I admit to having acquired a Windows platform a couple of weeks ago, and
trying to get all the insidious junk off of it that the vendor piled on
(some of it is part of the default OS install). That, and fixing an ailing
Mac, have thrown me off a bit.
To RAND():
In a perverse way, I'm sorta pleased that negative numbers showed up; it
seems to indicate that at least MS attempted 32-bit integers on this
go-round.
I missed the flag Don had raised re the generator, but had my own issues:
while RAND was touted at passing DIEHARD, it wasn't verifiable, since we
weren't given the underlying longs, or some way to map the doubles back to
them, in order to test the claim. Despite several requests, neither was the
generator documented, as per usual scientific standards.
Still, I have high hope the problem will be resolved, quickly. MS put a fair
amount of work into improving these functions; I would be surprised if they
were to let this undermine what otherwise is good (and
years-long-anticipated) work in other areas.
There's a well-documented, well-implemented, excellent alternative that is
*free* called PopTools:
http://www.cse.csiro.au/poptools/
The author provides quite a few tools; a number are undocumented, but if you
think it should be there, it likely is. E.g., he provides a routine for
Cholesky decomposition if you want to generate multinormals; he also
provides a direct call, mentioned in his covar demo. I've been using it with
great success on a current project.
Check it out--- I think you'll like it. Again, it's documented!
Oh, one last question--- has anyone actually seen a 1, 0 or -1? I ran the
generator for a while during beta, never saw the negatives, nor 0 or 1, and
wasn't going to test it without further info from MS as to what they were
attempting to do. It's easy enough to avoid the endpoints (much less the
negatives <bg>).
DaveB
.......