R devel - Dec 2002 - win/pentium 4 (was: can this happen?)

I reported an inconsistency in results in the same R program run on
different Windows computers and received some suggestions

It isn't due to the version of msvcrt.dll (Peter Dalgaard & Brian
Ripley)
and isn't fixed by Duncan Murdoch's patches to windows R.

More experimentation has shown that the set of results I like less
(involving internal error reports from optim()) happens on 3 out of 3
Pentium 4 machines, and not on 2 Pentium 3s (or on a dual Athlon under
Linux).


	-thomas

Thomas Lumley			Asst. Professor, Biostatistics
tlumley@u.washington.edu	University of Washington, Seattle

On Mon, 9 Dec 2002, Thomas Lumley wrote:
>
> I reported an inconsistency in results in the same R program run on
> different Windows computers and received some suggestions
>
> It isn't due to the version of msvcrt.dll (Peter Dalgaard & Brian
Ripley)
> and isn't fixed by Duncan Murdoch's patches to windows R.
>
> More experimentation has shown that the set of results I like less
> (involving internal error reports from optim()) happens on 3 out of 3
> Pentium 4 machines, and not on 2 Pentium 3s (or on a dual Athlon under
> Linux).
>
Using Rterm instead of Rgui seems to fix the optim() errors (in one
example on one machine, so far).


	-thomas