[boost] Re: The Mersenne Twister -- a C++ implementation

From: jeetsukumaran_at_[hidden]
Date: 1999-11-05 15:49:42

> A basic uniform random generator and a set of transformation
functions to
> convert to a different distribution.
> It may be better to separate the two functions such that any rand can
be
> used with any transformation function.

I like this idea.  Using templates to maintain orthogonal
decomposition, with different uniform prng passed to the transformation
functions ...

> 
> It may be prudent to provide several test routines to test both the
> generators and the transformation functions. (against other MT
> implementations such as the one suggested above?)
> 
> Paul

The original C-source code for the Mersenne Twister is available at:
ftp://random.mat.sbg.ac.at/pub/data/mt19937b-int.c
ftp://random.mat.sbg.ac.at/pub/data/mt19937b-real.c
You can use this to test the sequences of your MT19937B for accuracy.

Actually, the algorithm has a homepage.  But I've never managed to get
to it --I connect to the server, but cannot download a thing.  It's
been that way for a week.  The links I gave above are the original
source code maintained at a different site.

If you are interested in testing the worth of the algorithm itself (or
any others), then you might be interested in Marsaglia's diehard
battery, which appears to be the benchmark of choice for prng's
currently:

http://stat.fsu.edu/~geo/diehard.html

Warning -- it requires something like 10 - 12 million numbers generated
per test!

The MT19937B has passed all the diehard tests.  A full report can be
seen at

ftp://random.mat.sbg.ac.at/pub/data/pLabReport7.ps

-- jeet