Author: Robert Hyatt
Date: 10:51:08 09/11/02
Go up one level in this thread
I just watched _another_ commercial engine vs crafty on ICC. The engine had tables, because it realized that KRB vs KR is a draw. I know this because near the end of the game, Crafty was on the weak side of a KRB vs KRP ending and it offered to give up the pawn and the opponent refused to take it realizing the KRB vs KR ending was a dead draw. But the program voluntarily entered this ending thinking it was winning because it was a piece for a pawn up... It actually had winning chances earlier, but a lack of knowledge in the evaluation (that KRB vs KRP means that the only possible winner is the KRP side) and the peculiar way that a tree search with extensions can postpone the final capture long enough to push the KRB vs KR egtb probe beyond the search horizon, leads to a problem that only changes to the eval can fix. This also happens in KRB vs KRPP cases, the KRB has no chance whatsoever to win that, yet certain programs will dive right in thinking the score is +1
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.