Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: simplifying evaluation

Author: Alvaro Jose Povoa Cardoso

Date: 03:25:50 08/03/04

Go up one level in this thread


On August 02, 2004 at 21:10:38, Stuart Cracraft wrote:

>On August 02, 2004 at 12:12:04, Gian-Carlo Pascutto wrote:
>
>>On August 02, 2004 at 00:02:07, Stuart Cracraft wrote:
>>
>>>On August 02, 2004 at 00:01:18, Stuart Cracraft wrote:
>>>
>>>>So I had nothing better to do and drastically
>>>>stripped down my evaluation (which was already
>>>>pc/sq only) to make some of the tables all zeros.
>>>>
>>>>The speedup was negligable but I was surprised to
>>>>see a 2.3% improvement in score on a problem set.
>>>>
>>>>Seems that some "knowledge" is really best left out.
>>>>
>>>>Stuart
>>>
>>>And yes I know the point that this will hurt in a
>>>real game.
>>>
>>>Sign me,
>>>
>>>The Problemist
>>
>>It's worth trying if that's really true.
>>
>>Someone once quoted Chrilly Donninger as saying (about Nimzo)
>>"The stupider I make it, the better it plays!".
>>
>>Broken knowledge often hurts more than it helps.
>>
>>--
>>GCP
>
>Well said.

No doubt. On the other hand the more knowledge an evaluation function has the
more cutoffs (pruning) it causes wich is good. Imagine an evaluation with only
piece values (no square values at all), it would evaluate different positions
with the same value. But if you add square evaluation and lots of other criteria
it will differentiate more positions so it will prune a lot more.
So knowledge correctly implemented must be good.

Best regards,
Alvaro Cardoso



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.