Computation of the test loss

[loubnabnl]

Hi @lululxvi,

I solved a 1D heat equation using DeepXDE, but I get test loss vectors for which only the first value is not equal to zero. So I don't understand why since I assume we are using the same loss function as the training set. And this always gives a training loss greater than the test loss, which I find a bit odd.

Thank you.

[lululxvi]

In the test loss, I only keep the PDE loss, which is more important.

[thierryhen]

Hi @lululxvi,

Please can you explain what are the different numbers figuring in the training loss or test loss list?

[lululxvi]

@thierryhen See FAQ.