Riccardo Tomada

For the plot, you can have a look at #634, last update. From my experience, a loss in the order of 1e-2 is generally not enough. My suggestions: - Impose...

For the hard bc you need to modify the output of the network, see 'apply_output_transform'. Briefly, imagine you have a domain (0,1)^2 and DirichletBC = 1.0 for u on the...

I think it's better to rescale the domain in this case

Thank you very much for your kind reply. I'm going to work on it and keep you and anyone interested updated

You are right, Adam adapts the learning rate of each parameter already on its own. But the value you set when you compile the model is an upper bound. What...

Update: First of all, thank you @lululxvi for having clarified me how to correctly use the functions on_boundary() and inside() when dealing with a Polygon. I think this may also...

Hi @ZazaCro, did you come up with a working implementation of the adaptive loss weights strategy?

I noticed that the algorithm of the "inside" method is quite slow, especially if the points to be checked are a lot and the sampling method is latin hypercube. For...

I just went through the same problem. If it may help anyone, here it is the full code which can be easily extended to the case in which N >>...