Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Evaluation Autotuning

Author: Dan Honeycutt

Date: 10:35:00 06/28/04

Go up one level in this thread


On June 28, 2004 at 12:53:17, Uri Blass wrote:

>On June 28, 2004 at 12:37:42, Dan Honeycutt wrote:
>
>>On June 28, 2004 at 08:54:00, Anthony Cozzie wrote:
>>
>>
>>>settings, and then N games with the new settings.  I am only really interested
>>>in longer timecontrols: 20 min + on an Athlon 2.0G or so (70 min on P-650, etc),
>>
>>
>>Why long time controls?  I thought you could test evaluation with shorter time
>>controls, search needed longer (or varied) time controls.  Am I out in left
>>field?
>>
>>Dan H.
>
>I think that it is possible that the same evaluation change that is productive
>at long time control is counter productive at blitz.
>It does not mean that testing at short time control to test evaluation changes
>is not useful but if you want to be sure that the new version is better and you
>care mainly about long time control then you need to test in long time control.
>
>I suspect for example that a small reduction of the evaluation of pawns may be
>productive at long time control and counter productive at blitz.
>
>The reason is that piece is usually better than 3 pawns in the middle game but
>at blitz you may not have enough time to find the plan how to use it espacially
>when the side with the pawn has another positional advantage that is not enough
>to compensate for the piece when at long time control you may find the right
>plan because of searching deeper.
>
>The right plan for the side with the pawns is more simple:push the pawns forward
>It means that program will have less problem to find the right plan even at
>blitz.
>
>I do not know if this theory is correct but it is possible that another theory
>for another evaluation term is correct.
>
>Uri

My evaluation covers, I think, most of the basic stuff but I don't have any more
sophisticated stuff like you describe - altering pawn value based on time
control or stage of the game.  I feel like I should tune what I've got before I
go adding more.  Even with my fairly basic evaluation I have over a hundred
values for penalties/bonuses.  Every one of them I pulled from the sky (or some
other part of my anatomy).

This thing Anthony is working on sounds like it could be ideal for someone in my
situation.  If I can get to a decent "ground zero" then I can twiddle and tweak
from there.  I have lots of things that need work so I think I'm going to put
evaluation aside for now and see where this leads.

Dan H.



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.