math functions to move to SpecialFunctions.jl

[StefanKarpinski]

• lgamma
• gamma
• lfact
• beta
• lbeta

[ViralBShah]

Sounds good. Where should they go?

[JeffBezanson]

SpecialFunctions.jl?

[StefanKarpinski]

[l]gamma and [l]beta should likely go in https://github.com/JuliaMath/SpecialFunctions.jl with the rest of the *gamma functions. lfact should probably be called lfactorial and live there as well, largely because it's implemented in terms of the lgamma function.

[dpsanders]

Or better loggamma, logfactorial etc.

[StefanKarpinski]

The lfunction convention is pretty standard but sure. We could also do log(gamma)(x)... not entirely sure if ing or not...

[dpsanders]

log(gamma) is a neat idea.

[alanedelman]

+1 on log(gamma)

[freeboson]

FWIW, they're called lbeta, lgamma in R and Tensorflow

[StefanKarpinski]

I do think that lbeta, lgamma, lfactorial and lbinomial are fine names.

[ararslan]

log(gamma) is a fun idea in theory but you then lose the ability to sensibly overload lgamma, since log(gamma) would then return an anonymous function.

[andreasnoack]

It doesn't have to be anonymous. You could have Base.log(::typeof(gamma)) = lgamma

[ararslan]

Then what's the point of having log(gamma) if you still have lgamma...

[andreasnoack]

It looks better and lgamma doesn't need to be exported.

[martinholters]

(log∘gamma) would be more sensible than log(gamma), but having this overload of ∘ would also be a very-hard-to-discover feature...

[StefanKarpinski]

Let's just stick with the more traditional names, i.e. lgamma, etc.

[pkofod]

Speaking of lgamma, I have a port of openlibm's lgamma that can be used in SpecialFunctions.jl.

[simonbyrne]

lgamma etc. are pretty much standard, and tend to be more useful then their non-log counterparts (since they're harder to overflow) so should have convenient names. In fact, the gamma in C originally gave the log version, which is why C has a tgamma ("true gamma") function