Fix for special.zeta nan handling - follow-up PR #138653 #146618
Labels
module: special
Functions with no exact solutions, analogous to those in scipy.special
triaged
This issue has been looked at a team member, and triaged and prioritized into an appropriate module
Comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
🚀 The feature, motivation and pitch
Continuing PR #138653.
I hereby attach the suggestion of Albert Steppi (@steppi):
Now that we have this background out of the way. I think my preference in SciPy would be to change zeta(x, q) to be nan and to codify this as a recommendation in a special function array API extension as considered data-apis/array-api#725. I'm not sure what the downstream implications of this change might be though, and would be happy to hear feedback.
If there's no interest for PyTorch extending zeta to x < 1, then having zeta(1, q) return +inf makes sense, and by the principle guiding such special cases in the C99 standard, having zeta(1, nan) return +inf also makes sense in my opinion. However, through the work we are doing in SciPy discussed here, scipy/xsf#1, it would become straightforward for PyTorch to extend zeta if we extend it in SciPy, by using the xsf library as a shared dependency (although you would lose the ability to test against SciPy as an independent reference).
@rgommers @janeyx99 @mruberry
Alternatives
No response
Additional context
No response
cc @mruberry @kshitij12345
The text was updated successfully, but these errors were encountered: