实战1-DNN训练MNIST

实战1-DNN训练MNIST

这里放一个其他的设计方法

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data

data_dir = 'mnist/input_data'
mnist = input_data.read_data_sets(data_dir,one_hot=True)


def save_model(sess):
    Saver = tf.train.Saver()
    Saver.save(sess,'save/save_net.ckpt')

#创建模型
def mnist_train(middle_size1 = 400,learning_rate = 0.65,mini_batch = 100):

    x = tf.placeholder(tf.float32,[None,784])
    y = tf.placeholder(tf.float32,[None,10])
    W = tf.Variable(tf.truncated_normal([784,middle_size1],stddev=0.1))     #高斯初始化，stddev的值也很重要
    b = tf.Variable(tf.zeros([middle_size1]))
    layer1 = tf.nn.leaky_relu(tf.matmul(x,W) + b)       #先尝试relu，考虑到神经元死亡的情况，可以使用leaky_relu
    #layer1 = tf.nn.dropout(layer1,0.5) #防止过拟合

    W2 = tf.Variable(tf.truncated_normal([middle_size1,10],stddev=0.1))
    b2 = tf.Variable(tf.zeros([10]))

    pred = tf.matmul(layer1,W2) + b2
    cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y,logits=pred))   #计算交叉熵

    train_step = tf.train.GradientDescentOptimizer(learning_rate).minimize(cross_entropy)       #最小化交叉熵，反向传播，

    init = tf.global_variables_initializer()
    with tf.Session() as sess:
        sess.run(init)
        accuracy_ = 0
        for _ in range(10000):
            batch_xs,batch_ys = mnist.train.next_batch(mini_batch)
            sess.run(train_step,feed_dict={x:batch_xs,y:batch_ys})
            if _ % 100 == 0 :
                correct_prediction = tf.equal(tf.argmax(pred,1),tf.argmax(y,1))
                accuracy = tf.reduce_mean(tf.cast(correct_prediction,tf.float32))
                temp = sess.run(accuracy,feed_dict={x:mnist.test.images,y:mnist.test.labels})
                if temp > accuracy_:
                    print(_,'accuracy_rate:',temp)
                    accuracy_= temp


        X_, Y_ = mnist.train.images, mnist.train.labels
        pred_test = tf.nn.leaky_relu(tf.matmul(tf.nn.leaky_relu(tf.matmul(X_, W) + b), W2) + b2)
        accuracy_rate = tf.reduce_mean(tf.cast(tf.equal(tf.argmax(pred_test, 1), tf.argmax(Y_, 1)), tf.float32))
        print('accuracy_rate in train:', sess.run(accuracy_rate))
        X_, Y_ = mnist.test.images,mnist.test.labels
        pred_test = tf.nn.leaky_relu(tf.matmul(tf.nn.leaky_relu(tf.matmul(X_, W) + b), W2) + b2)
        accuracy_rate = tf.reduce_mean(tf.cast(tf.equal(tf.argmax(pred_test, 1), tf.argmax(Y_, 1)), tf.float32))
        print('accuracy_rate in test:',sess.run(accuracy_rate))
        if sess.run(accuracy_rate) > 0.982:
            save_model(sess)

mnist_train(middle_size1 = 400,learning_rate = 0.65,mini_batch = 100)