Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Hash collisions in chess programs?

Author: KarinsDad

Date: 15:09:25 02/19/99

Go up one level in this thread


On February 18, 1999 at 17:04:14, Dann Corbit wrote:

>On February 18, 1999 at 14:45:59, Inmann Werner wrote:
>[snip]
>>What does this mean in particular for chess programming?
>>
>>I use one hashingadress between 0 and 2000000 for the hash adress and one
>>extra calculated long int for verification. Each of the both hashcodes result
>>out of 2 different sets of 64 long int numbers.
>>How high is the probability of a collision in this case, and do you use the
>>same algorithm?
>I think we have to know what hash we are using in order to talk about
>collisions.  For instance, I could choose a hash for people's names which uses
>the first six letters as base 26 numbers.  This would probably be a rather bad
>hash, because people's names are not spelled at random, and because we are
>likely to receive them already in order (unless, for some reason we *want*
>collisions).
>It seems that most chess programs use Zobrist hash.  I think it would be
>interesting to count collisions for opening, early middle game,  mid-game,
>endgame, and late-endgame and see if the proportion stays the same as the board
>clears.  I am fairly sure that it will not be exactly as probability would
>dictate if we had a perfectly random distribution of positions.  It might also
>be interesting to investigate the properties of other hashing algorithms.

I have been wondering if changing the Zobrist hash from a set of random number
to a set of non-random and very specific numbers could result in a more even
distribution in the hash table. Has anyone done any work in this area?

KarinsDad



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.