Handwritten digit analysis

tensorflow.py

import tensorflow as tf

from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)

# 784 pixels (28 x 28 pixel images), 10 possibile outputs
x = tf.placeholder(tf.float32, [None, 784]) # placeholder
W = tf.Variable(tf.zeros([784, 10])) # weights
b = tf.Variable(tf.zeros([10])) # biases

# y is the predicted probability distribution
y = tf.nn.softmax(tf.matmul(x, W) + b)

# y prime is the true distribution
y_ = tf.placeholder(tf.float32, [None, 10])

# Use cross-entropy for the cost function
cross_entropy = -tf.reduce_sum(y_*tf.log(y))

# Use gradient descent on the cost function for training
train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cross_entropy)

init = tf.initialize_all_variables()
sess = tf.Session()
sess.run(init)

for i in range(1000):
    batch_xs, batch_ys = mnist.train.next_batch(100)
    sess.run(train_step, feed_dict={x: batch_xs, y_: batch_ys})

# Measure the accuracy of the algorithm
correct_prediction = tf.equal(tf.argmax(y,1), tf.argmax(y_,1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
print(sess.run(accuracy, feed_dict={x: mnist.test.images, y_: mnist.test.labels}))