Thanks all!
On Thu, Jan 20, 2011 at 9:35 PM, Robert Dodier <robert.dodier at gmail.com> wrote:
> IIRC Maxima's MT implement generates chunks of 32 bits at a time,
> and uses those to construct the output which is some number
> of bits wide. For each N-bit output, Maxima uses 1 + floor(N/32) chunks.
> Any leftover bits are thrown away (not used for the next output).
Ah yes, I've found this now. Is this described in the mt19937
standard? Or is this simply one way out of many to do this?
>
> All correct implementations of MT19937 should yield the same output
> given the same initial state, should they not?
> (Otherwise, it would seem, the nice properties of MT prng's aren't guaranteed.)
Indeed, although I guess it's how one then casts the result down into
the correct range that might be different. I'll keep looking into it.
Thanks!
Matthew Gwynne