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}))