Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Random keys and hamming distance

Author: Dieter Buerssner

Date: 13:52:14 08/17/02

Go up one level in this thread


On August 17, 2002 at 03:38:04, Daniel Clausen wrote:

Some comments to Daniel, some other comments to various points made in this
tread.

>The Crafty-sources contain a reference to a very good random generator.

I have no doubt, that the lagged Fibonacci generator used by Crafty is more than
adequate for Zorbrist hashing. However, I would not call it a good pseudo random
number generator. (Almost) all lagged Fibonacci generators have at least one
serious and well known flaw - they totally fail Marsaglia's duplicate birthday
test - the failure can be statistically very significant be seen even with
rather few calls to the generator. So certainly, for serious statistical
simulation (which more or less is done by the mentioned test), they should be
handled with very much care - or better avoided when not combined with another
algorithm.

It is some time ago, when I tried various generators for zorbrist hashing. I
could not detect any systematic difference in test positions. Of course, the
very worst PRNGs must be avoided. For example some (i.e. one used in some Unix
systems) only toggle the lowest significant bit, and when one naivly combines
several calls to such a PRNG to create one 64 bit random number, one will have
several basically useless bits.

For slightly serious PRNGs, I see no problem to combine several calls to get a
64 bit number.

Another point: it is to my knowledge not clear at all, that good randomness of
the keys is needed for Zorbrist hashing. Also, it at least seems doubtful, that
having a very large mininimal Hammond distance is needed or especially good (we
combine more than 2 of those keys ...).

Regards,
Dieter




This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.