RNN(Recurrent NeuralNetwork)和LSTM(Long Short Term Memory)

 

RNN(Recurrent NeuralNetwork)

04-循环神经网络(RNN)和LSTM

04-循环神经网络(RNN)和LSTM

04-循环神经网络(RNN)和LSTM

 

RNN:存在随着时间的流逝,信号会不断的衰弱(梯度消失)

 

LSTM(Long Short Term Memory):

很好的解决梯度消失

控制信号的衰减

控制信号输出信号本身值的百分之多少

04-循环神经网络(RNN)和LSTM

 04-循环神经网络(RNN)和LSTM

04-循环神经网络(RNN)和LSTM

 

 

只有时间1的的信号可以被传入

只有时间4和6的信号可以被输出

 

 

练习:

04-循环神经网络(RNN)和LSTM

 

循环(递归)神经网络RNN代码

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

#载入数据集
mnist =input_data.read_data_sets("MNIST_data/",one_hot=True)

#输入层
#
输入图片是28*28
n_inputs = 28 #输入一行,一行有28个数据
max_time = 28 #一共28
lstm_size = 100 #隐层单元
n_classes = 10 # 10个分类
batch_size = 50 #每批次50个样本
n_batch =mnist.train.num_examples // batch_size #计算一共有多少个批次

#这里的none表示第一个维度可以是任意的长度
x =tf.placeholder(tf.float32,[None,784])
#正确的标签
y =tf.placeholder(tf.float32,[None,10])

#初始化权值
weights = tf.Variable(tf.truncated_normal([lstm_size, n_classes], stddev=0.1))
#初始化偏置值
biases =tf.Variable(tf.constant(0.1, shape=[n_classes]))

#定义RNN网络
def RNN(X,weights,biases):
   
# inputs=[batch_size, max_time, n_inputs]
   
inputs = tf.reshape(X,[-1,max_time,n_inputs])
   
#定义LSTM基本CELL
    #lstm_cell =tf.contrib.rnn.core_rnn_cell.BasicLSTMCell(lstm_size)
    lstm_cell =tf.contrib.rnn.core_rnn_cell.BasicLSTMCell(lstm_size)
   
# final_state[0]是cell state
    # final_state[1]
是hidden_state
    outputs,final_state = tf.nn.dynamic_rnn(lstm_cell,inputs,dtype=tf.float32)
    results =tf.nn.softmax(tf.matmul(final_state[
1],weights)+ biases)
   
return results
   
   
#计算RNN的返回结果
prediction= RNN(x, weights, biases) 
#损失函数
cross_entropy =tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=prediction,labels=y))
#使用AdamOptimizer进行优化
train_step =tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
#结果存放在一个布尔型列表中
correct_prediction= tf.equal(tf.argmax(y,1),tf.argmax(prediction,1))#argmax返回一维张量中最大的值所在的位置
#
求准确率
accuracy =tf.reduce_mean(tf.cast(correct_prediction,tf.float32))#把correct_prediction变为float32类型
#初始化
init =tf.global_variables_initializer()

with tf.Session() as sess:
    sess.run(init)
   
for epochin range(6):
       
for batchin range(n_batch):
            batch_xs
,batch_ys = mnist.train.next_batch(batch_size)
            sess.run(train_step
,feed_dict={x:batch_xs,y:batch_ys})
       
        acc = sess.run(accuracy
,feed_dict={x:mnist.test.images,y:mnist.test.labels})
        
print ("Iter " + str(epoch) + ", Testing Accuracy= " + str(acc))