完成卷积神经网络的训练

with tf.Session() as sess:
    merged_summary_op = tf.summary.merge_all()
    summary_writer = tf.summary.FileWriter('/tmp/mnist_log/1', sess.graph)
    summary_writer.add_graph(sess.graph)

    sess.run(tf.global_variables_initializer())
    for i in range(20000):
        batch = data.train.next_batch(50)
        if i % 100 == 0:
            train_accuracy = accuracy.eval(feed_dict={x: batch[0], _y: batch[1], keep_prob: 1.0})
            print("step %d, train accuracy=%d" % (i, train_accuracy))
        sess.run(train_step, feed_dict={x: batch[0], _y: batch[1], keep_prob: 0.5})
        print(sess.run(accuracy, feed_dict={x: data.test.images, _y: data.test.labels, keep_prob: 1.0}))
        path = saver.save(sess, os.path.join(os.path.dirname(__file__), 'data', 'convalution.ckpt'),
                          write_meta_graph=False, write_state=False)
        print("Saved:", path)

1.在GPU上运行可能会稍微快一些比在CPU上。

2.    for i in range(20000):   # 对于这样的卷积训练一般要做10000-20000次的循环

        batch = data.train.next_batch(50)   # 定义batch的大小

        if i % 100 == 0:    # 每隔100次准确率做一次打印

            train_accuracy = accuracy.eval(feed_dict={x: batch[0], y_: batch[1], keep_prob: 1.0})

            print('step %d, training accuracy %g' % (i, train_accuracy))    # 打印

        sess.run(train_step, feed_dict={x: batch[0], y_: batch[1], keep_prob: 0.5})

    print(sess.run(accuracy, feed_dict={x: data.test.images, y_: data.test.labels, keep_prob: 1.0}))

    # 保存

    path = saver.save(

        sess, os.path.join(os.path.dirname(__file__), 'data', 'convolutional.ckpt'),

        write_meta_graph=False, write_state=False

    )

    print('Saved:', path)