OK, I am forwarding this to the r-help. In the original message
I simply hit reply and I thought I send to ALL r-help list. (apparent not)
also I recieved another piece of mail.....about the RNG in
Splus 4.0 or later indicating it was modified and getting worse.
I only have Splus 3.4 for HP. So I cannot try it.
This makes the idea of using the
pre-generated RN on CDROM more appealing..
Mai Zhou
Forwarded message:> From cribari at de.ufpe.br Fri Nov 6 12:40:49 1998
> Message-Id: <199811061638.OAA12522 at de.ufpe.br>
> From: "Francisco Cribari" <cribari at de.ufpe.br>
> Organization: UFPE
> To: Mai Zhou <mai at ms.uky.edu>
> Date: Fri, 6 Nov 1998 14:39:37 -0000
> MIME-Version: 1.0
> Content-type: text/plain; charset=US-ASCII
> Content-transfer-encoding: 7BIT
> Subject: Re: [R] simple questions about R
> Priority: normal
> In-reply-to: <199811061620.LAA00381 at t3.mscf.uky.edu>
> References: <3640785F.46F65937 at de.ufpe.br> from "Francisco
Cribari" at Nov 4, 98 03:53:03 pm
> X-mailer: Pegasus Mail for Win32 (v3.01c)
>
>
> Mai,
>
> This is indeed interesting. Have you posted this to the R mailing
> list? If not, may I suggest that you do?
>
> I hope next versions of R will implement a better RNG!
>
> Francisco
>
>
> From: Mai Zhou <mai at ms.uky.edu>
> Subject: Re: [R] simple questions about R
> To: cribari at de.ufpe.br (Francisco Cribari)
> Date sent: Fri, 6 Nov 1998 11:20:05 -0500 (EST)
>
> > > 2) Can anyone send me details on the random number generator
> > > used by R (period, etc.). Also, why is it different from
> > > the SuperDuper RNG in S?
> >
> > I am teaching a bootstrap course, I asked my students to do bootstrap
> > and simulation with R and find the following: (we use Rjune with some
patch)
> >
> > First we found the bootstrap estimate of the bias do not seem to work
well,
> > then we tried simulation and it also do not work. Here is the details
> > about simulation:
> >
> > Generate 19 min( exp(0.3), 4 ) observations
> > and estimate how big is E (X_max) - 4 = the bias of MLE.
> > Using as much as 5000 to 10000 runs, you get something around -0.158
> > (depending on different seed there is some variation but all
> > between -0.15 and -0.165 )
> >
> > Using Splus (version 3.4 on HP) we get something around -0.35
> >
> > Just curious we then used Maple to compute the theoretical value of
the
> > bias and it is -0.35092
> >
> > So, Splus's RNG is better than R's. And I think a simpler case
can be
> > found with just the Uniform[0 theta], instead of truncated exp.
> >
===================================================================> >
> > George Marsaglia has a CDROM (also online) of pre-generated RN's
> > Has anybody tried reading them into S/Splus/R/C/f77 and start from
there?
> > (Any potential problems? speed/memory use etc...?)
> >
> > Mai Zhou
> > mai at ms.uky.edu
>
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help 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: r-help-request at
stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._