Albert Steppi
Albert Steppi
> I just think we shouldn't have two different kinds of error reporting in `xsf`. We've got the `set_error` system, which we inherited and ugpraded from what was in SciPy...
> Can this be closed now after #21441 @steppi ? Thanks for pointing this out. Not yet. The issue here is that we only allow evaluating this functions when the...
@tk-yoshimura, can you share the script you used to generate these csv files and plots? > * `cot` > `cotdg`, `cotpi` exist but `cot` does not. > Should it be...
This is the same underlying problem as https://github.com/scipy/scipy/issues/21303 and will require adjusting the Boost policies of `ibeta_inv_wrap` as was done for that issue. We had only adjusted the policies for...
+1 for adding this to Boost math and using Boost’s implementation in SciPy.
Yeah, there should be a NaN check in this part of the corresponding scalar kernel https://github.com/scipy/scipy/blob/b7e45413a75c6dda6b8f914a4b262d6c456e516f/scipy/special/xsf/cephes/igam.h#L328-L333 This goes back to the original cephes implementation. I guess we should do an...
> Can that be an infrastructure thing so it does not need to be added to each function? There are exceptions to NaN in / NaN out in `stats` (e.g....
@rlucas7, I've cleaned up the merge conflicts. This is ready to review again.
Ah, more merge conflicts after merging #21401. I’ll have time to fix in a few hours.
This PR also fixes `test_xsf_cuda.py` in general, which is broken on main because the include path for the headers used in this file was not changed from `special` to `xsf`...