mnist_tensorflow_tutorial_done_right.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Extracting /tmp/mnist/train-images-idx3-ubyte.gz\n",
      "Extracting /tmp/mnist/train-labels-idx1-ubyte.gz\n",
      "Extracting /tmp/mnist/t10k-images-idx3-ubyte.gz\n",
      "Extracting /tmp/mnist/t10k-labels-idx1-ubyte.gz\n",
      "iteration 0\t accuracy: 0.071\n",
      "iteration 1000\t accuracy: 0.814\n",
      "iteration 2000\t accuracy: 0.851\n",
      "iteration 3000\t accuracy: 0.864\n",
      "iteration 4000\t accuracy: 0.873\n",
      "iteration 5000\t accuracy: 0.876\n",
      "iteration 6000\t accuracy: 0.879\n",
      "iteration 7000\t accuracy: 0.884\n",
      "iteration 8000\t accuracy: 0.888\n",
      "iteration 9000\t accuracy: 0.890\n",
      "iteration 10000\t accuracy: 0.891\n",
      "iteration 11000\t accuracy: 0.893\n",
      "iteration 12000\t accuracy: 0.894\n",
      "iteration 13000\t accuracy: 0.897\n",
      "iteration 14000\t accuracy: 0.897\n",
      "iteration 15000\t accuracy: 0.899\n",
      "iteration 16000\t accuracy: 0.898\n",
      "iteration 17000\t accuracy: 0.899\n",
      "iteration 18000\t accuracy: 0.901\n",
      "iteration 19000\t accuracy: 0.901\n"
     ]
    }
   ],
   "source": [
    "from __future__ import absolute_import\n",
    "from __future__ import division\n",
    "from __future__ import print_function\n",
    "\n",
    "import tensorflow as tf\n",
    "from tensorflow.contrib.learn.python.learn.datasets.mnist import read_data_sets\n",
    "\n",
    "def main():\n",
    "    # import data\n",
    "    mnist = read_data_sets('/tmp/mnist/', one_hot=True)\n",
    "\n",
    "    with tf.Graph().as_default() as g:\n",
    "        # where are you going to allocate memory and perform computations\n",
    "        with tf.device(\"/cpu:0\"):\n",
    "            \n",
    "            # define model \"input placeholders\", i.e. variables that are\n",
    "            # going to be substituted with input data on train/test time\n",
    "            x_ = tf.placeholder(tf.float32, [None, 784])\n",
    "            y_ = tf.placeholder(tf.float32, [None, 10])\n",
    "\n",
    "            # create the actual model\n",
    "            scope_args = {'initializer': tf.random_normal_initializer()}\n",
    "            with tf.variable_scope(\"weights\", **scope_args):\n",
    "                W = tf.get_variable('W', shape=[784, 10])\n",
    "                b = tf.get_variable('b', shape=[10])\n",
    "                y_logits = tf.matmul(x_, W) + b\n",
    "\n",
    "            # naive implementation of loss:\n",
    "            # > losses = y_ * tf.log(tf.nn.softmax(y_logits))\n",
    "            # > tf.reduce_mean(-tf.reduce_sum(losses, 1))\n",
    "            # can be numerically unstable.\n",
    "            #\n",
    "            # so here we use tf.nn.softmax_cross_entropy_with_logits on the raw\n",
    "            # outputs of 'y', and then average across the batch.\n",
    "            \n",
    "            losses = tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=y_logits)\n",
    "            cross_entropy_loss = tf.reduce_mean(losses)\n",
    "            train_step = tf.train.GradientDescentOptimizer(0.1).minimize(cross_entropy_loss)\n",
    "            \n",
    "            y_pred = tf.argmax(tf.nn.softmax(y_logits), dimension=1)\n",
    "            correct_prediction = tf.equal(y_pred, tf.argmax(y_, 1))\n",
    "            accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))\n",
    "\n",
    "    with g.as_default(), tf.Session() as sess:\n",
    "        # that is how we \"execute\" statements \n",
    "        # (return None, e.g. init() or train_op())\n",
    "        # or compute parts of graph defined above (loss, output, etc.)\n",
    "        # given certain input (x_, y_)\n",
    "        sess.run(tf.initialize_all_variables())\n",
    "        \n",
    "        # train\n",
    "        for iter_i in range(20000):\n",
    "            batch_xs, batch_ys = mnist.train.next_batch(100)\n",
    "            sess.run(train_step, feed_dict={x_: batch_xs, y_: batch_ys})\n",
    "            \n",
    "            # test trained model\n",
    "            if iter_i % 1000 == 0:\n",
    "                tf_feed_dict = {x_: mnist.test.images, y_: mnist.test.labels}\n",
    "                acc_value = sess.run(accuracy, feed_dict=tf_feed_dict)\n",
    "                print('iteration %d\\t accuracy: %.3f'%(iter_i, acc_value))\n",
    "                \n",
    "main()"
   ]
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "kernelspec": {
   "display_name": "Python [conda env:tf11]",
   "language": "python",
   "name": "conda-env-tf11-py"
  },
  "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": 1
}