elementwise versions of unary functions

[bob-carpenter]

Summary:

There are functions like log2() and log1p() that are still not vectorized for elementwise application to containers.

Description:

All of our unary functions should be vectorized. Review the index of the manual and make sure every unary function is vectorized. Here are the ones for this issue:

• acosh
• asinh
• atanh
• exp2
• log1m_inv_logit
• log1p
• log2
• log_inv_logit
• logit
• Phi_approx
• trigamma
• trunc

Current Version:

v2.11.0

[bob-carpenter]

See #202, which was only partially completed.

[bob-carpenter]

See also #272 for logit() and #271 for log1m_inv_logit

[bob-carpenter]

This was blocked before due to ambiguous functions. It's fixed with a judicious application of namespace qualifications and Boost enable_if and disable_if qualifications. Now that I have log1p working I should be able to knock off the rest in no time. It will not require reverse-mode specializations as suggested in #272 and #271 (now closed).

[MatsuuraKentaro]

Stan manual v2.17 says pow is not vectorized yet. Do you have a plan?

[bob-carpenter]

Yes, but it's going slowly. We should just vectorize that function (and the ^ syntax to go with it) rather than trying the general solution which Rayleigh Lei is working on.