Dave Tweed wrote:
> On Fri, 4 Dec 1998, Keith Wansbrough wrote:
>
> > Surely it would be better to split the one stream into several infinite ones:
> > - - -
>
> Closer inspection reveals this is not a necessarily a the best idea
> (particularly if you're going to repeat the trick several times on various
> subsets) because you can get nasty correlations between the various
> streams even from quite a good random number generator. There was a paper
> published in the JFP about a better way of splitting streams which I think
> appeared sometime between January 1996--October 1996.
This reminds me slightly of the Functional Pearl: On generating unique names
(Augustsson, Rittri, Synek), which appeared in JFP in January 1994. [vol 4(1)]
We showed how to simulate an infinite binary tree of unique names,
by having a hidden call to gensym().
Instead of gensym(), I suppose one could hide a call to a random number
generator. But I don't know if nasty correlations can appear in this case.
-- Mikael Rittri
