MNIST的简单实现

MNIST的第一次简单实现

import tensorflow as tf

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


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('float',[None,10])

cross_entropy=-tf.reduce_sum(y_*tf.log(y))

train_step=tf.train.GradientDescentOptimizer(0.001).minimize(cross_entropy)


init=tf.global_variables_initializer()

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})

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}))