Author: Uri Blass
Date: 09:08:52 05/13/01
Go up one level in this thread
On May 13, 2001 at 10:07:27, Robert Hyatt wrote: >On May 13, 2001 at 03:50:15, Gian-Carlo Pascutto wrote: > >>...when it could happen to you. >> >>On ICC my program reached the following position: >> >>[D]2k2r2/2P3R1/1P6/P7/8/1K6/8/8 w - - 0 56 >> >>Convinced it was winning, it now played a6?? >>after which black draws easily. >> >>I remember having seen this theme before in problems, >>but not in a real game. Ouch! >> >>Most other programs don't seem to have a clue about >>this position either, though :) >> >>-- >>GCP > >It's a null-move killer because it is based on zugzwang problems. White is >up material + whatever bonuses it gets for three connected passers, which is >quite a bit normally. Which means that any null move done will think white is >simply winning. Yace is a null mover and it has no problem to see that moves like Kb4 are better than a6. You do not need to see that a6 is a draw in order to avoid it and you only need to see that other moves are better. The sequence of checks to reach the draw is also quite long >which aggravates things... The main problem of a lot of chess programs is their evaluation, They can see a lot of checks and after the checks capturing the rook and believe that white is a rook and 3 pawns up because they do not see the stalemate. Programs evaluate the position as more than +9 for white and it is ridicilous(I guess that the +9 is evaluation of stalemate positions). Yace is more inteligent and evaluates a6 only as +5 for white and it is enough for it to avoid a6. It will be interesting to see if Crafty can avoid the mistake if you add the following if in your evaluation function: If the side to move has only a king then if the side to move is in stalemate then return 0. The side to move has only a king is only a private case and you may add more cases but the price is being slower in other positions. Uri
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.