Re: R standalone Rmath library - missing log1pexp(), rnbeta()

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

Re: R standalone Rmath library - missing log1pexp(), rnbeta()

Martin Maechler
>>>>> William Birkett
>>>>>     on Mon, 1 Jun 2020 15:16:07 -0400 writes:

    > I built the the standalone Rmath library,
    > https://cran.r-project.org/doc/manuals/r-release/R-admin.html#The-standalone-Rmath-library
    > <https://cran.r-project.org/doc/manuals/r-release/R-admin.html#The-standalone-Rmath-library>

    > The build process went smoothly.

    > But, there are two functions, log1pexp() and rnbeta(),
    > which appear in the header file, and not in
    > ‘libRmath.dylib’.

    > There is a comment in the header, '// <--
    > ../nmath/plogis.c’ next to log1pexp(). That file,
    > ‘plogis.c’ contains a function ‘Rf_log1pexp’, which is how
    > the function would be named without '#define
    > MATHLIB_STANDALONE’.

    > The other function, 'rnbeta()’ doesn’t appear in any of
    > the source files in the ‘nmath’ folder. Maybe a file is
    > missing.

Incidentally (not being an Apple costumer), I've seen this; as
I'm the one who had added  log1pexp()  (8 years ago), I've been intrigued.

The problems you are seeing are *not* at all related to R on the
Mac. ==> I'm exceptionally crossposting to R-devel.
=======> Please followup there, *not* on R-SIG-mac.

1) An R core colleague of mine had decided log1pexp() should be
   Rf_log1pexp() but had forgotten to update the documentation.

   This should be fixed one way (= documentation),  or then, as
   it is the *only* function with name Rf_*() in the whole of standalone
   Rmathlib.{so,dynlib} by renaming it back to log1pexp().

2) rnbeta(): This has been in the headers forever, but there has
   not been a corresponding function in R's C code.

   As you, can see, R's  rbeta(*, ncp = v) currently just calls R code,
   and so there was never a need for a C level rnbeta().  OTOH,
   this is a pity/asymmetry, and I had wanted to change this
   (and also add 'rnt' or 'rnf') to the standalone Rmathlib for
   a long time.  The reason this has not happened (yet?):
   Because of vectorization working differently in the current R
   code of the r*(*, ncp=*) functions and the corresponding C
   code, this would lead to different random numbers in this
   case... and so such a change --- to make Rmathlib "complete" ---
   is something to be discussed.

Best regards,
Martin Maechler
ETH Zurich  and  R Core Team


    > William B. Birkett ....
    > .......[hidden email]

    > [[alternative HTML version deleted]]

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel