIrwA0Hl8a0+dLCj\nR48m6ytWrEjWx8bGkvW+vr5kfdmyZVVrc+bMSW7b39+frG/dujVZ37VrV9XaAw88kNx26dKlyfp0\nwJkfCIrwA0ERfiAowg8ERfiBoAg/EBThB4JinL8L1Lp8duoS1evXr09u++mnnybrd999d7I+PDyc\nrPf29ibrzdi0aVOyPnv27Kq1vXv3JrdlnB/AtEX4gaAIPxAU4QeCIvxAUIQfCIrwA0Exzt8FRkZG\nkvVaY/Eptaa53rJlS7KezdtQiM8++yxZnzVrVtXa7bffnnc7XYczPxAU4QeCIvxAUIQfCIrwA0ER\nfiAowg8EVXOc38y2SFom6bi7X5ctu1zSTklDkkYl3eHuf2ldm7G98MILDW+7cePGZP2hhx5K1osc\nx3/ssceS9SeffDJZP3PmTNXaq6++mtx20aJFyXqRxyUv9Zz5t0q6+YJlD0va6+7XSNqb3QfQRWqG\n3933STp5weLlkrZlt7dJujXnvgC0WKOv+ee5+/l5nD6RNC+nfgC0SdNv+Lm7S/JqdTNba2ZlMytX\nKpVmdwcgJ42G/5iZDUpS9vt4tRXdfdjdS+5eGhgYaHB3APLWaPh3S1qZ3V4p6dl82gHQLjXDb2bb\nJf2vpG+Z2WEzWy3pcUnfN7P3Jf1Ldh9AF6k5zu/ud1YpfS/nXlBFM+P8tcar58yZ0/BjS9LRo0eT\n9Xvvvbdq7b333ktu++GHHzbUUz1OnDiRrE+Hcfxa+IQfEBThB4Ii/EBQhB8IivADQRF+ICgu3d0F\n1qxZk6zv2bOnam3VqlXJbQ8ePJis33PPPcn64sWLk/WTJy/8Tlj7DA0NVa2tW7eufY1cpIlPzFeX\n1zAkZ34gKMIPBEX4gaAIPxAU4QeCIvxAUIQfCMpqjSnmqVQqeblcbtv+potaY+Xz58+vWvviiy/y\nbic3vb29yXqt3mfMSJ+7UlObX3vttcltu1WpVFK5XK7rgwCc+YGgCD8QFOEHgiL8QFCEHwiK8ANB\nEX4gKL7P3wVmz56drPf09FStFT3O39fXV7VWawan0dHRZH3Tpk3J+nQdy88LZ34gKMIPBEX4gaAI\nPxAU4QeCIvxAUIQfCKrmOL+ZbZG0TNJxd78uW7ZR0hpJlWy1R9z9+VY1GV1qrFySbrjhhqq1l19+\nOeduvqzWFN+pKbo3b96c3Paqq65K1u+7775kHWn1nPm3Srp5iuW/cPfF2Q/BB7pMzfC7+z5JxU27\nAqAlmnnNf7+ZvWFmW8zsstw6AtAWjYb/l5K+KWmxpDFJP6u2opmtNbOymZUrlUq11QC0WUPhd/dj\n7j7u7uck/UrSksS6w+5ecvdSrS9yAGifhsJvZoOT7v5I0lv5tAOgXeoZ6tsu6buS+s3ssKR/l/Rd\nM1ssySWNSvpJC3sE0AJct38aOHXqVNXaJZdckty21rXza9VPnz6drKc88cQTyfqGDRuS9bzmqZ9O\nuG4/gJoIPxAU4QeCIvxAUIQfCIrwA0F11aW7U5ehrjUkNZ3NnTu3am3WrFnJbWfOTP8XqDUNdi2p\ny4qvXr06uS1Dea3FmR8IivADQRF+ICjCDwRF+IGgCD8QFOEHguqqcf5aY9L4qs8//zxZX7hwYbJe\na5rsWlKfQah12W+0Fmd+ICjCDwRF+IGgCD8QFOEHgiL8QFCEHwiqqwbO+X53/kqlUrLe7Dh/apYm\nxvmLxZkfCIrwA0ERfiAowg8ERfiBoAg/EBThB4KqOc5vZgskPSVpniSXNOzum83sckk7JQ1JGpV0\nh7v/pXWtohVeeumllj7+qlWrqtaanRMAzann6J+VtN7dF0m6QdI6M1sk6WFJe939Gkl7s/sAukTN\n8Lv7mLu/nt0+JWlE0hWSlkvalq22TdKtrWoSQP4u6nmXmQ1J+rakP0ma5+5jWekTTbwsANAl6g6/\nmX1N0m8l/dTd/zq55u6uifcDptpurZmVzaxcqVSaahZAfuoKv5n1aiL4v3H332WLj5nZYFYflHR8\nqm3dfdjdS+5eSn3JA0B71Qy/TXyV7teSRtz955NKuyWtzG6vlPRs/u0BaJV6vtL7HUk/lvSmmR3I\nlj0i6XFJu8xstaQ/S7qjNS12v3PnziXrRQ559fX1JesnT55M1m+88cZkfcOGDRfdE9qjZvjd/Y+S\nqn2R/nv5tgOgXfiUBRAU4QeCIvxAUIQfCIrwA0ERfiCorrp0d5HGx8er1vr7+5Pb3nTTTcn61Vdf\nnawPDg4m6wcOHKhae+WVV5LbHjp0KFmvZdeuXcl6b29vU4+P1uHMDwRF+IGgCD8QFOEHgiL8QFCE\nHwiK8ANBMc5fp56enqq1hQsXJrd97rnnGn5sSTp79myyPnEVtcZceumlyfrIyEiyPn/+/Ib3jWJx\n5geCIvxAUIQfCIrwA0ERfiAowg8ERfiBoBjnz8E777yTrM+dOzdZf/TRR5P1Bx98MFmfObP6P2O5\nXE5ue/311yfrmL448wNBEX4gKMIPBEX4gaAIPxAU4QeCIvxAUDXH+c1sgaSnJM2T5JKG3X2zmW2U\ntEZSJVv1EXd/vlWNdrLTp08Xuv9z584Vun90p3o+5HNW0np3f93M5krab2Z7stov3P0/WtcegFap\nGX53H5M0lt0+ZWYjkq5odWMAWuuiXvOb2ZCkb0v6U7bofjN7w8y2mNllVbZZa2ZlMytXKpWpVgFQ\ngLrDb2Zfk/RbST91979K+qWkb0parIlnBj+bajt3H3b3kruXBgYGcmgZQB7qCr+Z9Woi+L9x999J\nkrsfc/dxdz8n6VeSlrSuTQB5qxl+MzNJv5Y04u4/n7R88tSxP5L0Vv7tAWiVet7t/46kH0t608zO\nzwX9iKQ7zWyxJob/RiX9pCUdAmiJet7t/6Mkm6IUckwfmC74hB8QFOEHgiL8QFCEHwiK8ANBEX4g\nKMIPBEX4gaAIPxAU4QeCIvxAUIQfCIrwA0ERfiAoc/f27cysIunPkxb1SzrRtgYuTqf21ql9SfTW\nqDx7+0d3r+t6eW0N/1d2blZ291JhDSR0am+d2pdEb40qqjee9gNBEX4gqKLDP1zw/lM6tbdO7Uui\nt0YV0luhr/kBFKfoMz+AghQSfjO72czeNbMPzOzhInqoxsxGzexNMztgZuWCe9liZsfN7K1Jyy43\nsz1m9n72e8pp0grqbaOZHcmO3QEzu6Wg3haY2f+Y2UEze9vM/jVbXuixS/RVyHFr+9N+M+uR9J6k\n70s6LOk1SXe6+8G2NlKFmY1KKrl74WPCZnaTpL9Jesrdr8uWPSHppLs/nv3hvMzdH+qQ3jZK+lvR\nMzdnE8oMTp5ZWtKtku5Wgccu0dcdKuC4FXHmXyLpA3f/yN3PSNohaXkBfXQ8d98n6eQFi5dL2pbd\n3qaJ/zxtV6W3juDuY+7+enb7lKTzM0sXeuwSfRWiiPBfIenQpPuH1VlTfrukP5jZfjNbW3QzU5iX\nTZsuSZ9ImldkM1OoOXNzO10ws3THHLtGZrzOG2/4fdVSd79e0g8lrcue3nYkn3jN1knDNXXN3Nwu\nU8ws/XdFHrtGZ7zOWxHhPyJpwaT738iWdQR3P5L9Pi7pGXXe7MPHzk+Smv0+XnA/f9dJMzdPNbO0\nOuDYddKM10WE/zVJ15jZQjObJWmFpN0F9PEVZtaXvREjM+uT9AN13uzDuyWtzG6vlPRsgb18SafM\n3FxtZmkVfOw6bsZrd2/7j6RbNPGO/4eS/q2IHqr0dZWk/8t+3i66N0nbNfE08AtNvDeyWtLXJe2V\n9L6kFyVd3kG9/ZekNyW9oYmgDRbU21JNPKV/Q9KB7OeWoo9doq9Cjhuf8AOC4g0/ICjCDwRF+IGg\nCD8QFOEHgiL8QFCEHwiK8ANB/T8M3VkHtZqaugAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f07ba6e7c50>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f07ba462160>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f07ba2dbb70>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "saver = tf.train.Saver()\n",
    "with tf.Session() as sess:\n",
    "    saver.restore(sess, 'pretrained-model/pretrained_gan.ckpt')\n",
    "    z_batch = np.random.normal(0, 1, size=[10, z_dimensions])\n",
    "    z_placeholder = tf.placeholder(tf.float32, [None, z_dimensions], name='z_placeholder') \n",
    "    generated_images = generator(z_placeholder, 10, z_dimensions)\n",
    "    images = sess.run(generated_images, {z_placeholder: z_batch})\n",
    "    for i in range(10):\n",
    "        plt.imshow(images[i].reshape([28, 28]), cmap='Greys')\n",
    "        plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "And here we have our trained GAN output. Although it's a little blurred but the generated images are very close to the handwritten digits. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "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": "3.5.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}