MNIST tensorflow

原始地址

import tensorflow as tf
import tensorflow.examples.tutorials.mnist.input_data as input_data
mnist=input_data.read_data_sets("MNIST_data/",one_hot=True)
print (mnist.train.images.shape)
print (mnist.train.labels.shape)
print (mnist.validation.images.shape)
print (mnist.validation.labels.shape)
print (mnist.test.images.shape)
print (mnist.test.labels.shape)

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(0.01).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}))