Question

是否可以在keras中添加外部输入到Merge图层？我有简单的嵌入，我想与外部值结合，但每次我尝试做我总是得到错误。有没有办法将外部输入添加到Keras图层？

models = []
inputs =Input(shape=(10,))
models.append(inputs)
for i in range(2):
    model_s = Sequential()
    model_s.add(Embedding(1115, 10, input_length=1, name='Hb_{}'.format(i)))
    model_s.add(Reshape(target_shape=(10,)))
    models.append(model_s)



model = Sequential()
model.add(Merge(models, mode='concat'))
model.add(Dense(1000, kernel_initializer='uniform'))
model.add(Activation('relu'))
model.add(Dense(500, kernel_initializer='uniform'))
model.add(Activation('relu'))
model.add(Dense(1))
model.add(Activation('sigmoid'))
model.summary()

错误

Tensor' object has no attribute 'get_output_shape_at'

这是用于测试的小代码。

model.compile(optimizer='rmsprop',
              loss='binary_crossentropy',
              metrics=['accuracy'])


m={'Hb_0_input': np.array([0,1, 2, 3, 4, 5, 6, 7, 8, 9]),'Hb_1_input': np.array([0,1, 2, 3, 4, 5, 6, 7, 8, 9]), 'x': np.array([0,1, 2, 3, 4, 5, 6, 7, 8, 9])}
y=np.array([0, 1, 0, 1, 0, 1, 0, 0, 0, 0])
model.fit(m, y)

Answer 1

不确定您的意图是什么，但是您将模型和张量混合在同一列表models = [inputs, model1, model2]中。这是错误的原因。

现在，我们不知道你有什么样的输入，所以我们无法进一步帮助，但假设一些事情（可能是错的），这段代码可以帮助你：

inputForEmbedding = Input((length,)) #where length seems to be 1
extraInput = Input(shapeOfTheExtraInput) #I don't know this shape

model1Out = Embedding(1115, 10, input_length=1 name='Hb_{}'.format(i))(inputForEmbedding)
model1Out = Reshape((10,))(model1Out)

....you must make the shapes compatible for `model1Out` and `extraInput`...

outputs = Concatenate(axis=...)([model1Out,extraInput])
outputs = Dense(1000, kernel_initializer='uniform')(outputs)
outputs = Activation('relu')(outputs)
outputs = Dense(500, kernel_initializer='uniform')(outputs)
outputs = Activation('relu')(outputs)
outputs = Dense(1)(outputs)
outputs = Activation('sigmoid')(outputs)

model = Model([inputForEmbedding, extraInput])
model.summary()

将图层与输入相结合

1 个答案: