Merge output layer into DirectParams

[ruigangwang7]

Hi Nic,

Since the ExplicitParam has already included the output layer, maybe we should merge the output layer param into directparams as well. So it will correspond to the implicit G in the paper. How do you think?

Best
Ray

[nic-barbara]

Hi Ray, I've been thinking about doing this but haven't gotten around to it yet. What was Max's reasoning for keeping them separate in the first place?

I agree that they should be merged, particularly now that we have a D22 term that depends on all the direct parameters.

[ruigangwang7]

The possible reasons are as follows: 1) make it close to classic RNN,LSTM, having a separate output layer; 2) at the beginning we choose D22=0, so the output layer is decoupled from internal dynamics; 3) for the observer example we just need the state dynamics.