Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Anecdotal data

Author: Keith Evans

Date: 08:45:51 07/10/02

Go up one level in this thread


On July 10, 2002 at 03:24:52, Gian-Carlo Pascutto wrote:

>A version of my program that played crazyhouse chess used two
>hashkeys, one for the pieces in hand and one for the pieces on
>the board. Both 32 bits. Used for both evaluation caching and
>ttable.
>
>After about 1.5 years I discovered a bug that erronously
>always returned a hit regardless of the hash of the pieces
>in hand. Because in crazyhouse this totally changes kingsafety,
>it was essentially returning random evaluations each time.
>
>I fixed this. Some of the weird scores I saw were gone, but
>playing strength was basically unaffected.
>
>The reasoning is simply this. I am doing about 599 999 totally
>wrong evaluations per second. The search corrects for them.
>What are 100 hash collisions going to change?
>
>--
>GCP

Is this effect related to the "Random minimaxing" introduced by Beal and Smith?

"Beal and Smith, using random minimaxing in Chess, showed that
 the strength of play increases as the depth of the lookahead
 is increased"

I think that I saw a reference to that in the Knuth paper on Alpha Beta, and it
really mystified me at the time. I never did track down the original papers
though.

Regards,
Keith



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.