Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Asymmetric Crafty Evaluation

Author: Robert Hyatt

Date: 06:56:28 03/17/03

Go up one level in this thread


On March 17, 2003 at 01:14:38, Ralph Patriquin wrote:

>On March 16, 2003 at 18:54:18, Robert Hyatt wrote:
>
>>On March 16, 2003 at 18:32:01, Ralph Patriquin wrote:
>>
>>>Bob, could you explain what asymmetric evaluation is vs symmetric evaluation,
>>>and pro and con for both? Why is it used? Does it impact the end user in what he
>>>sees for a numeric score from the computer? Also, I seem to remember some config
>>>file setting for Crafty having to do with asymmetric eval, is this the same
>>>thing you're talking about here?
>>>
>>>Thanks,
>>>
>>>Ralph
>>
>>
>>Asymmetric evaluation is an evaluation that not the same when computed for
>>black and white.
>>
>>One example is blocked pawn positions.  Most believe a computer is bad in
>>these positions and that such positions make it easier for the human to sneak
>>into a drawn position.  The solution is to have the program dislike such
>>positions so that it won't get into them.  Of course, avoiding blocked positions
>>all the time is not a good idea, so it needs to be somehow controlled so that
>>a program that thinks it is in a bad position might block things to avoid
>>opening up a bad position into a much worse position.
>>
>>It's a matter of taste.  Without it, many programs will make ugly looking moves
>>such as d4 d5 c4 e6 c5, which starts things toward a blocked queen-side.  I
>>won't say it is a "cure-all" but it can (if used carefully) help.
>>
>>Another case is king safety.  Programs are bad about understanding an open
>>king-side until they see the roof falling in.  The fix is to ramp up king safety
>>scores, so that it won't try to win a pawn by letting its kingside get wrecked.
>>But then it may well offer pawns to wreck the opponent's king-side when it is
>>not sound, and lose.  Asymmetric king safety will let it avoid wrecking its own
>>kingside, while not unsoundly offering material or positional gambits to open
>>the opponent's king...
>
>OK, if I understand this correctly, Crafty will value some feature of the
>position (like King safety) higher for itself than its opponent for reasons of
>avoiding particular problems. Does Crafty do this however when in analyze mode,
>where it is evaluating both sides' play, or is that a symmetric eval? If it's
>still asymmetric, how should output be interpreted?


No.  The asymmetry is turned off in analysis mode so that the score doesn't
flip-flop as real moves are made in the game being analyzed.  I use this mode
on ICC to provide real-time commentary on games being played, and I added the
ability to disable asymmetry to avoid the questions that came up when a player
made a move and the score went from +1 to -1 because of the asymmetry in the
code.



>
>As an example of what I'm getting at is Stephen Ham's opinion that Fritz'
>analysis is more objective, whereas Shredder's is optimistic (for example when
>Fritz is showing +0.5 Shredder might be showing +0.8). When using Crafty to
>analyze with and it's showing say +0.5, can the user assume this to be a true
>half pawn, or does asymmetric eval cause the result to be on the high or low
>side, or change with side to move rather than stay fairly stable?
>
>Ralph


in analysis mode the evaluation should be pretty stable, regardless of the side
to move...



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.