Add inf special cases in C implementation of gamma related functions

[amyoshino]

Description

gammainc C code does not have a special case for np.inf.

Instead of returning 0, 1 or nan it always return nan.

The solution is to add special cases conditions following scipy implementation: https://github.com/scipy/scipy/blob/4ecf5638c36c3a87b473056f5fffecfeddffd368/scipy/special/cephes/igam.c#L149-L151

Related Issue

• Closes BUG: gammainc C code does not have a special case for np.inf #633
• Related to Add icdf functions for Beta, Gamma, Chisquared and StudentT distributions pymc#6845

Checklist

• Checked that the pre-commit linting/style checks pass
• Included tests that prove the fix is effective or that the new feature works
• Added necessary documentation (docstrings and/or example notebooks)
• If you are a pro: each commit corresponds to a relevant logical change

Type of change

• New feature / enhancement
• Bug fix
• Documentation
• Maintenance
• Other (please specify):

[codecov-commenter]

Codecov Report

Attention: 6 lines in your changes are missing coverage. Please review.

Comparison is base (1091b44) 80.82% compared to head (a2b242c) 80.83%.
Report is 3 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #634   +/-   ##
=======================================
  Coverage   80.82%   80.83%           
=======================================
  Files         162      162           
  Lines       46744    46758   +14     
  Branches    11418    11420    +2     
=======================================
+ Hits        37783    37795   +12     
- Misses       6716     6717    +1     
- Partials     2245     2246    +1     

Files	Coverage Δ
pytensor/scalar/math.py	87.34% <60.00%> (-0.45%)	⬇️

... and 1 file with indirect coverage changes

[ricardoV94]

You'll need to define methods like these for the old caches to be ignored, for all Ops that use these two functions.

pytensor/pytensor/scalar/math.py

Lines 1391 to 1396 in 34bfe31

	def c_code_cache_version(self):
	v = super().c_code_cache_version()
	if v:
	return (3, *v)
	else:
	return v

That and some tests and we're good to go!

[amyoshino]

You'll need to define methods like these for the old caches to be ignored, for all Ops that use these two functions.

pytensor/pytensor/scalar/math.py

Lines 1391 to 1396 in 34bfe31

	def c_code_cache_version(self):
	v = super().c_code_cache_version()
	if v:
	return (3, *v)
	else:
	return v

That and some tests and we're good to go!

Thanks for this information! I have added them to any function that uses GammaP or GammaQ in gamma.c. From

pytensor/pytensor/compile/ops.py

Line 74 in ee7c946

def c_code_cache_version(self):

I see that we need to return a "version" number, which I suppose it is 2, since it is the second time we change it. Let me know if my reasoning is wrong, I didn't find any other pointers that justify another choice.

I also added tests to test_math.py, I guess this is sufficient but let me know if I forget to add any other test place.

[ricardoV94]

@amyoshino yes your reasoning is correct

[ricardoV94]

This is great @amyoshino

[amyoshino]

Cool! One more step to close pymc-devs/pymc#6845 😄

[ricardoV94]

Cool! One more step to close pymc-devs/pymc#6845 😄

Any day now xD

We plan to release sometime this week