Add option to squash UserWarning about double derivatives

[sirmarcel]

Hello, it seems that #141 added a warning on initialisation about issues with second derivatives. Is it possible to disable this warning, or at least only issue it when the second derivatives are called? It's not a huge issue, but I'm not a big fan of unnecessary entries in my log files.

[ceriottm]

This should be triggered only once, and it's actually kind of important as it can lead to silent failures down the line. You can disable it with the usual mechanism you mute warnings, but was discussed at length between @Luthaf and @frostedoyster and I thing this was the least evil. Don't ask me why it can't be triggered when second derivatives are computed, but I think it's a torch limitation (?).

[sirmarcel]

Silent failure is indeed bad, is there any reason to not just return nan in this case? Seems like silently returning something that is a number is a bit of a foot gun, as they say.

[frostedoyster]

Perhaps we could make the switch to torch.compile, which should solve these issues. IDK if that would integrate well with our software stack @Luthaf

[frostedoyster]

I agree that, as an easier fix, we can have a warn argument (true by default)

[sirmarcel]

I don't think it's worth adding an argument tbh, I'll just live with it.

[sirmarcel]

Alright, I think this is a clear wontfix, I'm closing this.

[ceriottm]

yeah one can use `warnings.filterwarnings` if it gets annoying

…

On Mon, 2 Sept 2024 at 17:53, Marcel ***@***.***> wrote: Alright, I think this is a clear wontfix, I'm closing this. — Reply to this email directly, view it on GitHub <#147 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAIREZ5WPMKEIPZLYG37CGTZUSCY7AVCNFSM6AAAAABNQQ4YRKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGMRVGAZDGNBZGQ> . You are receiving this because you commented.Message ID: ***@***.***>