In [1]:
# import modules
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data

In [2]:
# load the input data
mnist = input_data.read_data_sets('MNIST_data', one_hot=True)


Extracting MNIST_data/train-images-idx3-ubyte.gz
Extracting MNIST_data/train-labels-idx1-ubyte.gz
Extracting MNIST_data/t10k-images-idx3-ubyte.gz
Extracting MNIST_data/t10k-labels-idx1-ubyte.gz

In [3]:
# setup an interactive session required for working in ipython
sess = tf.InteractiveSession()

# design of a softmax regression model

# place holders for the data
x = tf.placeholder(tf.float32, shape=[None, 784])
y_ = tf.placeholder(tf.float32, shape=[None, 10])

# create variables for regression model
W = tf.Variable(tf.zeros([784,10]))
b = tf.Variable(tf.zeros([10]))

# it is important to initialize the variables before they can be used in a session
sess.run(tf.initialize_all_variables())

# implement output of regression model
y = tf.matmul(x,W) + b

# use cross-entropy as the loss function for training
cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(y, y_))

# use gradient descent to optimize the parameters such that the loss is minimized
train_step = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy)

# this train step is run for some iterations
for i in range(1000):
    batch = mnist.train.next_batch(100)
    train_step.run(feed_dict={x:batch[0], y_:batch[1]})
    if i%100 == 0:
        print 'Running'
        
# check the accuracy of the model
correct_prediction = tf.equal(tf.argmax(y,1), tf.argmax(y_,1))

# compute accuracy from all predictions
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
print accuracy.eval(feed_dict={x: mnist.test.images, y_:mnist.test.labels})


Running
Running
Running
Running
Running
Running
Running
Running
Running
Running
0.9209

In [ ]:
# create a session
sess = tf.InteractiveSession()

# train convolutional neural network model

# functions to initialize weight parameters
def weight_variable(shape):
    initial = tf.truncated_normal(shape, stddev=0.1)
    return tf.Variable(initial)

def bias_variable(shape):
    initial = tf.constant(0.1, shape=shape)
    return tf.Variable(initial)

# functions to perform convolutions and max pooling
def conv_2d(x,W):
    return tf.nn.conv2d(x, W, strides=[1,1,1,1], padding='SAME')

def max_pool_2x2(x):
    return tf.nn.max_pool(x, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')

# reshape image variable
x = tf.placeholder(tf.float32, shape=[None, 784])
x_image = tf.reshape(x, [-1,28,28,1])

# variable for true labels
y_ = tf.placeholder(tf.float32, shape=[None, 10])

# create the convolutional layers
W_conv1 = weight_variable([5,5,1,32])
b_conv1 = bias_variable([32])

W_conv2 = weight_variable([5,5,32,64])
b_conv2 = bias_variable([64])

# apply convolutional layers
h_conv1 = tf.nn.relu(conv_2d(x_image, W_conv1)+b_conv1)
h_pool1 = max_pool_2x2(h_conv1)

h_conv2 = tf.nn.relu(conv_2d(h_pool1, W_conv2)+b_conv2)
h_pool2 = max_pool_2x2(h_conv2)

# create a fully connected layer
W_fc1 = weight_variable([7*7*64, 1024])
b_fc1 = bias_variable([1024])

# apply fully connected layer
h_pool2_flat = tf.reshape(h_pool2, [-1,7*7*64])
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1)+b_fc1)

# apply dropout before computing softmax probabilities
keep_prob = tf.placeholder(tf.float32)
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob=keep_prob)

# obtain output
W_fc2 = weight_variable([1024,10])
b_fc2 = bias_variable([10])

y_conv = tf.matmul(h_fc1_drop, W_fc2)+b_fc2

# evaluate loss function
cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(y_conv, y_))

# create optimizer
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)

# formulation for computing accuracy
correct_prediction = tf.equal(tf.argmax(y_conv,1),tf.argmax(y_,1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

# initialize all variables for the session
sess.run(tf.initialize_all_variables())

# train the model
for i in range(20000):
    batch = mnist.train.next_batch(50)
    
    if i%100 == 0:
        train_accuracy = accuracy.eval(feed_dict={x:batch[0], y_:batch[1], keep_prob:1.0})
        print "Step %d: Accuracy %g"%(i,train_accuracy)
        
    train_step.run(feed_dict={x:batch[0], y_:batch[1], keep_prob:0.5})
    
# obtain the test accuracy
print "Test accuracy %g" % accuracy.eval(feed_dict={x: mnist.test.images, 
                                                    y_: mnist.test.labels, 
                                                    keep_prob: 1.0})


Step 0: Accuracy 0.06
Step 100: Accuracy 0.8
Step 200: Accuracy 0.88
Step 300: Accuracy 0.96
Step 400: Accuracy 1
Step 500: Accuracy 0.96
Step 600: Accuracy 0.96
Step 700: Accuracy 1
Step 800: Accuracy 0.94
Step 900: Accuracy 0.86
Step 1000: Accuracy 0.98
Step 1100: Accuracy 1
Step 1200: Accuracy 0.98
Step 1300: Accuracy 0.98
Step 1400: Accuracy 0.94
Step 1500: Accuracy 0.98
Step 1600: Accuracy 0.98
Step 1700: Accuracy 1

In [ ]: