1 import tensorflow as tf
 2 from tensorflow.examples.tutorials.mnist import input_data
 3 
 4 mnist=input_data.read_data_sets("MNIST_data/",one_hot=True)
 5 
 6 # print(mnist.train.images.shape,mnist.train.labels.shape)
 7 # print(mnist.test.images.shape,mnist.test.labels.shape)
 8 # print(mnist.validation.images.shape,mnist.validation.labels.shape)
 

 9 
10 sess=tf.InteractiveSession()
11 x=tf.placeholder(tf.float32,[None,784])
12 
13 W=tf.Variable(tf.zeros([784,10]))
14 b=tf.Variable(tf.zeros([10]))
15 
16 y=tf.nn.softmax(tf.matmul(x,W)+b)
17 
18 y_=tf.placeholder(tf.float32,[None,10])
19 cross_entropy=tf.reduce_mean(-tf.reduce_sum(y_*tf.log(y),reduction_indices=[1]))
 
20 
21 train_step=tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy)
22 tf.initialize_all_variables().run()
23 
24 for i in range(1000):
25     batch_xs,batch_ys=mnist.train.next_batch(100)
26     train_step.run({x:batch_xs,y_:batch_ys})
27 
28 correct_prediction=tf.equal(tf.arg_max(y,1),tf.arg_max(y_,1))
 
29 accuracy=tf.reduce_mean(tf.cast(correct_prediction,tf.float32))
30 
31 print(accuracy.eval({x:mnist.test.images,y_:mnist.test.labels}))

TensorFlow入門測試程序