Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Hash Table Size Versus Performance.

Author: Randall Shane

Date: 07:30:24 06/05/98

Go up one level in this thread


On June 05, 1998 at 09:34:39, Robert Hyatt wrote:

>I don't agree with him on collisions.  In general, the collision rate
>increases with the depth of the search tree, something that has been
>known for many years, since lots of folks tested the "Zobrist hashing"
>scheme in chess programs.  A single move greatly perturbs the hash
>signature.  and the liklihood of a collision after one ply is close
>enough to zero to be zero, particularly if you check your random numbers
>to be sure that they average a hamming distance of 32, with *none* that
>drop below 16, which I have done.

I think I could use some advice on how to better generate such numbers.
Inspired by Crafty, I'm writing my own program from scratch (currently
called BadChess, not public yet).  My current set of 896 64-bit hash
numbers has a minimum distance of 23 and a median distance of 32.  I'm
currently generating random numbers and tossing out those that drop the
distance below 23.  It takes about 5 minutes to run on a P-90 laptop.
Is there a better and faster algorithm for generating the set than Monte
Carlo?

Thanks!



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.