How to give different inputs for subnetworks in PFNN

[bottle88]

Discussed in #1733

^{Originally posted by bottle88 May 11, 2024}
Dear All, I have a question of using the implementation of PFNN.

I have a 2D heat equation where T(x,y,t) is the solution. And I have a inverse varaible parameter in my BC. The parameter only applies on a particular boundary. so it is supposed to dependent on x only.
In the PFNN implementation. The First NN has x,y,t as inputs and T(x,y,t) as ouput , and second NN also has x,y,t as input and C(x,y,t) as output, as far I understand. But I need the second NN to have x as input and C(x) as output. How do I do this

I have seen the poison equation example, where they had unknown source field as output, and tried to implement it. But the function prediction was not at all correct. https://deepxde.readthedocs.io/en/stable/demos/pinn_inverse/elliptic.inverse.field.html.

This is how I had defined, seeing the example given.
layer_size = [3, [30,30], [30,30], [30,30], [30,30], [30,30], 2] activation = "tanh" initializer = "Glorot uniform" net = dde.nn.PFNN(layer_size, activation, initializer).

I am having doubt on what do I need to change to implement according to my condition. I am confused.

I am eager to hear your response. Thankyou in Advance.