使用Keras创建具有多个输入的NN体系结构

Question

I want to create a model with the following architecture:

          i_1    i_2     i_3 (3 input nodes)
       h1_1 ..... h1_j ...... h1_n ('n' nodes in 1st hidden layer)
     h2_1 ....... h2_j......... h2_m ('m' nodes in 2nd hidden layer)
                  o         (1 output)

Further Details

1. Each input node has a descriptor vector of constant length (let's say the length for each descriptor vector is 10)
2. I'd like each node in the 1st hidden layer to be "connected" to all 3 input nodes (ie for a weight to exist for each combination of hidden and input node) and for each node in the 2nd hidden layer to be connected to each node in the 1st hidden layer.
3. The NN is feed-forward

How can I achieve this in Keras?

Answer 1

This looks like a basic feed-forward network:

model = Sequential()
model.add(Dense(n, input_dim=3))
model.add(Dense(m))
model.add(Dense(1))

You can specify activation parameters independently for each layer.