tensorflow实现循环神经网络

      循环神经网络出现于20世纪80年代，在其发展早期，应用部署特别丰富。最近几年由于神经网络结构的进步和GPU上深度学习训练效率的突破，RNN变得越来越流行。RNN对时间序列数据非常有效，其每个神经元可通过内部组件保存之前输入的信息。

     人每次思考时不会重头开始，而是保留之前思考的一些结果为现在的决策提供支持。

     下面我们将RNN用在手写数字识别上。

# tensorflow 实现递归神经网络

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
from tensorflow.contrib import rnn
mnist = input_data.read_data_sets("MNIST_data", one_hot=True)
# 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行
lstm_size = 200#隐层单元
n_class = 10#分类个数
batch_size = 50#每个批次样本大小
n_batch = mnist.train.num_examples // batch_size #批次个数

x = tf.placeholder(tf.float32, [None, 784])
y = tf.placeholder(tf.float32, [None, 10])

#初始化权值
weights = tf.Variable(tf.truncated_normal([lstm_size, n_class], stddev = 0.1))
biase = tf.Variable(tf.constant(0.1, shape=[n_class]))

#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 = rnn.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

def LSTM(X, weights, biase):
    #inputs format : [batch_size, max_time, n_inputs]
    inputs = tf.reshape(X, [-1, max_time, n_inputs])
    #定义LSTM基本cell
    lstm_cell = rnn.BasicLSTMCell(lstm_size)
    outputs, final_state = tf.nn.dynamic_rnn(lstm_cell, inputs, dtype=tf.float32)
    results = tf.nn.softmax(tf.matmul(final_state[1], weights) + biase)
    return results

#返回结果
prediction = LSTM(x, weights, biase) # RNN
#损失函数
cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y, logits=prediction))
#优化器
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
#计算准确率
correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(prediction, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

init = tf.global_variables_initializer()

with tf.Session() as sess:
    sess.run(init)
    for epoch in range(61):
        for batch in range(batch_size):
            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))

最终迭代结果为：

Iter 54, Testing Accuracy= 0.8411
Iter 55, Testing Accuracy= 0.8555
Iter 56, Testing Accuracy= 0.8968
Iter 57, Testing Accuracy= 0.9022
Iter 58, Testing Accuracy= 0.9109
Iter 59, Testing Accuracy= 0.907
Iter 60, Testing Accuracy= 0.9185