并行LSTM,每个LSTM在输入的不同部分工作

Keras Functional API

from keras.layers import Input, LSTM, concatenate, Dense
from keras.models import Model

input_1 = Input(shape=(15, 6), name='input_1')
input_2 = Input(shape=(15, 6), name='input_2')
lstm1 = LSTM(256, name='lstm1')(input_1)
lstm2 = LSTM(256, name='lstm2')(input_2)
concat = concatenate([lstm1, lstm2]) 
output = Dense(6, activation='tanh', name='dense')(concat)
model = Model(inputs=[input_1, input_2], outputs=output)

如果您不想使用多个输入 Lambda

from keras.layers import Input, LSTM, concatenate, Dense, Lambda
from keras.models import Model

input_ = Input(shape=(30, 6), name='input')
input_1 = Lambda(lambda x: x[:, :15, :])(input)
input_2 = Lambda(lambda x: x[:, 15:, :])(input)
lstm1 = LSTM(256, name='lstm1')(input_1)
lstm2 = LSTM(256, name='lstm2')(input_2)
concat = concatenate([lstm1, lstm2]) 
output = Dense(6, activation='tanh', name='dense')(concat)
model = Model(inputs=input_, outputs=output)

你会打电话给 fit 每个示例的函数分别如下:

model.fit(x=[input_1, input_2], y=y)

或者

model.fit(x={'input_1': input_1, 'input_2': input_2}, y=y)

单输入:

model.fit(x=input_, y=y)