keras搭建多层lstm层

记住中间的lstm层需要返回所有timestep的输出作为下一层lstm的输入，所以出了最后一层lstm外其它层的return_sequences=True

from keras.layers import LSTM, Input
inputs = Input(shape=[4, 1])  # num_step(4) input_size(1)
lstm1 = LSTM(units=32, return_sequences=True)(inputs)
lstm2 = LSTM(units=64)(lstm1)