Hello!
At the moment I am trying the functionality of the framework on the Kirsch problem. And for a good solution, it requires a lot of density around the cutout. Is it possible to adaptively compact the sample closer to the edges during training? At the moment I have divided the area into two parts - the small area around the cutout and everything else. Are there any smarter ways?
Dividing the interior domain into two low and high-resolution regions is a good approach for this problem. We have an example similar to the Kirsch problem that is using the same approach, and can be found in examples/fuselage_panel. Another approach you can try is the importance sampling. Please refer to the user guide and the sample example in examples/annular_ring/annular_ring_importance_sampling.py for more details.
Thanks, but while working, I noticed that while the annular_ring example was running, no logging occurred. If you pause the model and then continue training, sampling starts from the beginning.