【转载】 【TensorFlow】static_rnn 和dynamic_rnn的区别

原文地址：

https://blog.csdn.net/qq_20135597/article/details/88980975

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

tensorflow中提供了rnn接口有两种，一种是静态的rnn，一种是动态的rnn

通常用法：

1、静态接口：static_rnn

主要使用 tf.contrib.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)

静态 rnn 的意思就是在图中创建一个固定长度（n_steps）的网络。这将导致

缺点：

1. 生成过程耗时更长，占内存更多，导出的模型更大；
2. 无法传递比最初指定的更长的序列（> n_steps）。

优点：

模型中带有某个序列中间台的信息，便与调试。

2、动态接口：dynamic_rnn

主要使用 tf.nn.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)

动态的tf.nn.dynamic_rnn被执行时，它使用循环来动态构建图形。这意味着

优点：

1. 图形创建速度更快，占用内存更少；
2. 并且可以提供可变大小的批处理。

缺点：

1. 模型中只有最后的状态。

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

区别：

1、输入输出不同：

dynamic_rnn实现的功能就是可以让不同迭代传入的batch可以是长度不同数据，但同一次迭代一个batch内部的所有数据长度仍然是固定的。例如，第一时刻传入的数据shape=[batch_size, 10]，第二时刻传入的数据shape=[batch_size, 12]，第三时刻传入的数据shape=[batch_size, 8]等等。

但是static_rnn不能这样，它要求每一时刻传入的batch数据的[batch_size, max_seq]，在每次迭代过程中都保持不变。

2、训练方式不同：

具体参见参考文献1

多层LSTM的代码实现对比：

1、静态多层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])

gru = tf.contrib.rnn.GRUCell(n_hidden*2)
lstm_cell = tf.contrib.rnn.LSTMCell(n_hidden)
mcell = tf.contrib.rnn.MultiRNNCell([lstm_cell,gru])

x1 = tf.unstack(x, n_steps, 1)
outputs, states = tf.contrib.rnn.static_rnn(mcell, x1, dtype=tf.float32)

pred = tf.contrib.layers.fully_connected(outputs[-1],n_classes,activation_fn = None)

learning_rate = 0.001
# 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))

training_iters = 100000

display_step = 10

# 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 = 100
    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}))

2、动态多层RNN

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])

gru = tf.contrib.rnn.GRUCell(n_hidden*2)
lstm_cell = tf.contrib.rnn.LSTMCell(n_hidden)
mcell = tf.contrib.rnn.MultiRNNCell([lstm_cell,gru])

outputs,states  = tf.nn.dynamic_rnn(mcell,x,dtype=tf.float32)#(?, 28, 256)
outputs = tf.transpose(outputs, [1, 0, 2])#(28, ?, 256) 28个时序，取最后一个时序outputs[-1]=(?,256)

pred = tf.contrib.layers.fully_connected(outputs[-1],n_classes,activation_fn = None)

learning_rate = 0.001
# 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))

training_iters = 100000

display_step = 10

# 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 = 100
    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}))

【参考文献】：

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

2、What's the difference between tensorflow dynamic_rnn and rnn?

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