[R-pkg-devel] For reproducibility issue

[Jeff Newmiller]

As an existing package I suspect that it already calls the R RNG, but the default RNG changed [1] so you need to call
RNGkind(sample.kind = "Rounding") before calling set.seed in your test code to compare with old results.

[1] https://stackoverflow.com/questions/47199415/is-set-seed-consistent-over-different-versions-of-r-and-ubuntu/56381613#56381613

On January 17, 2020 6:39:14 AM PST, Ivan Krylov <krylov.r00t using gmail.com> wrote:
>On Fri, 17 Jan 2020 13:55:39 +0000
>وليد خلف معوض المطيرى <wkmtierie using qu.edu.sa> wrote:
>
>> So, does anyone have an idea of how to solve this issue.
>
>"Writing R Extensions", 1.6. Writing portable packages:
>
>>> Compiled code should not call the system random number generators
>>> such as rand, drand48 and random, but rather use the interfaces to
>>> R’s RNGs described in Random numbers. In particular, if more than
>>> one package initializes the system RNG (e.g. via srand), they will
>>> interfere with each other.
>
>>> Nor should the C++11 random number library be used, nor any other
>>> third-party random number generators such as those in GSL.
>
>It somewhat less convenient to call the R random number generator from
>Fortran than it would be from C or C++, but still possible. There is a
>F77-style example of such use [1], but since you are already using
>iso_c_binding, you should be able to declare the C API [2] right in the
>Fortran source:
>
>subroutine GetRNGState() bind(c)
>end subroutine
>
>subroutine PutRNGstate() bind(c)
>end subroutine
>
>As a bonus, you get to use the R distribution functions [3], without
>the need to implement them yourself from uniformly distributed samples:
>
>function rnorm(mu, sigma) bind(c) result(ret)
> use intrinsic, iso_c_binding, only: c_double
> real(c_double), value :: mu, sigma
> real(c_double) ret
>end function
>
>function rgamma(shape, scale) bind(c) result(ret)
> use intrinsic, iso_c_binding, only: c_double
> real(c_double), value :: shape, scale
> real(c_double) ret
>end function
>
>(The prototypes above are unchecked; I haven't written any Fortran 2003
>in more than a year.)

-- 
Sent from my phone. Please excuse my brevity.