Purpose of num_test argument in TimePDE

[sibyjackgrove]

Could you tell me how the num_test argument is used in TimePDE? It seems even if I set it to 1 and call data.test_x. I still get a large number of test points.

[lululxvi]

This is the testing points for PDE loss, not including BC points.

[sibyjackgrove]

I see from the docs that testing points are randomly sampled from the domain and so should be different from the training points. But what I don't understand is why the training loss is equal to the testing loss?

[lululxvi]

"If None, then the training points will be used for testing."

[sibyjackgrove]

So if num_test is assigned a non-zero integer, then the testing loss should be different from training loss?

[lululxvi]

Yes.

[sibyjackgrove]

Ok, thanks! I get it now.