【转载】 tensorflow的单层静态与动态RNN比较

原文地址：

https://www.jianshu.com/p/1b1ea45fab47

yanghedada

-----------------------------------------------------------------------------------

static_rnn和dynamic_rnn

1:     static_rnn

x = tf.placeholder("float", [None, n_steps, n_input])
x1 = tf.unstack(x, n_steps, 1)
lstm_cell = tf.contrib.rnn.BasicLSTMCell(n_hidden, forget_bias=1.0)
outputs, states = tf.contrib.rnn.static_rnn(lstm_cell, x1, dtype=tf.float32)
pred = tf.contrib.layers.fully_connected(outputs[-1],n_classes,activation_fn = None)

2:     dynamic_rnn

x = tf.placeholder("float", [None, n_steps, n_input])
lstm_cell = tf.contrib.rnn.BasicLSTMCell(n_hidden, forget_bias=1.0)
outputs,_  = tf.nn.dynamic_rnn(lstm_cell ,x,dtype=tf.float32)
outputs = tf.transpose(outputs, [1, 0, 2])
pred = tf.contrib.layers.fully_connected(outputs[-1],n_classes,activation_fn = None)

BasicLSTMCell：
（num_units: 是指一个Cell中神经元的个数,forget_bias:忘记门记住多少，1.0代表全部记住）

 

 

tf.contrib.rnn.static_rnn：
静态 rnn的意思就是按照样本时间序列个数（n_steps）展开，在图中创建（n_steps）个序列的cell

 

 

tf.nn.dynamic_rnn：
动态rnn的意思是只创建样本中的一个序列RNN，其他序列数据会通过循环进入该RNN运算。  通过静态static_rnn生成的RNN网络，生成过程所需的时间会更长，网络所占有的内存会更多，导出的模型会更大。static_rnn模型中会带有第个序列中间态的信息，利于调试。static_rnn在使用时必须与训练的样本序列个数相同。dynamic_rnn通过动态生成的RNN网络，所占用内存较少。dynamic_rnn模型中只会有最后的状态，在使用时还能支持不同的序列个数。

 

 

 

区别

1.tf.nn.dynamic_rnn与tf.contrib.rnn.static_rnn输入格式不同。
2.tf.nn.dynamic_rnn与tf.contrib.rnn.static_rnn输出格式不同。
3.tf.nn.dynamic_rnn与tf.contrib.rnn.static_rnn内部训练方式。

 

 

 

请仔细对比以下区别：

可以参考：https://blog.csdn.net/mzpmzk/article/details/80573338

 

 

 

 

 

 

 

 

 

动态rnn

import tensorflow as tf
# 导入 MINST 数据集
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("c:/user/administrator/data/", one_hot=True)
n_input = 28 # MNIST data 输入 (img shape: 28*28)
n_steps = 28 # timesteps
n_hidden = 128 # hidden layer num of features
n_classes = 10  # MNIST 列别 (0-9 ，一共10类)
batch_size = 128
tf.reset_default_graph()

# tf Graph input
x = tf.placeholder("float", [None, n_steps, n_input])
y = tf.placeholder("float", [None, n_classes])
lstm_cell = tf.contrib.rnn.LSTMCell(n_hidden, forget_bias=1.0)
outputs,_  = tf.nn.dynamic_rnn(lstm_cell,x,dtype=tf.float32)
outputs = tf.transpose(outputs, [1, 0, 2])
#取最后一条输出信息，(outputs[-1]）
pred = tf.contrib.layers.fully_connected(outputs[-1],n_classes,activation_fn = None)

learning_rate = 0.001
training_iters = 100000

display_step = 10

# Define loss and optimizer
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(logits=pred, labels=y))
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)

# Evaluate model
correct_pred = tf.equal(tf.argmax(pred,1), tf.argmax(y,1))
accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))

# 启动session
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    step = 1
    # Keep training until reach max iterations
    while step * batch_size < training_iters:
        batch_x, batch_y = mnist.train.next_batch(batch_size)
        # Reshape data to get 28 seq of 28 elements
        batch_x = batch_x.reshape((batch_size, n_steps, n_input))
        # Run optimization op (backprop)
        sess.run(optimizer, feed_dict={x: batch_x, y: batch_y})
        if step % display_step == 0:
            # 计算批次数据的准确率
            acc = sess.run(accuracy, feed_dict={x: batch_x, y: batch_y})
            # Calculate batch loss
            loss = sess.run(cost, feed_dict={x: batch_x, y: batch_y})
            print ("Iter " + str(step*batch_size) + ", Minibatch Loss= " + 
                  "{:.6f}".format(loss) + ", Training Accuracy= " + 
                  "{:.5f}".format(acc))
        step += 1
    print (" Finished!")

    # 计算准确率 for 128 mnist test images
    test_len = 128
    test_data = mnist.test.images[:test_len].reshape((-1, n_steps, n_input))
    test_label = mnist.test.labels[:test_len]
    print ("Testing Accuracy:", 
        sess.run(accuracy, feed_dict={x: test_data, y: test_label}))

静态RNN

import tensorflow as tf
# 导入 MINST 数据集
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("c:/user/administrator/data/", one_hot=True)
n_input = 28 # MNIST data 输入 (img shape: 28*28)
n_steps = 28 # timesteps
n_hidden = 128 # hidden layer num of features
n_classes = 10  # MNIST 列别 (0-9 ，一共10类)
batch_size = 128
tf.reset_default_graph()

# tf Graph input
x = tf.placeholder("float", [None, n_steps, n_input])
y = tf.placeholder("float", [None, n_classes])
lstm_cell = tf.contrib.rnn.LSTMCell(n_hidden, forget_bias=1.0)
x1 = tf.unstack(x, n_steps, 1)
lstm_cell = tf.contrib.rnn.LSTMCell(n_hidden, forget_bias=1.0)
outputs, states = tf.contrib.rnn.static_rnn(lstm_cell, x1, dtype=tf.float32)
#取最后一条输出信息，(outputs[-1]）
pred = tf.contrib.layers.fully_connected(outputs[-1],n_classes,activation_fn = None)

learning_rate = 0.001
training_iters = 100000

display_step = 10

# Define loss and optimizer
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(logits=pred, labels=y))
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)

# Evaluate model
correct_pred = tf.equal(tf.argmax(pred,1), tf.argmax(y,1))
accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))

# 启动session
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    step = 1
    # Keep training until reach max iterations
    while step * batch_size < training_iters:
        batch_x, batch_y = mnist.train.next_batch(batch_size)
        # Reshape data to get 28 seq of 28 elements
        batch_x = batch_x.reshape((batch_size, n_steps, n_input))
        # Run optimization op (backprop)
        sess.run(optimizer, feed_dict={x: batch_x, y: batch_y})
        if step % display_step == 0:
            # 计算批次数据的准确率
            acc = sess.run(accuracy, feed_dict={x: batch_x, y: batch_y})
            # Calculate batch loss
            loss = sess.run(cost, feed_dict={x: batch_x, y: batch_y})
            print ("Iter " + str(step*batch_size) + ", Minibatch Loss= " + 
                  "{:.6f}".format(loss) + ", Training Accuracy= " + 
                  "{:.5f}".format(acc))
        step += 1
    print (" Finished!")

    # 计算准确率 for 128 mnist test images
    test_len = 128
    test_data = mnist.test.images[:test_len].reshape((-1, n_steps, n_input))
    test_label = mnist.test.labels[:test_len]
    print ("Testing Accuracy:", 
        sess.run(accuracy, feed_dict={x: test_data, y: test_label}))

本代码源自：
凯文自学TensorFlow

# -*- coding: utf-8 -*-


import tensorflow as tf
# 导入 MINST 数据集
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("c:/user/administrator/data/", one_hot=True)
n_input = 28 # MNIST data 输入 (img shape: 28*28)
n_steps = 28 # timesteps
n_hidden = 128 # hidden layer num of features
n_classes = 10  # MNIST 列别 (0-9 ，一共10类)
batch_size = 128
tf.reset_default_graph()

# tf Graph input
x = tf.placeholder("float", [None, n_steps, n_input])
y = tf.placeholder("float", [None, n_classes])
#重置x以适合tf.contrib.rnn.static_rnn所要求的格式
#x1 = tf.unstack(x, n_steps, 1)

#BasicLSTMCell（num_units: 是指一个Cell中神经元的个数,forget_bias:忘记门记住多少，1.0代表全部记住）
#静态 （tf.contrib.rnn.static_rnn）的意思就是按照样本时间序列个数（n_steps）展开，在图中创建（n_steps）个序列的cell;
#动态（tf.nn.dynamic_rnn）的意思是只创建样本中的一个序列RNN，其他序列数据会通过循环进入该RNN运算
"""
通过静态生成的RNN网络，生成过程所需的时间会更长，网络所占有的内存会更多，导出的模型会更大
。模型中会带有第个序列中间态的信息，利于调试。在使用时必须与训练的样本序列个数相同。通过动
态生成的RNN网络，所占用内存较少。模型中只会有最后的状态，在使用时还能支持不同的序列个数。
"""
#lstm_cell = tf.contrib.rnn.BasicLSTMCell(n_hidden, forget_bias=1.0)
#outputs, states = tf.contrib.rnn.static_rnn(lstm_cell, x1, dtype=tf.float32)
"""
#2 LSTMCell，LSTM实现的一个高级版本（use_peepholes:默认False,True表示启用peephole连接）
cell_clip:是否在输出前对cell状态按照给定值进行截断处理
initializer:指定初始化函数
num_proj:通过projection进行模型压缩的输出维度
proj_clip：将num_proj按照给定的proj_clip截断
"""
#lstm_cell = tf.contrib.rnn.LSTMCell(n_hidden, forget_bias=1.0)
#outputs, states = tf.contrib.rnn.static_rnn(lstm_cell, x1, dtype=tf.float32)

#3 gru类定义
#gru = tf.contrib.rnn.GRUCell(n_hidden)
#outputs = tf.contrib.rnn.static_rnn(gru, x1, dtype=tf.float32)

#4 创建动态RNN，此时的输入是x，是动态的[None, n_steps, n_input]LIST
#具体定义参考https://blog.csdn.net/mzpmzk/article/details/80573338
gru = tf.contrib.rnn.GRUCell(n_hidden)
outputs,_  = tf.nn.dynamic_rnn(gru,x,dtype=tf.float32)
outputs = tf.transpose(outputs, [1, 0, 2])
#取最后一条输出信息，(outputs[-1]）
pred = tf.contrib.layers.fully_connected(outputs[-1],n_classes,activation_fn = None)



learning_rate = 0.001
training_iters = 100000

display_step = 10

# Define loss and optimizer
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(logits=pred, labels=y))
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)

# Evaluate model
correct_pred = tf.equal(tf.argmax(pred,1), tf.argmax(y,1))
accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))

# 启动session
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    step = 1
    # Keep training until reach max iterations
    while step * batch_size < training_iters:
        batch_x, batch_y = mnist.train.next_batch(batch_size)
        # Reshape data to get 28 seq of 28 elements
        batch_x = batch_x.reshape((batch_size, n_steps, n_input))
        # Run optimization op (backprop)
        sess.run(optimizer, feed_dict={x: batch_x, y: batch_y})
        if step % display_step == 0:
            # 计算批次数据的准确率
            acc = sess.run(accuracy, feed_dict={x: batch_x, y: batch_y})
            # Calculate batch loss
            loss = sess.run(cost, feed_dict={x: batch_x, y: batch_y})
            print ("Iter " + str(step*batch_size) + ", Minibatch Loss= " + 
                  "{:.6f}".format(loss) + ", Training Accuracy= " + 
                  "{:.5f}".format(acc))
        step += 1
    print (" Finished!")

    # 计算准确率 for 128 mnist test images
    test_len = 128
    test_data = mnist.test.images[:test_len].reshape((-1, n_steps, n_input))
    test_label = mnist.test.labels[:test_len]
    print ("Testing Accuracy:", 
        sess.run(accuracy, feed_dict={x: test_data, y: test_label}))

-----------------------------------------------------------------------------------