--- Raymond Toy wrote:
> I think it is good that all platforms use the same rng by default.
> It should be something that has been well-tested and is reasonably
> fast. And whatever we pick, it should be good so that we don't go
> changing it again at some later time just because something faster
> or better comes along. I think we should proceed cautiously, but
> this is a good time to make the change since so much is also
> happening.
I think MT19937 is a good choice. We can avoid second-guessing
by using MT, which seems to be the ne plus ultra of rng's these days.
The CMUCL MT19937 implementation has already been tested,
so all we need to do is make sure I didn't goof up something
that was already working.
> I used to do lots of Monte Carlo simulations, and it's quite annoying
> to have rng's change so that the old results differ from the new when
> you're trying to debug code. Even starting from a different seed is
> annoying.
Making it easier to test the correctness of some Maxima
code is a strong argument in favor of a single cross-platform rng.
For what it's worth,
Robert Dodier
__________________________________
Do you Yahoo!?
Yahoo! Mail - now with 250MB free storage. Learn more.
http://info.mail.yahoo.com/mail_250