About the choice of loss_weights

[ZSTanone]

Hi Dr.Lu and deepxde users,
I've recently been working on the inverse problem of the odes to estimate parameters. I find set different loss_weights will lead to changes in results.

We know that in solving some stiffness equations，their solutions will differ by many orders of magnitude，so we can add output_transform to fix that .However,in this case, is there a conventional way to set the loss_weights? A common idea is to keep the loss of the initialized model on the same order of magnitude.And there are also proposals that set the loss_weights of obsevers bigger than odes in inverse problem ,Does this make any sense？

[lululxvi]

Yes, it makes sense.