Author: Jay Scott
Date: 15:17:12 01/14/98
It's the last round and your program needs to win. Either a draw or a loss will put you at the same standing. Or, it's the last round and you can't afford to lose. With a win or a draw, you take clear first. Your opponent is stronger than you, and you'll be satisfied with a draw. Or, your opponent is much weaker and you want to try for a win even in a worse position. Humans are pretty good at adjusting their play to the situation. But programs tend to think that an equal sharp position is about the same as an equal drawish position--if they have a preference, then it's always the same preference regardless of circumstances. Some programs have a "contempt factor" that adjusts the value of a forced draw, but I've never heard of a program that tries to take into account the likelihood and importance of an eventual draw that can't yet be seen in the search. The problem can be solved in principle by including a separate drawish<->sharp measurement in the evaluator. If the pawn structure is symmetrical, the position is more drawish; if there's one open file that all the heavy pieces will be exchanged on, it's more drawish; opposite color bishops are drawish in the endgame and sharp in the middlegame; a big concentration of enemy pieces around the king suggests sharpness; that kind of thing. Of course it's hard to measure this accurately, but then, it's also hard to tell whether you have enough compensation for a pawn. Measuring drawishness doesn't seem any more difficult than that. Once you have a drawishness measure, you can use it to adjust the evaluation depending on how valuable a draw is for this game. If you need to win, then a drawish position is similar to a losing position. If you only need a draw, a drawish position is similar to a winning position. Does anybody know of a program that does something like this? Tell us about it! I think this would be good for getting more half points from grandmasters and more full points from weaker opponents. I'm puzzled why I haven't heard of a program doing it. Jay
This page took 0.01 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.