Re: need a good implementation of pseudorandom generators


David A. Heiser wrote:
Yes, we can say that KISS is a suspect RNG or defective generator. A lot of
the inventions of the 1960's, 1970', 1980'3, etc. ended up as being included
in the main frame (and unix) computers ( the age of FORTRAN). Some of the C+
packages included these historic wrecks. One always finds these resurrected
from junkyards of the past. The 'Numerical Recipe's books are collections
from these junkyards.

The first ever big surprise I got from a random number generator was
when using the built in random function on a VAX machine running Ultrix
(a version of unix). I was generating random x,y coordinates by taking
random integers modulus the size of a 2d area. For some strange reason,
the number of 'hits' I got on various pixels was roughly uniform for
about half the pixels, and 0 for all the others. A quick inspection of
the pixels hit showed that the random numbers generated were
alternately even and odd, so that sampling pairs resulted in points
that were all (even x, odd y) or vice versa.

When I last did similation, I used a Java implementation of Mersenne
Twister, as recommended in this group at the time.

BTW: I wasn't recommending the use of correlations as a method of
testing random numbers. Even I know that. I was just looking at the
OP's claim that similar seeds (by some definition of similar) resulted
in near identical sequences.

If the OP can post some of his/her seeds, I'd like to see whether I get
the same sequence of numbers from two "similar" seeds or not.

Cheers,

Ross-c

.