tf_conv_np.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "from tensorflow.examples.tutorials.mnist import input_data\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Extracting ../MNIST/MNIST_data/train-images-idx3-ubyte.gz\n",
      "Extracting ../MNIST/MNIST_data/train-labels-idx1-ubyte.gz\n",
      "Extracting ../MNIST/MNIST_data/t10k-images-idx3-ubyte.gz\n",
      "Extracting ../MNIST/MNIST_data/t10k-labels-idx1-ubyte.gz\n"
     ]
    }
   ],
   "source": [
    "mnist = input_data.read_data_sets(\"../MNIST/MNIST_data/\", one_hot=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0xf0c6d30>"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAO4AAADuCAYAAAA+7jsiAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nqXsl2qE1AAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0xf0c6cf8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# get a random image from the MNIST data\n",
    "id = np.random.randint(1000)\n",
    "image = mnist.train.images[id:id+1]\n",
    "\n",
    "# display the image\n",
    "plt.tick_params( axis='both', which='both', bottom='off', left='off', labelbottom='off', labelleft='off' )\n",
    "plt.imshow(image.reshape(28,28), cmap='gray')\n",
    "# append '_r' to cmap for reversed color map"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# we know that all the image are 28 by 28\n",
    "# this is the input placeholder to provide data to our network\n",
    "x = tf.placeholder(tf.float32, [None,28*28])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# create a session\n",
    "sess = tf.InteractiveSession()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Reshape the image to use in a convolutional network\n",
    "xshaped = tf.reshape(x, [-1,28,28,1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1, 28, 28, 1)"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# this is reshaped image\n",
    "out = sess.run(xshaped, feed_dict={x:image}) \n",
    "out.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# how many features do we want to extract from the image\n",
    "# each feature is an image with a filter applied to it\n",
    "features = 5\n",
    "# weights\n",
    "W = tf.Variable( tf.truncated_normal([5,5,1,features], stddev=0.1))\n",
    "# biases\n",
    "b = tf.Variable( tf.constant(0.1, shape=[features]) )\n",
    "# first 2 dimensions are patch size (5x5)\n",
    "#   filters are applied to a section of the image (denoted by the patch size)\n",
    "#   the section is then moved in the image by some value (denoted by a stride)\n",
    "# next is number of input channels (we are only using 1 'greyscale')\n",
    "#   this could also be the number of features coming from previous convolutional layer\n",
    "# next is number of output channels or features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# initialize the variables\n",
    "sess.run(tf.global_variables_initializer())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# create a convolutional layer\n",
    "# stride denotes how patch (defined above) is moved in the image\n",
    "# truncate the bottom of the output (<0 = 0) with relu (rectified linear unit)\n",
    "conv1 = tf.nn.relu(tf.nn.conv2d(xshaped, W, strides=[1,1,1,1], padding='SAME') + b)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.0 0.97853\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(1, 28, 28, 5)"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "out = sess.run(conv1, feed_dict={x:image}) # this is convoluted image\n",
    "print( out.min(), out.max() )\n",
    "out.shape\n",
    "# last dimension is the number of output channels in W\n",
    "# (1,image width, image height, output channels / features)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAyIAAACcCAYAAACDZR/GAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0xf29d780>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Let see how all the features look like ...\n",
    "# split the array to images with different features\n",
    "img = np.dsplit(out[0], features)\n",
    "width = out.shape[1]\n",
    "height = out.shape[2]\n",
    "# display the different features as images\n",
    "plt.figure( figsize=( 14,10 ) )\n",
    "for i in range( features ):\n",
    "    plt.subplot(1, 5, i+1)\n",
    "    plt.tick_params( axis='both', which='both', bottom='off', left='off', labelbottom='off', labelleft='off' )\n",
    "    plt.imshow(img[i].reshape(width, height), cmap='gray')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Create a pooling layer\n",
    "#   a pooling layer is used to reduce the size of the inputs going forward in the network.\n",
    "#   with a convolutional layer the size has already risen by a factor of feature count.\n",
    "# this esssentially scales down all the images/features by a factor\n",
    "scale_factor = 2\n",
    "# add the pooling layer to the graph\n",
    "pool1 = tf.nn.max_pool( conv1, ksize=[1,scale_factor,scale_factor,1], strides=[1,scale_factor,scale_factor,1], padding='SAME' )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1, 14, 14, 5)"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# run the pooling layer\n",
    "out = sess.run(pool1, feed_dict={x:image})\n",
    "out.shape\n",
    "# the middle 2 dimensions have reduced by half"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAyIAAACcCAYAAACDZR/GAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAIB0hggAAJDOEAEAANL9F0tTbyZchYrPAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0xf2810f0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# split the array to images with different features\n",
    "img = np.dsplit(out[0], features)\n",
    "width = out.shape[1]\n",
    "height = out.shape[2]\n",
    "# display the different features as images\n",
    "plt.figure( figsize=( 14,10 ) )\n",
    "for i in range( features ):\n",
    "    plt.subplot(1, 5, i+1)\n",
    "    plt.tick_params( axis='both', which='both', bottom='off', left='off', labelbottom='off', labelleft='off' )\n",
    "    plt.imshow(img[i].reshape(width, height), cmap='gray')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Most of the graphs attach another convolution and pooling layer\n",
    "# then the graph is attached to a fully connected layer"
   ]
  }
 ],
 "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": "3.5.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}