Author: Graham Laight
Date: 07:24:36 01/16/02
Go up one level in this thread
On January 16, 2002 at 10:11:02, Albert Silver wrote: >On January 16, 2002 at 08:09:58, Graham Laight wrote: > >>On January 16, 2002 at 07:49:25, Albert Silver wrote: >> >>>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. >>>> >>>>-g >>> >>>Not so strange considering that chess is a game of absolutes, whether we know >>>them or not. A positions is either a win (with best play), a draw, or a loss. >>>How are you going to estimate how many times a player _should_ win? --> >>> >>>Hmmm... Normally, I'd say John is going to win this, but having seen him drink 5 >>>beers during lunch shortly before the game, I'd say he only has a 60% chance.... >>>:-) >>> >> >>I don't say that the evaluations would be more "accurate" - but I do say that >>the number produced, accurate or not, would be on a more sensible scale. >> >>-g > >In backgammon, evaluations are done according to probability since there is no >way to control the outcome of the dice, so you can only calculate your chances. >Chess is a different animal altogether. In chess, there is a theoretical >absolute answer. In other words, if I were to calculate very single possibility, >I'd know what the result would be with best play. Either the position is a win, >a draw or a loss. If chess were to be solved, then the evals would indeed only >have 3 evals possible. There is no probability though, unless you consider the >human chance to err, as I joked about above. Objectively, there is never a 70% >chance to win since best play must _always_ be assumed in order to play the best >moves. Suppose I were choosing two moves. In move A, the theoretical absolute >outcome is a win though the road is a very thin line (a long series of "only" >moves - anything else in the entire road would in fact lose), so it must be >considered the best move. Move B however presents more opportunities for the >opponent to make a fatal mistake and lead to easier wins, however if the >opponent doesn't fall for any of these traps move B is theoretically an absolute >loss (with best play). Move A has less "probability" to win, and Move B has >more. Which should it be? > > Albert These are good points - but they would also apply to pawn evaluation. Saying "white is a pawn ahead" is only meaningful if the evaluation is done on material only. The probability, while clearly not accurate, should be regarded as the best guess given the available information, IMO. What we're doing with pawn evaluation is the equivalent of saying, "Taking into account the comfort, economy, price, cornering ability, and equipment level of this car, our evaluation of its suitability for you is 0-60 in 9.1 seconds, sir." -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.