Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Secrets of Rybka and Fruit from my point of view

Author: Robert Hyatt

Date: 07:58:12 12/16/05

Go up one level in this thread


On December 16, 2005 at 01:28:19, Chrilly Donninger wrote:

>On December 15, 2005 at 21:55:02, Robert Hyatt wrote:
>
>>On December 15, 2005 at 21:14:45, Sergei S. Markoff wrote:
>>
>>>>>>I find that VERY hard to believe. That concept has been around a
>>>>>>very long-time.
>>>>>
>>>>>Really? Where it was introduced by the first time?
>>>>
>>>>Don't know about the first time, but GNUChess (v4) uses the same approach,
>>>>albeit only for some king eval terms, and many programs scale things by
>>>>material, although not necessarily as nicely as Fruit does.
>>>
>>>Well, I think such material dependence for some factors was in Kaissa too :) But
>>>I never seen before Fruit the united concept of opening/endgame evaluation. Also
>>>material dependence of some factors seems to be appeared at first time in Fruit.
>>>For example -- piece mobility depends of stage.
>>
>>
>>That's not new at all.  I can think of several programs that have done this
>>"scaled" eval approach since the late 1970's...  If you look at Crafty's eval,
>>you will see it done much more subtly than what is done in Fruit.  Individual
>>eval terms are scaled by material when appropriate.  Endgame things like passed
>>pawns, distant passed pawns, pawn majorities/candidate passers, all scale up as
>>material comes off.  King safety scales down in the same way.
>
>
>I scale the terms also individually. I think most programms do that. But the
>Fruit approach has some beauty. With the individual approach one has more
>freedom for tuning. The relation between material and the eval terms differ. But
>one can also introduce inconsistencies in the eval. Its hard to keep everything
>balanced. The formula is typically for Fruits KISS approach. Its hard to do any
>stupid things with the Fruit-formula.
>
>Chrilly
>
>Chrilly


Would not disagree.  Am in the middle of a complete re-vamp of the crafty eval
code, and I am going back to "simple is better".  There was simply way too much
complexity in the old way with "3rd order terms" (where groups of pieces
contribute to a term (2nd order) and then these 2nd order terms interact to
produce a 3rd order result.)  Nearly impossible to understand at times...

I've come up with a better way to scale eval terms for each piece so that I can
control the total positional score a piece can contribute, without having to
tune each individual weight to adjust this.  Hopefully, simple turns out to be
better in addition to easier...



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.