softmax_regression完成mnist手写体数据集的识别

---恢复内容开始---

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])
y_ = tf.placeholder(tf.float32, [None, 10])

W = tf.Variable(tf.zeros([784, 10]))
b = tf.Variable(tf.zeros([10]))
y = tf.nn.softmax(tf.matmul(x, W) + b)

# 根据y, y_构造交叉熵损失
cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_ * tf.log(y)))

# 有了损失，我们就可以用随机梯度下降针对模型的参数（W和b）进行优化
train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cross_entropy)
correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
sess = tf.InteractiveSession() 

tf.global_variables_initializer().run() 

print('start training...') 

# 进行1000步梯度下降 

for i in range(5000):
　　batch = mnist.train.next_batch(50)# 默认 shuffle=True
　　if i%100==0:
　　　　train_accuracy = accuracy.eval({x:batch[0], y_:batch[1]})
　　　　print("step:%d accuracy is %g" %(i , train_accuracy) )

sess.run(train_step, feed_dict={x: batch[0], y_: batch[1]})

#print(sess.run(accuracy, feed_dict={x: mnist.test.images, y_: mnist.test.labels})) 

print(accuracy.eval(feed_dict={x:mnist.test.images,y_:mnist.test.labels}))

---恢复内容结束---