Author: José Carlos
Date: 00:46:33 11/16/01
Go up one level in this thread
On November 15, 2001 at 14:52:01, Dieter Buerssner wrote: >On November 15, 2001 at 06:51:23, José Carlos wrote: > >>>[D] 8/8/6p1/6Q1/2K4p/6kP/5p2/8 b - - 0 55 > >> This is the kind of position programs can chose the right moves, but don't >>understand at all. My program says +7.xx for white, but it defends correctly >>with black. > >Yes. > >> I guess this position won't be solved by any program in a long time because it >>requires reasoning. You can't write code for positions like this easily, because >>an extra pawn here or there makes a difference. > >Certainly not easy. But this has bothered me for a while. I have seen it in >games of Yace, and just accidently I saw this game yesterday. > >Perhaps a naive idea would be: Dectect the cases of drawn Q vs. P (R or B-P on >7th rank and some rules for K positions). If all pawns of the Q-side are >blocked, it gets tricky. When from no blocking pawn a check can be given, or the >promotion square can be controlled, give the P on 7th the value of the Q. It is >of course still no guarantee for draw, because the Q side may be able to force >the promotion, exchange Qs and have a won pawn endgame. > >Perhaps a table with all such blocked pawn positions. Or some special table >base. > >Perhaps also a special search rule, although this won't be easy. When I play >against Yace in the above position after about 20 moves, I get a draw score. >When I go back, immediately a draw score is seen for any move almost until back >to the position (Yace remembers the positions together with search depth and >score, that where analyzed). Only at one point, where the move Qxg6 was, this >capture will be avoided, again with a bogus high score for white. When I do >again about 10 moves forward I see the draw score again. No I can go back until >the root, and the draw score will be seen there. This is a totally mechanical >manner - so it should be possible to do this automatically. Note that when you do that, you know what you're looking for. A human could do this even if he's not sure whether it is a draw or not, because humans 'feel' what to look for. Now for a program you can hard-code some rules and catch a percent of the cases, but: a) you'd do this special-search many times it shouldn't be done, hurting normal search; b) you'd miss many positions where you should have tried it. My (pessimistic?) intuition tells me the overall is that you'd lose strength with such a special-search. But it's true that if you don't try you don't lose... and you don't win. :) José C. >I think, essentially >some sort of tablebase will be built for this position. > >Cheers, >Dieter
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.