使用Keras

Question

当我创建模型时，我会这样做：

model <- keras_model_sequential()
model %>% 
  layer_1(...)
  layer_2(...)
  layer_3(...)

当我使用摘要时，summary(model)：

Layer  -- Connected to
layer_1 --   
layer_2 -- layer_1
layer_3 -- layer_2

图层是一个接一个地添加的，但我该怎么做：

Layer  -- Connected to
layer_1 --   
layer_2 -- layer_1
layer_3 -- layer_1

我希望第2层和第3层连接到第1层。

Answer 1

您可以使用keras_model：

执行此操作

library(keras)

layer_1 <- layer_input(1)

layer_2 <- layer_1 %>% layer_dropout(0.4)
layer_3 <- layer_1 %>% layer_dropout(0.6)

model <- keras_model(
  inputs = layer_1, 
  outputs = c(layer_2, layer_3)
)

summary(model)

输出：

___________________________________________________________________
Layer (type)          Output Shape  Param # Connected to           
===================================================================
input_1 (InputLayer)  (None, 1)     0                              
___________________________________________________________________
dropout_1 (Dropout)   (None, 1)     0       input_1[0][0]          
___________________________________________________________________
dropout_2 (Dropout)   (None, 1)     0       input_1[0][0]          
===================================================================
Total params: 0
Trainable params: 0
Non-trainable params: 0
___________________________________________________________________

这个例子非常虚拟，只是为了说明这一点。

Answer 2

代码片段的另一个例子：

input1 = keras.layers.Input(shape=(16,))
x1 = keras.layers.Dense(8, activation='relu')(input1)
x2 = keras.layers.Dense(8, activation='relu')(input1)
added = keras.layers.add([x1, x2])

model = keras.models.Model(inputs=[input1], outputs=added)

model.summary()

__________________________________________________________________________________________________
Layer (type)                    Output Shape         Param #     Connected to                     
==================================================================================================
input_44 (InputLayer)           (None, 16)           0                                            
__________________________________________________________________________________________________
dense_65 (Dense)                (None, 8)            136         input_44[0][0]                   
__________________________________________________________________________________________________
dense_66 (Dense)                (None, 8)            136         input_44[0][0]                   
__________________________________________________________________________________________________
add_92 (Add)                    (None, 8)            0           dense_65[0][0]                   
                                                                 dense_66[0][0]                   
==================================================================================================
Total params: 272
Trainable params: 272
Non-trainable params: 0
__________________________________________________________________________________________________