mnist.py

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data

#load dataset
mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)

#build modle arg
x = tf.placeholder("float", [None, 784])
W = tf.Variable(tf.zeros([784,10]))
b = tf.Variable(tf.zeros([10]))
y = tf.nn.softmax(tf.matmul(x,W) + b)

#built loss
y_ = tf.placeholder("float", [None,10])
cross_entropy = -tf.reduce_sum(y_*tf.log(y))

#choose algorithm
train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cross_entropy)

#start training
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)
  #run modle via train_step
  sess.run(train_step, feed_dict={x: batch_xs, y_: batch_ys})

#test modle
correct_prediction = tf.equal(tf.argmax(y,1), tf.argmax(y_,1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
print (sess.run(accuracy, feed_dict={x: mnist.test.images, y_: mnist.test.labels}))