Calculating the NN output in "losses" function of PDE class

[tsarikahin]

Dear Lu Lu,

I have one question regarding calculating NN output during the loss calculation of the PDE class. I am reformulating PDE class so that I can pursue calculations based on elements, not collocation points (PINNs based on weak formulation, Variational PINNs). To achieve that, I decided to make a loop for each element, which requires output of NN for that provided element. As shown below, outputs term contains the NN for the input, basically (train_x_all). However, I would like to obtain NN output for any given placeholder, not the prediction of course since we still build losses. How can I achieve that? I just need to calculate NN output based on updated network. I tried couple of things but I could not obtain that.

def losses(self, targets, outputs, loss_fn, inputs, model, aux=None):
.
.
.

[tsarikahin]

@lululxvi

For instance, model argument in losses function has already net. But I could not see that how can I calculate NN output from there.

[tsarikahin]

For instance, for torch as backend, there is forward function in fnn module. What about tensorflow.compat_v1?

[praksharma]

Hey I think the forward pass for fnn with Tensorflow 1.x is here.