R-alpha: random number generator
Martin Maechler <email@example.com>
Fri, 21 Nov 1997 18:45:03 +0100
Which reminds me of a remark that Luke Tierney made a while ago,
and I very much agree with:
LT> From: Luke Tierney <firstname.lastname@example.org>
LT> Subject: Re: R-alpha: compatibility
LT> Date: Mon, 15 Sep 1997 11:55:02 -0500 (CDT)
LT> Cc: Remail@example.com
LT> This is one point where compatibility is not necessarily a good
LT> thing. S, like most systems, gives you no real control over the random
LT> number generator expect re-seeding. Usually this is probably good. But
LT> most simulation books recommend running any simulations you care about
LT> with two very different generators.
Yes! -- That's why I think it would be nice if R allowed for SEVERAL
random number generators, one of them super-duper, incidentally.
The code is basically in V&R 1 and 2; V&R2 on p.167.
I have it as a C function that I used to dyn.load into S-plus in order to
prove that S-plus was using it.
my function does NOT YET contain the ``zero skipping'' which is mentioned
in the yellow book ..
LT> Most people don't do this because
LT> one generator is usually easy to use (the built-in one in your
LT> favorite system) and anything else is impossible. Another issue that S
LT> faces has to do with running separate bits of a simulation on separate
LT> workstations. Unless you are careful, the separate bits will share
LT> parts of random number streams, and being careful is not really
LT> possible with S's generator (I think) -- it is possible but not easy
LT> with R's. (Of course overlap can be a good thing if you know you have
LT> it -- the point here is that lots of folks run separate simulations
LT> thinking they have independent runs when they probably do not in any
LT> reasonable sense).
LT> The bottom line is that it would be nice to eventually provide "power
LT> users" with more control over the random number generator system. Now
LT> is clearly not the time to think about how to do that, but it would be
LT> unfortunate if decisions were made at this point in the name of
LT> compatibility that might limit the freedom of doing something "right"
LT> later on.
maybe the time to think about this, could be now..?
-- Martin M
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !) To: firstname.lastname@example.org