4_convolutions.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "4embtkV0pNxM"
   },
   "source": [
    "Deep Learning\n",
    "=============\n",
    "\n",
    "Assignment 4\n",
    "------------\n",
    "\n",
    "Previously in `2_fullyconnected.ipynb` and `3_regularization.ipynb`, we trained fully connected networks to classify [notMNIST](http://yaroslavvb.blogspot.com/2011/09/notmnist-dataset.html) characters.\n",
    "\n",
    "The goal of this assignment is make the neural network convolutional."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "cellView": "both",
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     }
    },
    "colab_type": "code",
    "collapsed": true,
    "id": "tm2CQN_Cpwj0"
   },
   "outputs": [],
   "source": [
    "# These are all the modules we'll be using later. Make sure you can import them\n",
    "# before proceeding further.\n",
    "from __future__ import print_function\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "from six.moves import cPickle as pickle\n",
    "from six.moves import range"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "cellView": "both",
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     },
     "output_extras": [
      {
       "item_id": 1
      }
     ]
    },
    "colab_type": "code",
    "collapsed": false,
    "executionInfo": {
     "elapsed": 11948,
     "status": "ok",
     "timestamp": 1446658914837,
     "user": {
      "color": "",
      "displayName": "",
      "isAnonymous": false,
      "isMe": true,
      "permissionId": "",
      "photoUrl": "",
      "sessionId": "0",
      "userId": ""
     },
     "user_tz": 480
    },
    "id": "y3-cj1bpmuxc",
    "outputId": "016b1a51-0290-4b08-efdb-8c95ffc3cd01"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training set (200000, 28, 28) (200000,)\n",
      "Validation set (10000, 28, 28) (10000,)\n",
      "Test set (10000, 28, 28) (10000,)\n"
     ]
    }
   ],
   "source": [
    "pickle_file = 'notMNIST.pickle'\n",
    "\n",
    "with open(pickle_file, 'rb') as f:\n",
    "  save = pickle.load(f)\n",
    "  train_dataset = save['train_dataset']\n",
    "  train_labels = save['train_labels']\n",
    "  valid_dataset = save['valid_dataset']\n",
    "  valid_labels = save['valid_labels']\n",
    "  test_dataset = save['test_dataset']\n",
    "  test_labels = save['test_labels']\n",
    "  del save  # hint to help gc free up memory\n",
    "  print('Training set', train_dataset.shape, train_labels.shape)\n",
    "  print('Validation set', valid_dataset.shape, valid_labels.shape)\n",
    "  print('Test set', test_dataset.shape, test_labels.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "L7aHrm6nGDMB"
   },
   "source": [
    "Reformat into a TensorFlow-friendly shape:\n",
    "- convolutions need the image data formatted as a cube (width by height by #channels)\n",
    "- labels as float 1-hot encodings."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "cellView": "both",
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     },
     "output_extras": [
      {
       "item_id": 1
      }
     ]
    },
    "colab_type": "code",
    "collapsed": false,
    "executionInfo": {
     "elapsed": 11952,
     "status": "ok",
     "timestamp": 1446658914857,
     "user": {
      "color": "",
      "displayName": "",
      "isAnonymous": false,
      "isMe": true,
      "permissionId": "",
      "photoUrl": "",
      "sessionId": "0",
      "userId": ""
     },
     "user_tz": 480
    },
    "id": "IRSyYiIIGIzS",
    "outputId": "650a208c-8359-4852-f4f5-8bf10e80ef6c"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training set (200000, 28, 28, 1) (200000, 10)\n",
      "Validation set (10000, 28, 28, 1) (10000, 10)\n",
      "Test set (10000, 28, 28, 1) (10000, 10)\n"
     ]
    }
   ],
   "source": [
    "image_size = 28\n",
    "num_labels = 10\n",
    "num_channels = 1 # grayscale\n",
    "\n",
    "import numpy as np\n",
    "\n",
    "def reformat(dataset, labels):\n",
    "  dataset = dataset.reshape(\n",
    "    (-1, image_size, image_size, num_channels)).astype(np.float32)\n",
    "  labels = (np.arange(num_labels) == labels[:,None]).astype(np.float32)\n",
    "  return dataset, labels\n",
    "train_dataset, train_labels = reformat(train_dataset, train_labels)\n",
    "valid_dataset, valid_labels = reformat(valid_dataset, valid_labels)\n",
    "test_dataset, test_labels = reformat(test_dataset, test_labels)\n",
    "print('Training set', train_dataset.shape, train_labels.shape)\n",
    "print('Validation set', valid_dataset.shape, valid_labels.shape)\n",
    "print('Test set', test_dataset.shape, test_labels.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "cellView": "both",
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     }
    },
    "colab_type": "code",
    "collapsed": true,
    "id": "AgQDIREv02p1"
   },
   "outputs": [],
   "source": [
    "def accuracy(predictions, labels):\n",
    "  return (100.0 * np.sum(np.argmax(predictions, 1) == np.argmax(labels, 1))\n",
    "          / predictions.shape[0])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "5rhgjmROXu2O"
   },
   "source": [
    "Let's build a small network with two convolutional layers, followed by one fully connected layer. Convolutional networks are more expensive computationally, so we'll limit its depth and number of fully connected nodes."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "cellView": "both",
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     }
    },
    "colab_type": "code",
    "collapsed": true,
    "id": "IZYv70SvvOan"
   },
   "outputs": [],
   "source": [
    "batch_size = 16\n",
    "patch_size = 5\n",
    "depth = 16\n",
    "num_hidden = 64\n",
    "\n",
    "graph = tf.Graph()\n",
    "\n",
    "with graph.as_default():\n",
    "\n",
    "  # Input data.\n",
    "  tf_train_dataset = tf.placeholder(\n",
    "    tf.float32, shape=(batch_size, image_size, image_size, num_channels))\n",
    "  tf_train_labels = tf.placeholder(tf.float32, shape=(batch_size, num_labels))\n",
    "  tf_valid_dataset = tf.constant(valid_dataset)\n",
    "  tf_test_dataset = tf.constant(test_dataset)\n",
    "  \n",
    "  # Variables.\n",
    "  layer1_weights = tf.Variable(tf.truncated_normal(\n",
    "      [patch_size, patch_size, num_channels, depth], stddev=0.1))\n",
    "  layer1_biases = tf.Variable(tf.zeros([depth]))\n",
    "  layer2_weights = tf.Variable(tf.truncated_normal(\n",
    "      [patch_size, patch_size, depth, depth], stddev=0.1))\n",
    "  layer2_biases = tf.Variable(tf.constant(1.0, shape=[depth]))\n",
    "  layer3_weights = tf.Variable(tf.truncated_normal(\n",
    "      [image_size // 4 * image_size // 4 * depth, num_hidden], stddev=0.1))\n",
    "  layer3_biases = tf.Variable(tf.constant(1.0, shape=[num_hidden]))\n",
    "  layer4_weights = tf.Variable(tf.truncated_normal(\n",
    "      [num_hidden, num_labels], stddev=0.1))\n",
    "  layer4_biases = tf.Variable(tf.constant(1.0, shape=[num_labels]))\n",
    "  \n",
    "  # Model.\n",
    "  def model(data):\n",
    "    conv = tf.nn.conv2d(data, layer1_weights, [1, 2, 2, 1], padding='SAME')\n",
    "    hidden = tf.nn.relu(conv + layer1_biases)\n",
    "    conv = tf.nn.conv2d(hidden, layer2_weights, [1, 2, 2, 1], padding='SAME')\n",
    "    hidden = tf.nn.relu(conv + layer2_biases)\n",
    "    shape = hidden.get_shape().as_list()\n",
    "    reshape = tf.reshape(hidden, [shape[0], shape[1] * shape[2] * shape[3]])\n",
    "    hidden = tf.nn.relu(tf.matmul(reshape, layer3_weights) + layer3_biases)\n",
    "    return tf.matmul(hidden, layer4_weights) + layer4_biases\n",
    "  \n",
    "  # Training computation.\n",
    "  logits = model(tf_train_dataset)\n",
    "  loss = tf.reduce_mean(\n",
    "    tf.nn.softmax_cross_entropy_with_logits(labels=tf_train_labels, logits=logits))\n",
    "    \n",
    "  # Optimizer.\n",
    "  optimizer = tf.train.GradientDescentOptimizer(0.05).minimize(loss)\n",
    "  \n",
    "  # Predictions for the training, validation, and test data.\n",
    "  train_prediction = tf.nn.softmax(logits)\n",
    "  valid_prediction = tf.nn.softmax(model(tf_valid_dataset))\n",
    "  test_prediction = tf.nn.softmax(model(tf_test_dataset))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "cellView": "both",
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     },
     "output_extras": [
      {
       "item_id": 37
      }
     ]
    },
    "colab_type": "code",
    "collapsed": false,
    "executionInfo": {
     "elapsed": 63292,
     "status": "ok",
     "timestamp": 1446658966251,
     "user": {
      "color": "",
      "displayName": "",
      "isAnonymous": false,
      "isMe": true,
      "permissionId": "",
      "photoUrl": "",
      "sessionId": "0",
      "userId": ""
     },
     "user_tz": 480
    },
    "id": "noKFb2UovVFR",
    "outputId": "28941338-2ef9-4088-8bd1-44295661e628"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initialized\n",
      "Minibatch loss at step 0: 2.850859\n",
      "Minibatch accuracy: 12.5%\n",
      "Validation accuracy: 10.0%\n",
      "Minibatch loss at step 50: 1.782504\n",
      "Minibatch accuracy: 31.2%\n",
      "Validation accuracy: 49.4%\n",
      "Minibatch loss at step 100: 1.055282\n",
      "Minibatch accuracy: 62.5%\n",
      "Validation accuracy: 69.6%\n",
      "Minibatch loss at step 150: 1.396942\n",
      "Minibatch accuracy: 56.2%\n",
      "Validation accuracy: 74.4%\n",
      "Minibatch loss at step 200: 0.914393\n",
      "Minibatch accuracy: 62.5%\n",
      "Validation accuracy: 77.6%\n",
      "Minibatch loss at step 250: 0.429481\n",
      "Minibatch accuracy: 87.5%\n",
      "Validation accuracy: 78.9%\n",
      "Minibatch loss at step 300: 0.390205\n",
      "Minibatch accuracy: 87.5%\n",
      "Validation accuracy: 79.5%\n",
      "Minibatch loss at step 350: 0.998649\n",
      "Minibatch accuracy: 75.0%\n",
      "Validation accuracy: 80.5%\n",
      "Minibatch loss at step 400: 0.481183\n",
      "Minibatch accuracy: 87.5%\n",
      "Validation accuracy: 78.6%\n",
      "Minibatch loss at step 450: 0.499964\n",
      "Minibatch accuracy: 81.2%\n",
      "Validation accuracy: 80.2%\n",
      "Minibatch loss at step 500: 0.871840\n",
      "Minibatch accuracy: 81.2%\n",
      "Validation accuracy: 80.3%\n",
      "Minibatch loss at step 550: 1.063585\n",
      "Minibatch accuracy: 68.8%\n",
      "Validation accuracy: 81.1%\n",
      "Minibatch loss at step 600: 0.462179\n",
      "Minibatch accuracy: 81.2%\n",
      "Validation accuracy: 80.2%\n",
      "Minibatch loss at step 650: 0.991397\n",
      "Minibatch accuracy: 68.8%\n",
      "Validation accuracy: 81.7%\n",
      "Minibatch loss at step 700: 0.696619\n",
      "Minibatch accuracy: 81.2%\n",
      "Validation accuracy: 81.7%\n",
      "Minibatch loss at step 750: 0.885388\n",
      "Minibatch accuracy: 81.2%\n",
      "Validation accuracy: 81.9%\n",
      "Minibatch loss at step 800: 1.021665\n",
      "Minibatch accuracy: 75.0%\n",
      "Validation accuracy: 82.2%\n",
      "Minibatch loss at step 850: 0.446952\n",
      "Minibatch accuracy: 87.5%\n",
      "Validation accuracy: 82.2%\n",
      "Minibatch loss at step 900: 0.341800\n",
      "Minibatch accuracy: 93.8%\n",
      "Validation accuracy: 82.7%\n",
      "Minibatch loss at step 950: 0.482371\n",
      "Minibatch accuracy: 93.8%\n",
      "Validation accuracy: 82.7%\n",
      "Minibatch loss at step 1000: 0.453269\n",
      "Minibatch accuracy: 81.2%\n",
      "Validation accuracy: 82.4%\n",
      "Test accuracy: 88.9%\n"
     ]
    }
   ],
   "source": [
    "num_steps = 1001\n",
    "\n",
    "with tf.Session(graph=graph) as session:\n",
    "  tf.global_variables_initializer().run()\n",
    "  print('Initialized')\n",
    "  for step in range(num_steps):\n",
    "    offset = (step * batch_size) % (train_labels.shape[0] - batch_size)\n",
    "    batch_data = train_dataset[offset:(offset + batch_size), :, :, :]\n",
    "    batch_labels = train_labels[offset:(offset + batch_size), :]\n",
    "    feed_dict = {tf_train_dataset : batch_data, tf_train_labels : batch_labels}\n",
    "    _, l, predictions = session.run(\n",
    "      [optimizer, loss, train_prediction], feed_dict=feed_dict)\n",
    "    if (step % 50 == 0):\n",
    "      print('Minibatch loss at step %d: %f' % (step, l))\n",
    "      print('Minibatch accuracy: %.1f%%' % accuracy(predictions, batch_labels))\n",
    "      print('Validation accuracy: %.1f%%' % accuracy(\n",
    "        valid_prediction.eval(), valid_labels))\n",
    "  print('Test accuracy: %.1f%%' % accuracy(test_prediction.eval(), test_labels))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "KedKkn4EutIK"
   },
   "source": [
    "---\n",
    "Problem 1\n",
    "---------\n",
    "\n",
    "The convolutional model above uses convolutions with stride 2 to reduce the dimensionality. Replace the strides by a max pooling operation (`nn.max_pool()`) of stride 2 and kernel size 2.\n",
    "\n",
    "---"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[16, 7, 7, 16]\n",
      "[10000, 7, 7, 16]\n",
      "[10000, 7, 7, 16]\n"
     ]
    }
   ],
   "source": [
    "batch_size = 16\n",
    "patch_size = 5\n",
    "depth = 16\n",
    "num_hidden = 64\n",
    "\n",
    "graph = tf.Graph()\n",
    "\n",
    "with graph.as_default():\n",
    "\n",
    "  # Input data.\n",
    "  tf_train_dataset = tf.placeholder(\n",
    "    tf.float32, shape=(batch_size, image_size, image_size, num_channels))\n",
    "  tf_train_labels = tf.placeholder(tf.float32, shape=(batch_size, num_labels))\n",
    "  tf_valid_dataset = tf.constant(valid_dataset)\n",
    "  tf_test_dataset = tf.constant(test_dataset)\n",
    "  \n",
    "  # Variables.\n",
    "  layer1_weights = tf.Variable(tf.truncated_normal(\n",
    "      [patch_size, patch_size, num_channels, depth], stddev=0.1))\n",
    "  layer1_biases = tf.Variable(tf.zeros([depth]))\n",
    "  layer2_weights = tf.Variable(tf.truncated_normal(\n",
    "      [patch_size, patch_size, depth, depth], stddev=0.1))\n",
    "  layer2_biases = tf.Variable(tf.constant(1.0, shape=[depth]))\n",
    "  layer3_weights = tf.Variable(tf.truncated_normal(\n",
    "      [image_size // 4 * image_size // 4 * depth, num_hidden], stddev=0.1))\n",
    "  layer3_biases = tf.Variable(tf.constant(1.0, shape=[num_hidden]))\n",
    "  layer4_weights = tf.Variable(tf.truncated_normal(\n",
    "      [num_hidden, num_labels], stddev=0.1))\n",
    "  layer4_biases = tf.Variable(tf.constant(1.0, shape=[num_labels]))\n",
    "  \n",
    "  # Model.\n",
    "  def model(data):\n",
    "    conv = tf.nn.conv2d(data, layer1_weights, [1, 1, 1, 1], padding='SAME')\n",
    "    hidden = tf.nn.relu(conv + layer1_biases)\n",
    "    pool = tf.nn.max_pool(hidden, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')\n",
    "    conv = tf.nn.conv2d(pool, layer2_weights, [1, 1, 1, 1], padding='SAME')\n",
    "    hidden = tf.nn.relu(conv + layer2_biases)\n",
    "    pool = tf.nn.max_pool(hidden, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')\n",
    "    shape = pool.get_shape().as_list()\n",
    "    print(shape)\n",
    "    reshape = tf.reshape(pool, [shape[0], shape[1] * shape[2] * shape[3]])\n",
    "    hidden = tf.nn.relu(tf.matmul(reshape, layer3_weights) + layer3_biases)\n",
    "    return tf.matmul(hidden, layer4_weights) + layer4_biases\n",
    "  \n",
    "  # Training computation.\n",
    "  logits = model(tf_train_dataset)\n",
    "  loss = tf.reduce_mean(\n",
    "    tf.nn.softmax_cross_entropy_with_logits(labels=tf_train_labels, logits=logits))\n",
    "    \n",
    "  # Optimizer.\n",
    "  optimizer = tf.train.GradientDescentOptimizer(0.05).minimize(loss)\n",
    "  \n",
    "  # Predictions for the training, validation, and test data.\n",
    "  train_prediction = tf.nn.softmax(logits)\n",
    "  valid_prediction = tf.nn.softmax(model(tf_valid_dataset))\n",
    "  test_prediction = tf.nn.softmax(model(tf_test_dataset))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initialized\n",
      "Minibatch loss at step 0: 4.163422\n",
      "Minibatch accuracy: 6.2%\n",
      "Validation accuracy: 10.4%\n",
      "Minibatch loss at step 50: 2.030268\n",
      "Minibatch accuracy: 25.0%\n",
      "Validation accuracy: 36.7%\n",
      "Minibatch loss at step 100: 1.173907\n",
      "Minibatch accuracy: 68.8%\n",
      "Validation accuracy: 62.7%\n",
      "Minibatch loss at step 150: 1.047236\n",
      "Minibatch accuracy: 56.2%\n",
      "Validation accuracy: 68.3%\n",
      "Minibatch loss at step 200: 0.810297\n",
      "Minibatch accuracy: 75.0%\n",
      "Validation accuracy: 76.5%\n",
      "Minibatch loss at step 250: 0.542067\n",
      "Minibatch accuracy: 87.5%\n",
      "Validation accuracy: 78.8%\n",
      "Minibatch loss at step 300: 0.571049\n",
      "Minibatch accuracy: 81.2%\n",
      "Validation accuracy: 80.0%\n",
      "Minibatch loss at step 350: 0.890173\n",
      "Minibatch accuracy: 75.0%\n",
      "Validation accuracy: 81.2%\n",
      "Minibatch loss at step 400: 0.513948\n",
      "Minibatch accuracy: 93.8%\n",
      "Validation accuracy: 80.0%\n",
      "Minibatch loss at step 450: 0.341944\n",
      "Minibatch accuracy: 87.5%\n",
      "Validation accuracy: 81.1%\n",
      "Minibatch loss at step 500: 0.688224\n",
      "Minibatch accuracy: 81.2%\n",
      "Validation accuracy: 82.0%\n",
      "Minibatch loss at step 550: 1.036529\n",
      "Minibatch accuracy: 75.0%\n",
      "Validation accuracy: 82.7%\n",
      "Minibatch loss at step 600: 0.395581\n",
      "Minibatch accuracy: 87.5%\n",
      "Validation accuracy: 82.1%\n",
      "Minibatch loss at step 650: 0.931773\n",
      "Minibatch accuracy: 68.8%\n",
      "Validation accuracy: 82.0%\n",
      "Minibatch loss at step 700: 0.623311\n",
      "Minibatch accuracy: 87.5%\n",
      "Validation accuracy: 82.6%\n",
      "Minibatch loss at step 750: 0.871678\n",
      "Minibatch accuracy: 81.2%\n",
      "Validation accuracy: 83.4%\n",
      "Minibatch loss at step 800: 1.021408\n",
      "Minibatch accuracy: 68.8%\n",
      "Validation accuracy: 83.5%\n",
      "Minibatch loss at step 850: 0.544716\n",
      "Minibatch accuracy: 87.5%\n",
      "Validation accuracy: 83.8%\n",
      "Minibatch loss at step 900: 0.280557\n",
      "Minibatch accuracy: 93.8%\n",
      "Validation accuracy: 83.8%\n",
      "Minibatch loss at step 950: 0.679899\n",
      "Minibatch accuracy: 81.2%\n",
      "Validation accuracy: 84.1%\n",
      "Minibatch loss at step 1000: 0.308318\n",
      "Minibatch accuracy: 87.5%\n",
      "Validation accuracy: 83.7%\n",
      "Test accuracy: 90.0%\n"
     ]
    }
   ],
   "source": [
    "num_steps = 1001\n",
    "\n",
    "with tf.Session(graph=graph) as session:\n",
    "  tf.global_variables_initializer().run()\n",
    "  print('Initialized')\n",
    "  for step in range(num_steps):\n",
    "    offset = (step * batch_size) % (train_labels.shape[0] - batch_size)\n",
    "    batch_data = train_dataset[offset:(offset + batch_size), :, :, :]\n",
    "    batch_labels = train_labels[offset:(offset + batch_size), :]\n",
    "    feed_dict = {tf_train_dataset : batch_data, tf_train_labels : batch_labels}\n",
    "    _, l, predictions = session.run(\n",
    "      [optimizer, loss, train_prediction], feed_dict=feed_dict)\n",
    "    if (step % 50 == 0):\n",
    "      print('Minibatch loss at step %d: %f' % (step, l))\n",
    "      print('Minibatch accuracy: %.1f%%' % accuracy(predictions, batch_labels))\n",
    "      print('Validation accuracy: %.1f%%' % accuracy(\n",
    "        valid_prediction.eval(), valid_labels))\n",
    "  print('Test accuracy: %.1f%%' % accuracy(test_prediction.eval(), test_labels))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "klf21gpbAgb-"
   },
   "source": [
    "---\n",
    "Problem 2\n",
    "---------\n",
    "\n",
    "Try to get the best performance you can using a convolutional net. Look for example at the classic [LeNet5](http://yann.lecun.com/exdb/lenet/) architecture, adding Dropout, and/or adding learning rate decay.\n",
    "\n",
    "---"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[64, 7, 7, 32]\n",
      "[10000, 7, 7, 32]\n",
      "[10000, 7, 7, 32]\n",
      "Initialized\n",
      "Minibatch loss at step 0: 5.824993\n",
      "Minibatch accuracy: 6.2%\n",
      "Validation accuracy: 9.7%\n",
      "Minibatch loss at step 500: 0.703758\n",
      "Minibatch accuracy: 75.0%\n",
      "Validation accuracy: 82.7%\n",
      "Minibatch loss at step 1000: 0.408750\n",
      "Minibatch accuracy: 89.1%\n",
      "Validation accuracy: 85.4%\n",
      "Minibatch loss at step 1500: 0.242372\n",
      "Minibatch accuracy: 90.6%\n",
      "Validation accuracy: 86.6%\n",
      "Minibatch loss at step 2000: 0.527866\n",
      "Minibatch accuracy: 84.4%\n",
      "Validation accuracy: 87.3%\n",
      "Minibatch loss at step 2500: 0.646556\n",
      "Minibatch accuracy: 81.2%\n",
      "Validation accuracy: 87.9%\n",
      "Minibatch loss at step 3000: 0.309477\n",
      "Minibatch accuracy: 92.2%\n",
      "Validation accuracy: 87.8%\n",
      "Minibatch loss at step 3500: 0.153755\n",
      "Minibatch accuracy: 95.3%\n",
      "Validation accuracy: 88.4%\n",
      "Minibatch loss at step 4000: 0.432996\n",
      "Minibatch accuracy: 90.6%\n",
      "Validation accuracy: 88.6%\n",
      "Minibatch loss at step 4500: 0.262458\n",
      "Minibatch accuracy: 95.3%\n",
      "Validation accuracy: 88.7%\n",
      "Minibatch loss at step 5000: 0.250018\n",
      "Minibatch accuracy: 93.8%\n",
      "Validation accuracy: 89.2%\n",
      "Minibatch loss at step 5500: 0.394233\n",
      "Minibatch accuracy: 87.5%\n",
      "Validation accuracy: 89.3%\n",
      "Minibatch loss at step 6000: 0.351470\n",
      "Minibatch accuracy: 85.9%\n",
      "Validation accuracy: 89.1%\n",
      "Minibatch loss at step 6500: 0.370727\n",
      "Minibatch accuracy: 87.5%\n",
      "Validation accuracy: 89.2%\n",
      "Minibatch loss at step 7000: 0.102480\n",
      "Minibatch accuracy: 96.9%\n",
      "Validation accuracy: 89.3%\n",
      "Minibatch loss at step 7500: 0.197446\n",
      "Minibatch accuracy: 93.8%\n",
      "Validation accuracy: 89.5%\n",
      "Minibatch loss at step 8000: 0.289213\n",
      "Minibatch accuracy: 92.2%\n",
      "Validation accuracy: 89.8%\n",
      "Minibatch loss at step 8500: 0.414156\n",
      "Minibatch accuracy: 87.5%\n",
      "Validation accuracy: 90.0%\n",
      "Minibatch loss at step 9000: 0.313443\n",
      "Minibatch accuracy: 89.1%\n",
      "Validation accuracy: 89.5%\n",
      "Minibatch loss at step 9500: 0.216289\n",
      "Minibatch accuracy: 95.3%\n",
      "Validation accuracy: 90.0%\n",
      "Minibatch loss at step 10000: 0.169348\n",
      "Minibatch accuracy: 95.3%\n",
      "Validation accuracy: 90.1%\n",
      "Test accuracy: 95.3%\n"
     ]
    }
   ],
   "source": [
    "batch_size = 64\n",
    "patch_size = 5\n",
    "depth = 32\n",
    "num_hidden = 64\n",
    "\n",
    "graph = tf.Graph()\n",
    "\n",
    "with graph.as_default():\n",
    "\n",
    "  # Input data.\n",
    "  tf_train_dataset = tf.placeholder(\n",
    "    tf.float32, shape=(batch_size, image_size, image_size, num_channels))\n",
    "  tf_train_labels = tf.placeholder(tf.float32, shape=(batch_size, num_labels))\n",
    "  tf_valid_dataset = tf.constant(valid_dataset)\n",
    "  tf_test_dataset = tf.constant(test_dataset)\n",
    "  \n",
    "  # Variables.\n",
    "  layer1_weights = tf.Variable(tf.truncated_normal(\n",
    "      [patch_size, patch_size, num_channels, depth], stddev=0.1))\n",
    "  layer1_biases = tf.Variable(tf.zeros([depth]))\n",
    "  layer2_weights = tf.Variable(tf.truncated_normal(\n",
    "      [patch_size, patch_size, depth, depth], stddev=0.1))\n",
    "  layer2_biases = tf.Variable(tf.constant(1.0, shape=[depth]))\n",
    "  layer3_weights = tf.Variable(tf.truncated_normal(\n",
    "      [image_size // 4 * image_size // 4 * depth, num_hidden], stddev=0.1))\n",
    "  layer3_biases = tf.Variable(tf.constant(1.0, shape=[num_hidden]))\n",
    "  layer4_weights = tf.Variable(tf.truncated_normal(\n",
    "      [num_hidden, num_labels], stddev=0.1))\n",
    "  layer4_biases = tf.Variable(tf.constant(1.0, shape=[num_labels]))\n",
    "  \n",
    "  # Model.\n",
    "  def model(data):\n",
    "    conv = tf.nn.conv2d(data, layer1_weights, [1, 1, 1, 1], padding='SAME')\n",
    "    hidden = tf.nn.relu(conv + layer1_biases)\n",
    "    pool = tf.nn.avg_pool(hidden, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')\n",
    "    conv = tf.nn.conv2d(pool, layer2_weights, [1, 1, 1, 1], padding='SAME')\n",
    "    hidden = tf.nn.relu(conv + layer2_biases)\n",
    "    pool = tf.nn.avg_pool(hidden, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')\n",
    "    shape = pool.get_shape().as_list()\n",
    "    print(shape)\n",
    "    reshape = tf.reshape(pool, [shape[0], shape[1] * shape[2] * shape[3]])\n",
    "    hidden = tf.nn.relu(tf.matmul(reshape, layer3_weights) + layer3_biases)\n",
    "    return tf.matmul(hidden, layer4_weights) + layer4_biases\n",
    "  \n",
    "  # Training computation.\n",
    "  logits = model(tf_train_dataset)\n",
    "  loss = tf.reduce_mean(\n",
    "    tf.nn.softmax_cross_entropy_with_logits(labels=tf_train_labels, logits=logits))\n",
    "    \n",
    "  # Optimizer.\n",
    "  optimizer = tf.train.GradientDescentOptimizer(0.05).minimize(loss)\n",
    "  \n",
    "  # Predictions for the training, validation, and test data.\n",
    "  train_prediction = tf.nn.softmax(logits)\n",
    "  valid_prediction = tf.nn.softmax(model(tf_valid_dataset))\n",
    "  test_prediction = tf.nn.softmax(model(tf_test_dataset))\n",
    "\n",
    "num_steps = 10001\n",
    "\n",
    "with tf.Session(graph=graph) as session:\n",
    "  tf.global_variables_initializer().run()\n",
    "  print('Initialized')\n",
    "  for step in range(num_steps):\n",
    "    offset = (step * batch_size) % (train_labels.shape[0] - batch_size)\n",
    "    batch_data = train_dataset[offset:(offset + batch_size), :, :, :]\n",
    "    batch_labels = train_labels[offset:(offset + batch_size), :]\n",
    "    feed_dict = {tf_train_dataset : batch_data, tf_train_labels : batch_labels}\n",
    "    _, l, predictions = session.run(\n",
    "      [optimizer, loss, train_prediction], feed_dict=feed_dict)\n",
    "    if (step % 500 == 0):\n",
    "      print('Minibatch loss at step %d: %f' % (step, l))\n",
    "      print('Minibatch accuracy: %.1f%%' % accuracy(predictions, batch_labels))\n",
    "      print('Validation accuracy: %.1f%%' % accuracy(\n",
    "        valid_prediction.eval(), valid_labels))\n",
    "  print('Test accuracy: %.1f%%' % accuracy(test_prediction.eval(), test_labels))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": false,
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[64, 7, 7, 32]\n",
      "[10000, 7, 7, 32]\n",
      "[10000, 7, 7, 32]\n",
      "Initialized\n",
      "Minibatch loss at step 0: 2.396194\n",
      "Minibatch accuracy: 12.5%\n",
      "Validation accuracy: 10.0%\n",
      "Minibatch loss at step 500: 0.716294\n",
      "Minibatch accuracy: 78.1%\n",
      "Validation accuracy: 82.4%\n",
      "Minibatch loss at step 1000: 0.488117\n",
      "Minibatch accuracy: 89.1%\n",
      "Validation accuracy: 85.0%\n",
      "Minibatch loss at step 1500: 0.231508\n",
      "Minibatch accuracy: 93.8%\n",
      "Validation accuracy: 86.6%\n",
      "Minibatch loss at step 2000: 0.493171\n",
      "Minibatch accuracy: 85.9%\n",
      "Validation accuracy: 87.4%\n",
      "Minibatch loss at step 2500: 0.674689\n",
      "Minibatch accuracy: 78.1%\n",
      "Validation accuracy: 87.8%\n",
      "Minibatch loss at step 3000: 0.364573\n",
      "Minibatch accuracy: 87.5%\n",
      "Validation accuracy: 88.0%\n",
      "Minibatch loss at step 3500: 0.183410\n",
      "Minibatch accuracy: 95.3%\n",
      "Validation accuracy: 88.6%\n",
      "Minibatch loss at step 4000: 0.405490\n",
      "Minibatch accuracy: 90.6%\n",
      "Validation accuracy: 88.6%\n",
      "Minibatch loss at step 4500: 0.272295\n",
      "Minibatch accuracy: 95.3%\n",
      "Validation accuracy: 89.3%\n",
      "Minibatch loss at step 5000: 0.266061\n",
      "Minibatch accuracy: 89.1%\n",
      "Validation accuracy: 89.2%\n",
      "Minibatch loss at step 5500: 0.364884\n",
      "Minibatch accuracy: 89.1%\n",
      "Validation accuracy: 89.8%\n",
      "Minibatch loss at step 6000: 0.284050\n",
      "Minibatch accuracy: 90.6%\n",
      "Validation accuracy: 89.8%\n",
      "Minibatch loss at step 6500: 0.336731\n",
      "Minibatch accuracy: 85.9%\n",
      "Validation accuracy: 89.7%\n",
      "Minibatch loss at step 7000: 0.104319\n",
      "Minibatch accuracy: 96.9%\n",
      "Validation accuracy: 90.0%\n",
      "Minibatch loss at step 7500: 0.185874\n",
      "Minibatch accuracy: 95.3%\n",
      "Validation accuracy: 90.0%\n",
      "Minibatch loss at step 8000: 0.333703\n",
      "Minibatch accuracy: 87.5%\n",
      "Validation accuracy: 90.3%\n",
      "Minibatch loss at step 8500: 0.473503\n",
      "Minibatch accuracy: 90.6%\n",
      "Validation accuracy: 90.2%\n",
      "Minibatch loss at step 9000: 0.263206\n",
      "Minibatch accuracy: 92.2%\n",
      "Validation accuracy: 90.2%\n",
      "Minibatch loss at step 9500: 0.217754\n",
      "Minibatch accuracy: 92.2%\n",
      "Validation accuracy: 90.3%\n",
      "Minibatch loss at step 10000: 0.139210\n",
      "Minibatch accuracy: 95.3%\n",
      "Validation accuracy: 90.5%\n",
      "Minibatch loss at step 10500: 0.108445\n",
      "Minibatch accuracy: 96.9%\n",
      "Validation accuracy: 90.6%\n",
      "Minibatch loss at step 11000: 0.405528\n",
      "Minibatch accuracy: 87.5%\n",
      "Validation accuracy: 90.5%\n",
      "Minibatch loss at step 11500: 0.294593\n",
      "Minibatch accuracy: 92.2%\n",
      "Validation accuracy: 90.6%\n",
      "Minibatch loss at step 12000: 0.248036\n",
      "Minibatch accuracy: 93.8%\n",
      "Validation accuracy: 90.7%\n",
      "Minibatch loss at step 12500: 0.276134\n",
      "Minibatch accuracy: 92.2%\n",
      "Validation accuracy: 90.5%\n",
      "Minibatch loss at step 13000: 0.220214\n",
      "Minibatch accuracy: 93.8%\n",
      "Validation accuracy: 90.6%\n",
      "Minibatch loss at step 13500: 0.338955\n",
      "Minibatch accuracy: 87.5%\n",
      "Validation accuracy: 90.8%\n",
      "Minibatch loss at step 14000: 0.395957\n",
      "Minibatch accuracy: 87.5%\n",
      "Validation accuracy: 90.8%\n",
      "Minibatch loss at step 14500: 0.346205\n",
      "Minibatch accuracy: 89.1%\n",
      "Validation accuracy: 90.8%\n",
      "Minibatch loss at step 15000: 0.146459\n",
      "Minibatch accuracy: 95.3%\n",
      "Validation accuracy: 91.0%\n",
      "Minibatch loss at step 15500: 0.373725\n",
      "Minibatch accuracy: 87.5%\n",
      "Validation accuracy: 90.8%\n",
      "Minibatch loss at step 16000: 0.327357\n",
      "Minibatch accuracy: 87.5%\n",
      "Validation accuracy: 90.9%\n",
      "Minibatch loss at step 16500: 0.327153\n",
      "Minibatch accuracy: 92.2%\n",
      "Validation accuracy: 90.8%\n",
      "Minibatch loss at step 17000: 0.296966\n",
      "Minibatch accuracy: 90.6%\n",
      "Validation accuracy: 90.8%\n",
      "Minibatch loss at step 17500: 0.291693\n",
      "Minibatch accuracy: 89.1%\n",
      "Validation accuracy: 91.3%\n",
      "Minibatch loss at step 18000: 0.232083\n",
      "Minibatch accuracy: 93.8%\n",
      "Validation accuracy: 90.8%\n",
      "Minibatch loss at step 18500: 0.359376\n",
      "Minibatch accuracy: 87.5%\n",
      "Validation accuracy: 91.0%\n",
      "Minibatch loss at step 19000: 0.154676\n",
      "Minibatch accuracy: 96.9%\n",
      "Validation accuracy: 91.1%\n",
      "Minibatch loss at step 19500: 0.327577\n",
      "Minibatch accuracy: 90.6%\n",
      "Validation accuracy: 91.2%\n",
      "Minibatch loss at step 20000: 0.320832\n",
      "Minibatch accuracy: 89.1%\n",
      "Validation accuracy: 91.2%\n",
      "Test accuracy: 96.3%\n"
     ]
    }
   ],
   "source": [
    "batch_size = 64\n",
    "patch_size = 5\n",
    "depth = 32\n",
    "num_hidden = 64\n",
    "\n",
    "graph = tf.Graph()\n",
    "\n",
    "with graph.as_default():\n",
    "\n",
    "  # Input data.\n",
    "  tf_train_dataset = tf.placeholder(\n",
    "    tf.float32, shape=(batch_size, image_size, image_size, num_channels))\n",
    "  tf_train_labels = tf.placeholder(tf.float32, shape=(batch_size, num_labels))\n",
    "  tf_valid_dataset = tf.constant(valid_dataset)\n",
    "  tf_test_dataset = tf.constant(test_dataset)\n",
    "  \n",
    "  # Variables.\n",
    "  layer1_weights = tf.Variable(tf.truncated_normal(\n",
    "      [patch_size, patch_size, num_channels, depth], stddev=np.sqrt(2.0 / (image_size * image_size))))\n",
    "  layer1_biases = tf.Variable(tf.zeros([depth]))\n",
    "  layer2_weights = tf.Variable(tf.truncated_normal(\n",
    "      [patch_size, patch_size, depth, depth], stddev=np.sqrt(2.0 / (image_size * image_size))))\n",
    "  layer2_biases = tf.Variable(tf.constant(1.0, shape=[depth]))\n",
    "  layer3_weights = tf.Variable(tf.truncated_normal(\n",
    "      [image_size // 4 * image_size // 4 * depth, num_hidden], stddev=np.sqrt(2.0 / (image_size * image_size))))\n",
    "  layer3_biases = tf.Variable(tf.constant(1.0, shape=[num_hidden]))\n",
    "  layer4_weights = tf.Variable(tf.truncated_normal(\n",
    "      [num_hidden, num_labels], stddev=np.sqrt(2.0 / (image_size * image_size))))\n",
    "  layer4_biases = tf.Variable(tf.constant(1.0, shape=[num_labels]))\n",
    "  \n",
    "  # Model.\n",
    "  def model(data, train=False):\n",
    "    conv = tf.nn.conv2d(data, layer1_weights, [1, 1, 1, 1], padding='SAME')\n",
    "    hidden = tf.nn.relu(conv + layer1_biases)\n",
    "    pool = tf.nn.avg_pool(hidden, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')\n",
    "    conv = tf.nn.conv2d(pool, layer2_weights, [1, 1, 1, 1], padding='SAME')\n",
    "    hidden = tf.nn.relu(conv + layer2_biases)\n",
    "    pool = tf.nn.max_pool(hidden, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')\n",
    "    shape = pool.get_shape().as_list()\n",
    "    print(shape)\n",
    "    reshape = tf.reshape(pool, [shape[0], shape[1] * shape[2] * shape[3]])\n",
    "    hidden = tf.nn.relu(tf.matmul(reshape, layer3_weights) + layer3_biases)\n",
    "    if train:\n",
    "        hidden = tf.nn.dropout(hidden, keep_prob=0.3)\n",
    "    return tf.matmul(hidden, layer4_weights) + layer4_biases\n",
    "  \n",
    "  # Training computation.\n",
    "  logits = model(tf_train_dataset)\n",
    "  loss = tf.reduce_mean(\n",
    "    tf.nn.softmax_cross_entropy_with_logits(labels=tf_train_labels, logits=logits))\n",
    "    \n",
    "  # Optimizer.\n",
    "  optimizer = tf.train.GradientDescentOptimizer(0.05).minimize(loss)\n",
    "  \n",
    "  # Predictions for the training, validation, and test data.\n",
    "  train_prediction = tf.nn.softmax(logits)\n",
    "  valid_prediction = tf.nn.softmax(model(tf_valid_dataset))\n",
    "  test_prediction = tf.nn.softmax(model(tf_test_dataset))\n",
    "\n",
    "num_steps = 20001\n",
    "\n",
    "with tf.Session(graph=graph) as session:\n",
    "  tf.global_variables_initializer().run()\n",
    "  print('Initialized')\n",
    "  for step in range(num_steps):\n",
    "    offset = (step * batch_size) % (train_labels.shape[0] - batch_size)\n",
    "    batch_data = train_dataset[offset:(offset + batch_size), :, :, :]\n",
    "    batch_labels = train_labels[offset:(offset + batch_size), :]\n",
    "    feed_dict = {tf_train_dataset : batch_data, tf_train_labels : batch_labels}\n",
    "    _, l, predictions = session.run(\n",
    "      [optimizer, loss, train_prediction], feed_dict=feed_dict)\n",
    "    if (step % 500 == 0):\n",
    "      print('Minibatch loss at step %d: %f' % (step, l))\n",
    "      print('Minibatch accuracy: %.1f%%' % accuracy(predictions, batch_labels))\n",
    "      print('Validation accuracy: %.1f%%' % accuracy(\n",
    "        valid_prediction.eval(), valid_labels))\n",
    "  print('Test accuracy: %.1f%%' % accuracy(test_prediction.eval(), test_labels))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": false,
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[64, 7, 7, 32]\n",
      "[10000, 7, 7, 32]\n",
      "[10000, 7, 7, 32]\n",
      "Initialized\n",
      "Minibatch loss at step 0: 2.481456\n",
      "Minibatch accuracy: 4.7%\n",
      "Validation accuracy: 10.0%\n",
      "Minibatch loss at step 500: 0.806655\n",
      "Minibatch accuracy: 75.0%\n",
      "Validation accuracy: 80.8%\n",
      "Minibatch loss at step 1000: 0.572952\n",
      "Minibatch accuracy: 85.9%\n",
      "Validation accuracy: 84.1%\n",
      "Minibatch loss at step 1500: 0.291002\n",
      "Minibatch accuracy: 90.6%\n",
      "Validation accuracy: 85.1%\n",
      "Minibatch loss at step 2000: 0.484176\n",
      "Minibatch accuracy: 84.4%\n",
      "Validation accuracy: 86.2%\n",
      "Minibatch loss at step 2500: 0.797123\n",
      "Minibatch accuracy: 76.6%\n",
      "Validation accuracy: 86.7%\n",
      "Minibatch loss at step 3000: 0.313286\n",
      "Minibatch accuracy: 90.6%\n",
      "Validation accuracy: 87.1%\n",
      "Minibatch loss at step 3500: 0.152538\n",
      "Minibatch accuracy: 95.3%\n",
      "Validation accuracy: 87.7%\n",
      "Minibatch loss at step 4000: 0.471414\n",
      "Minibatch accuracy: 84.4%\n",
      "Validation accuracy: 88.2%\n",
      "Minibatch loss at step 4500: 0.299623\n",
      "Minibatch accuracy: 92.2%\n",
      "Validation accuracy: 88.4%\n",
      "Minibatch loss at step 5000: 0.272919\n",
      "Minibatch accuracy: 90.6%\n",
      "Validation accuracy: 88.5%\n",
      "Minibatch loss at step 5500: 0.395781\n",
      "Minibatch accuracy: 89.1%\n",
      "Validation accuracy: 88.9%\n",
      "Minibatch loss at step 6000: 0.329109\n",
      "Minibatch accuracy: 89.1%\n",
      "Validation accuracy: 89.0%\n",
      "Minibatch loss at step 6500: 0.412399\n",
      "Minibatch accuracy: 85.9%\n",
      "Validation accuracy: 89.1%\n",
      "Minibatch loss at step 7000: 0.131765\n",
      "Minibatch accuracy: 96.9%\n",
      "Validation accuracy: 89.2%\n",
      "Minibatch loss at step 7500: 0.179797\n",
      "Minibatch accuracy: 93.8%\n",
      "Validation accuracy: 89.6%\n",
      "Minibatch loss at step 8000: 0.330590\n",
      "Minibatch accuracy: 89.1%\n",
      "Validation accuracy: 89.9%\n",
      "Minibatch loss at step 8500: 0.396308\n",
      "Minibatch accuracy: 92.2%\n",
      "Validation accuracy: 89.9%\n",
      "Minibatch loss at step 9000: 0.316697\n",
      "Minibatch accuracy: 89.1%\n",
      "Validation accuracy: 89.8%\n",
      "Minibatch loss at step 9500: 0.244450\n",
      "Minibatch accuracy: 95.3%\n",
      "Validation accuracy: 89.8%\n",
      "Minibatch loss at step 10000: 0.155566\n",
      "Minibatch accuracy: 95.3%\n",
      "Validation accuracy: 90.0%\n",
      "Minibatch loss at step 10500: 0.140654\n",
      "Minibatch accuracy: 96.9%\n",
      "Validation accuracy: 90.1%\n",
      "Minibatch loss at step 11000: 0.454604\n",
      "Minibatch accuracy: 87.5%\n",
      "Validation accuracy: 90.3%\n",
      "Minibatch loss at step 11500: 0.355300\n",
      "Minibatch accuracy: 89.1%\n",
      "Validation accuracy: 90.1%\n",
      "Minibatch loss at step 12000: 0.296903\n",
      "Minibatch accuracy: 92.2%\n",
      "Validation accuracy: 90.4%\n",
      "Minibatch loss at step 12500: 0.320689\n",
      "Minibatch accuracy: 90.6%\n",
      "Validation accuracy: 90.5%\n",
      "Minibatch loss at step 13000: 0.222681\n",
      "Minibatch accuracy: 93.8%\n",
      "Validation accuracy: 90.4%\n",
      "Minibatch loss at step 13500: 0.307001\n",
      "Minibatch accuracy: 92.2%\n",
      "Validation accuracy: 90.5%\n",
      "Minibatch loss at step 14000: 0.416337\n",
      "Minibatch accuracy: 87.5%\n",
      "Validation accuracy: 90.6%\n",
      "Minibatch loss at step 14500: 0.355889\n",
      "Minibatch accuracy: 85.9%\n",
      "Validation accuracy: 90.7%\n",
      "Minibatch loss at step 15000: 0.140226\n",
      "Minibatch accuracy: 93.8%\n",
      "Validation accuracy: 90.7%\n",
      "Minibatch loss at step 15500: 0.419600\n",
      "Minibatch accuracy: 84.4%\n",
      "Validation accuracy: 90.8%\n",
      "Minibatch loss at step 16000: 0.281715\n",
      "Minibatch accuracy: 92.2%\n",
      "Validation accuracy: 90.7%\n",
      "Minibatch loss at step 16500: 0.363287\n",
      "Minibatch accuracy: 89.1%\n",
      "Validation accuracy: 90.8%\n",
      "Minibatch loss at step 17000: 0.361800\n",
      "Minibatch accuracy: 90.6%\n",
      "Validation accuracy: 90.9%\n",
      "Minibatch loss at step 17500: 0.276172\n",
      "Minibatch accuracy: 92.2%\n",
      "Validation accuracy: 90.8%\n",
      "Minibatch loss at step 18000: 0.252930\n",
      "Minibatch accuracy: 92.2%\n",
      "Validation accuracy: 90.7%\n",
      "Minibatch loss at step 18500: 0.410446\n",
      "Minibatch accuracy: 85.9%\n",
      "Validation accuracy: 91.0%\n",
      "Minibatch loss at step 19000: 0.169078\n",
      "Minibatch accuracy: 95.3%\n",
      "Validation accuracy: 90.9%\n",
      "Minibatch loss at step 19500: 0.343370\n",
      "Minibatch accuracy: 90.6%\n",
      "Validation accuracy: 90.9%\n",
      "Minibatch loss at step 20000: 0.318184\n",
      "Minibatch accuracy: 92.2%\n",
      "Validation accuracy: 91.3%\n",
      "Test accuracy: 96.3%\n"
     ]
    }
   ],
   "source": [
    "batch_size = 64\n",
    "patch_size = 5\n",
    "depth = 32\n",
    "num_hidden = 64\n",
    "\n",
    "graph = tf.Graph()\n",
    "\n",
    "with graph.as_default():\n",
    "\n",
    "  # Input data.\n",
    "  tf_train_dataset = tf.placeholder(\n",
    "    tf.float32, shape=(batch_size, image_size, image_size, num_channels))\n",
    "  tf_train_labels = tf.placeholder(tf.float32, shape=(batch_size, num_labels))\n",
    "  tf_valid_dataset = tf.constant(valid_dataset)\n",
    "  tf_test_dataset = tf.constant(test_dataset)\n",
    "  \n",
    "  # Variables.\n",
    "  layer1_weights = tf.Variable(tf.truncated_normal(\n",
    "      [patch_size, patch_size, num_channels, depth], stddev=np.sqrt(2.0 / (image_size * image_size))))\n",
    "  layer1_biases = tf.Variable(tf.zeros([depth]))\n",
    "  layer2_weights = tf.Variable(tf.truncated_normal(\n",
    "      [patch_size, patch_size, depth, depth], stddev=np.sqrt(2.0 / (image_size * image_size))))\n",
    "  layer2_biases = tf.Variable(tf.constant(1.0, shape=[depth]))\n",
    "  layer3_weights = tf.Variable(tf.truncated_normal(\n",
    "      [image_size // 4 * image_size // 4 * depth, num_hidden], stddev=np.sqrt(2.0 / (image_size * image_size))))\n",
    "  layer3_biases = tf.Variable(tf.constant(1.0, shape=[num_hidden]))\n",
    "  layer4_weights = tf.Variable(tf.truncated_normal(\n",
    "      [num_hidden, num_labels], stddev=np.sqrt(2.0 / (image_size * image_size))))\n",
    "  layer4_biases = tf.Variable(tf.constant(1.0, shape=[num_labels]))\n",
    "  \n",
    "  # Model.\n",
    "  def model(data, train=False):\n",
    "    conv = tf.nn.conv2d(data, layer1_weights, [1, 1, 1, 1], padding='SAME')\n",
    "    hidden = tf.nn.relu(conv + layer1_biases)\n",
    "    pool = tf.nn.avg_pool(hidden, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')\n",
    "    conv = tf.nn.conv2d(pool, layer2_weights, [1, 1, 1, 1], padding='SAME')\n",
    "    hidden = tf.nn.relu(conv + layer2_biases)\n",
    "    pool = tf.nn.avg_pool(hidden, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')\n",
    "    shape = pool.get_shape().as_list()\n",
    "    print(shape)\n",
    "    reshape = tf.reshape(pool, [shape[0], shape[1] * shape[2] * shape[3]])\n",
    "    hidden = tf.nn.relu(tf.matmul(reshape, layer3_weights) + layer3_biases)\n",
    "    if train:\n",
    "        hidden = tf.nn.dropout(hidden, keep_prob=0.3)\n",
    "    return tf.matmul(hidden, layer4_weights) + layer4_biases\n",
    "  \n",
    "  # Training computation.\n",
    "  logits = model(tf_train_dataset)\n",
    "  loss = tf.reduce_mean(\n",
    "    tf.nn.softmax_cross_entropy_with_logits(labels=tf_train_labels, logits=logits))\n",
    "    \n",
    "  # Optimizer.\n",
    "  optimizer = tf.train.GradientDescentOptimizer(0.05).minimize(loss)\n",
    "  \n",
    "  # Predictions for the training, validation, and test data.\n",
    "  train_prediction = tf.nn.softmax(logits)\n",
    "  valid_prediction = tf.nn.softmax(model(tf_valid_dataset))\n",
    "  test_prediction = tf.nn.softmax(model(tf_test_dataset))\n",
    "\n",
    "num_steps = 20001\n",
    "\n",
    "with tf.Session(graph=graph) as session:\n",
    "  tf.global_variables_initializer().run()\n",
    "  print('Initialized')\n",
    "  for step in range(num_steps):\n",
    "    offset = (step * batch_size) % (train_labels.shape[0] - batch_size)\n",
    "    batch_data = train_dataset[offset:(offset + batch_size), :, :, :]\n",
    "    batch_labels = train_labels[offset:(offset + batch_size), :]\n",
    "    feed_dict = {tf_train_dataset : batch_data, tf_train_labels : batch_labels}\n",
    "    _, l, predictions = session.run(\n",
    "      [optimizer, loss, train_prediction], feed_dict=feed_dict)\n",
    "    if (step % 500 == 0):\n",
    "      print('Minibatch loss at step %d: %f' % (step, l))\n",
    "      print('Minibatch accuracy: %.1f%%' % accuracy(predictions, batch_labels))\n",
    "      print('Validation accuracy: %.1f%%' % accuracy(\n",
    "        valid_prediction.eval(), valid_labels))\n",
    "  print('Test accuracy: %.1f%%' % accuracy(test_prediction.eval(), test_labels))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "colab": {
   "default_view": {},
   "name": "4_convolutions.ipynb",
   "provenance": [],
   "version": "0.3.2",
   "views": {}
  },
  "kernelspec": {
   "display_name": "Python [conda env:tensorflow]",
   "language": "python",
   "name": "conda-env-tensorflow-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}