R devel - Oct 2003 - Random Number Generator RNGkind() under "R CMD check" (PR#4691)

Full_Name: Wolfgang Huber
Version: 1.8.0
OS: Linux
Submission from: (NULL) (193.174.58.146)


The man page for RNGkind says that the default is Mersenne-Twister, and when I
start R interactively, I get in fact
> RNGkind()
[1] "Mersenne-Twister" "Inversion"

However, during the execution of "R CMD check" I
get
> > ### ** Examples
> >
> >  RNGkind()
> [1] "Wichmann-Hill" "Inversion"
This creates discrepancies between the behaviour of code that uses pseudo random
numbers during "check" and otherwise.
> version
         _
platform i686-pc-linux-gnu
arch     i686
os       linux-gnu
system   i686, linux-gnu
status
major    1
minor    8.0
year     2003
month    10
day      08
language R

>>>>> "WoHub" == w huber <w.huber@dkfz.de>
>>>>>     on Mon, 20 Oct 2003 22:15:04 +0200 (MET DST) writes:
    WoHub> Full_Name: Wolfgang Huber
    WoHub> Version: 1.8.0
    WoHub> OS: Linux
    WoHub> Submission from: (NULL) (193.174.58.146)


    WoHub> The man page for RNGkind says that the default is
    WoHub> Mersenne-Twister, and when I start R interactively, I
    WoHub> get in fact

    >> RNGkind()
    WoHub> [1] "Mersenne-Twister" "Inversion"

    WoHub> However, during the execution of "R CMD check" I get
    >> > ### ** Examples
    >> >
    >> >  RNGkind()
    >> [1] "Wichmann-Hill" "Inversion"

    WoHub> This creates discrepancies between the behaviour of
    WoHub> code that uses pseudo random numbers during "check"
    WoHub> and otherwise.

yes, but this has been so on purpose "forever".
If you really want reproducible example using random numbers
you must set both RNGkind() and set.seed().

I'd agree that this should be documented -- and further, we tend
to agree that we should consider changing the current behavior,
probably to something like
	 RNGkind("default","default"); set.seed(1)
{in the cleanupEx() function called before each example}.

But note again: You still may see "discrepancies" between the "R
CMD check" and the normal usage, unless you call at least
set.seed(.) yourself in your example.

Martin Maechler <maechler@stat.math.ethz.ch>
http://stat.ethz.ch/~maechler/
Seminar fuer Statistik, ETH-Zentrum  LEO C16	Leonhardstr. 27
ETH (Federal Inst. Technology)	8092 Zurich	SWITZERLAND
phone: x-41-1-632-3408		fax: ...-1228			<><