基于tensorflow数据集的dropout优化

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

#导入数据
mnist = input_data.read_data_sets(“MNIST_data”,one_hot=True)

batch_size = 100 #一次20
n_batch = mnist.train.num_examples //batch_size #次数

x = tf.placeholder(tf.float32,[None,784]) #60000，28*28
y = tf.placeholder(tf.float32,[None,10]) #0-9
keep_prob = tf.placeholder(tf.float32) #keep_prob神经元使用率
lr = tf.Variable(0.001,dtype=tf.float32)

#输入层
w1 = tf.Variable(tf.truncated_normal(([784,800]),stddev = 0.1)) #截断正态分布，stddev是标准差
b1 = tf.Variable(tf.zeros([800])+0.1)
L1 = tf.nn.sigmoid((tf.matmul(x,w1)+b1))
L1 = tf.nn.tanh((tf.matmul(x,w1)+b1))
L1_dropout = tf.nn.dropout(L1,keep_prob)

#中间层
w2 = tf.Variable(tf.truncated_normal(([800,400]),stddev = 0.1))
b2 = tf.Variable(tf.zeros([400])+0.1)
L2 = tf.nn.sigmoid((tf.matmul(L1_dropout,w2)+b2))
#L2 = tf.nn.tanh((tf.matmul(L1_dropout,w2)+b2))
L2_dropout = tf.nn.dropout(L2,keep_prob) #防止或减轻过拟合 （输入tensor，keep_prob: float类型，每个元素被保留下来的概率，设置神经元被选中的概率,在初始化时keep_prob是一个占位符, ）

#输出层
w3 = tf.Variable(tf.truncated_normal(([400,10]),stddev=0.1))
b3 = tf.Variable(tf.zeros([10])+0.1)
prediction = tf.matmul(L2_dropout,w3)+b3

#loss = tf.reduce_mean(tf.square(y-prediction)) 下面这个训练效果好
loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y,logits=prediction))

#labels：为神经网络期望的输出，logits：为神经网络最后一层的输出
#这个函数内部自动计算softmax，然后再计算交叉熵代价函数，
#也就是说logits必须是没有经过tf.nn.softmax函数处理的数据，
否则导致训练结果有问题。建议编程序时使用这个函数，而不必自己编写交叉熵代价函数。

train_step = tf.train.GradientDescentOptimizer(0.1).minimize(loss)
train_step = tf.train.AdamOptimizer(lr).minimize(loss)

#equal()返回布尔类型的列表，argmax(参数，0表示列，1表示行) 返回一维张量(1维向量)中最大值所在位置的索引
#y是10维向量，prediction2也是10维向量，所以correct_prediction是包含10个True和False的列表
correct_prediction = tf.equal(tf.argmax(y,1),tf.argmax(prediction,1))

#准确率即由0，1.0求和在求平均值
accurary = tf.reduce_mean(tf.cast(correct_prediction,tf.float32))

with tf.Session() as sess:
sess.run(tf.global_variables_initializer())

for step in range(31):
    sess.run(tf.assign(lr, 0.001*(0.95 ** step)))
    for batch in range(n_batch):
        # 图片数据保存在batch_xs，图片标签保存在batch_ys
        batch_xs,batch_ys = mnist.train.next_batch(batch_size)
        sess.run(train_step,feed_dict={x:batch_xs,y:batch_ys,keep_prob:1.0})

    learning_rate = sess.run(lr)
    Test_acc = sess.run(accurary,feed_dict={x:mnist.test.images,y:mnist.test.labels,keep_prob:1.0})
    Train_acc = sess.run(accurary, feed_dict={x: mnist.train.images, y: mnist.train.labels, keep_prob: 1.0})
    print("Iter "+str(step)+" Test_acc "+str(Test_acc) + " Train_acc " + str(Train_acc) + " 学习率 "+str(learning_rate))

#Iter 30 Test_acc 0.9819 Train_acc 0.9961636
#Iter 30 Test_acc 0.9819 Train_acc 1.0 学习率 0.00021463877