TensorFlow实现简单的手写字体识别

#读取mnist数据集

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

#设置参数的大小
sess = tf.InteractiveSession()
x = tf.placeholder(tf.float32,[None,784])

W = tf.Variable(tf.zeros([784,10]))
b = tf.Variable(tf.zeros([10]))

y = tf.nn.softmax(tf.matmul(x,W)+b)
y_ = tf.placeholder(tf.float32,[None,10])

#loss function为cross_entropy
cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_*tf.log(y),reduction_indices = [1]))

train_step = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy)
init = tf.initialize_all_variables()
sess = tf.Session()
sess.run(init)
#训练
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})
#测试
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})