Reputation: 121
How to make a not fully connected graph in Keras? I am trying to make a network with some nodes in input layer that are not connected to the hidden layer but to the output layer. Is there any way to do this easily in Keras? Thanks!
Upvotes: 5
Views: 1518
Reputation: 57739
Yes, this is possible. The easiest way to do this is to specify two inputs:
in_1 = Input(...)
in_2 = Input(...)
hidden = Dense(...)(in_1)
# combine secondary inputs and hidden outputs
in_2_and_hidden = merge([in_2, hidden], mode='concat')
# feed combined vector to output
output = Dense(...)(in_2_and_hidden)
The documentation is better at explaining what merge
does in detail. The general idea of multiple inputs and the functional model can be read here.
Upvotes: 5