Automatic differentiation

[mohsenkarkheiran]

I want to check the derivatives of a PDE's solution relative to the independent variables and parameters of the equation.
However dde.Model(data, net).predict(...) returns numpy array only and cannot be used for differentiation. so I need tf.tensor.

It seems I should use Model.net() instead of Model.predict(), but when I use that I get the following error:

" 'FNN' object is not callable "

So I cannot have access to the neural net...

Any hint/suggestion will be appreciated.
Thanks

[pescap]

Hi, did you check operator in predict?

deepxde/deepxde/model.py

Line 907 in 18400e5

def predict(self, x, operator=None, callbacks=None):