Tf.nn.dynamic_rnnとMultiRNNCellは多層動的LSTMを構築します

Tf Nn Dynamic_rnn Multirnncell Build Multi Layer Dynamic Lstm

import tensorflow as tf import numpy as np X = tf.random_normal(shape=[3, 5, 6], dtype=tf.float32) X = tf.reshape(X, [-1, 5, 6]) stacked_rnn=[] for i in range(3): stacked_rnn.append(tf.contrib.rnn.BasicLSTMCell(24)) # cell = tf.nn.rnn_cell.BasicLSTMCell(10) # can also be replaced with something else, such as GRUCell, BasicRNNCell, etc. lstm_multi = tf.contrib.rnn.MultiRNNCell(stacked_rnn) # state = lstm_multi.zero_state(3, tf.float32) output, state = tf.nn.dynamic_rnn(lstm_multi, X, time_major=False,dtype=tf.float32) # with tf.Session() as sess: # sess.run(tf.initialize_all_variables()) # print(output.get_shape()) # print(sess.run(state)) print(output.shape) Print(len(state))#Three LSTM hidden layers # LSTM hidden layer Print(state[0].h.shape)#hTM in hs state Print(state[0].c.shape)#cTM in cs state # LSTM hidden layer print(state[1].h.shape) print(state[1].c.shape) #third LSTM hidden layer print(state[2].h.shape) print(state[2].c.shape)