tf.contrib.rnn.static_rnn与tf.nn.dynamic_rnn区别

你要的答案或许都在这里：小鹏的博客目录

MachineLP的Github（欢迎follow）：https://github.com/MachineLP

我的GitHub：https://github.com/MachineLP/train_cnn-rnn-attention 自己搭建的一个框架，包含模型有：vgg(vgg16,vgg19), resnet(resnet_v2_50,resnet_v2_101,resnet_v2_152), inception_v4, inception_resnet_v2等。

chunk_size = 256
chunk_n = 160
rnn_size = 256
num_layers = 2
n_output_layer = MAX_CAPTCHA*CHAR_SET_LEN   # 输出层

单层rnn：

tf.contrib.rnn.static_rnn：

输入：[ 步长, batch,input] 

输出：[n_steps,batch,n_hidden] 

还有rnn中加dropout

def recurrent_neural_network(data):
    
    data = tf.reshape(data, [-1, chunk_n, chunk_size])
    data = tf.transpose(data, [1,0,2])
    data = tf.reshape(data, [-1, chunk_size])
    data = tf.split(data,chunk_n)
    
    # 只用RNN
    layer = {'w_':tf.Variable(tf.random_normal([rnn_size, n_output_layer])), 'b_':tf.Variable(tf.random_normal([n_output_layer]))}
    lstm_cell = tf.contrib.rnn.BasicLSTMCell(rnn_size)
    outputs, status = tf.contrib.rnn.static_rnn(lstm_cell, data, dtype=tf.float32)
    # outputs = tf.transpose(outputs, [1,0,2])
    # outputs = tf.reshape(outputs, [-1, chunk_n*rnn_size])
    ouput = tf.add(tf.matmul(outputs[-1], layer['w_']), layer['b_'])
    
    return ouput

多层rnn：

tf.nn.dynamic_rnn：

输入：[batch,步长,input] 
输出：[batch,n_steps,n_hidden] 
所以我们需要tf.transpose(outputs, [1, 0, 2])，这样就可以取到最后一步的output

def recurrent_neural_network(data):
    # [batch,chunk_n,input]
    data = tf.reshape(data, [-1, chunk_n, chunk_size])
    #data = tf.transpose(data, [1,0,2])
    #data = tf.reshape(data, [-1, chunk_size])
    #data = tf.split(data,chunk_n)
    
    # 只用RNN
    layer = {'w_':tf.Variable(tf.random_normal([rnn_size, n_output_layer])), 'b_':tf.Variable(tf.random_normal([n_output_layer]))}
    #1
    # lstm_cell1 = tf.contrib.rnn.BasicLSTMCell(rnn_size)
    # outputs1, status1 = tf.contrib.rnn.static_rnn(lstm_cell1, data, dtype=tf.float32)
    
    def lstm_cell():
        return tf.contrib.rnn.LSTMCell(rnn_size)
    def attn_cell():
        return tf.contrib.rnn.DropoutWrapper(lstm_cell(), output_keep_prob=keep_prob)
    # stack = tf.contrib.rnn.MultiRNNCell([attn_cell() for _ in range(0, num_layers)], state_is_tuple=True)
    stack = tf.contrib.rnn.MultiRNNCell([lstm_cell() for _ in range(0, num_layers)], state_is_tuple=True)
    # outputs, _ = tf.nn.dynamic_rnn(stack, data, seq_len, dtype=tf.float32)
    outputs, _ = tf.nn.dynamic_rnn(stack, data, dtype=tf.float32)
    # [batch,chunk_n,rnn_size] -> [chunk_n,batch,rnn_size]
    outputs = tf.transpose(outputs, (1, 0, 2))
    
    ouput = tf.add(tf.matmul(outputs[-1], layer['w_']), layer['b_'])
    
    return ouput