Author: Miguel A. Ballicora
Date: 10:11:12 01/16/02
Go up one level in this thread
On January 16, 2002 at 07:41:28, Graham Laight wrote: >It has occurred to me that it is wrong to evaluate a position in terms of >relative pawns (the "de facto" standard - whereby an evaluation of 2 means that >you're approximately the equivalent of 2 pawns ahead). > >This means that many aspects of evaluation have to be squeezed into a dimension >which is not appropriate at all. > >A better way would be to evaluate "winning probability". If a position was a >draw, the value would be 0.50 (or 50%). If the player should win 3 out of 4 >times, the eval should be 75%. If the player must win from here, then the >evaluation should be 100%. > >It seems strange when you think about it that all programmers have chosen to >adopt the traditional "pawn equivalence" standard. I agree with you, except in some details. In fact, every single programmer is doing what you say, they just do not know it :-). everytime they add bonuses, penalties, etc. etc. they are adjusting a score that could be expressed as "probability". For instance, a pawn does not worth the same in the opening than in an endgame. Why? because you have more chances to score a win in an endgame with a pawn extra than in the opening (of course you have to include exceptions an other parameters into the equation). That makes the engine go for "higher" probabilities, expressed in pawns. I think that it could be possible to calibrate (it won't be linear) a scale from 0 to 100% compared to -MATE to MATE. Mate in one, should not be 100% to allow the program to go "forward" to 100%. 100% is only a Mate in the board. So, mate in one could be 99.999% or whatever. The point is, nothing will change, just the output. In fact, I try to think in terms of probabilities: one PAWN is equal to the prob. to win with a PAWN advantage in the initial position. Regards, Miguel > >-g
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.