multiple functions from scipy.special do not work with quantities

[namurphy]

I just tried to use SciPy's error function, and Astropy politely requested that I raise this issue here:

>>> from astropy import units
>>> from scipy.special import erf, erfc
>>> r1 = 5*u.m
>>> r2 = 6*u.m
>>> erf(r1/r2)
TypeError: Unknown ufunc erf.  Please raise issue on https://github.com/astropy/astropy
>>> erfc(r1/r2)
TypeError: Unknown ufunc erfc.  Please raise issue on https://github.com/astropy/astropy
>>> erf(r1.value/r2.value)  # workaround
0.76140717068356445

In scipy.special, ufuncs that I would prioritize include:

• Error function, etc
• Bessel functions
• Gamma and related functions
• Legendre functions
• zeta and zetac - Riemann zeta functions

Thanks!

[namurphy]

Note: this is a superset of issue #4092.

[mhvk]

#6843 reminded me that it would be really nice to support scipy's special functions; it is not difficult in principle; see discussion in #4092 (which also gives a work-around).

[mhvk]

@kyleaoman - moving discussion to this open issue, and trying to be a bit more constructive: how about starting with a list of scipy.special functions that are dimensionless to dimensionless, and just add those? It could be something like the following at the end of quantity_helpers.py:

try:
    import scipy
except ImportError:
    pass
else:
    for special in ['erf', 'erfc', ...]:
        UFUNC_HELPERS[getattr(scipy.special, special)] = helper_dimensionless_to_dimensionless

but also needs some tests (sadly, I fear that very good requirement is the reason I didn't get around to this...)

[kyleaoman]

I had a go at starting to add support for scipy.special functions, I now have a branch in a state where I can pick away at it for the next couple of weeks in spare moments. This is my first foray into collaborative github work (though I use it for my own projects). I sense I should mark this as 'in progress' somehow?

[pllim]

@kyleaoman , you can always open a "WIP" pull request and add [ci skip] in your commits to skip the CI builds until you are ready.

[mhvk]

If you want me/others to have a look, the easiest would be to make it a pull request -- see http://docs.astropy.org/en/latest/development/workflow/development_workflow.html#pull-request; then we can add a label that it is in progress (do have a look at the documentation, and let us know if it is unclear). But if it is not in that state yet, you can also paste a link here.

... ah, I see @pllim was ahead of me ... was about to add something about testing.

[kyleaoman]

Sounds good, I think I can start a wip pull request shortly. I don't currently have [ci skip] on my commits, but each commit does pass the tests on my system. I suspect it will break on other systems since scipy.special seems to have shuffled around which functions are implemented as ufuncs over time (at least, the docs of the current version 1.0.0 only sort of reflect what's in the source...). Is it better to go back and add the tag to my commit messages in this scenario?