Author: Martin
Date: 09:31:16 03/02/00
Go up one level in this thread
On March 02, 2000 at 09:41:16, Robert Hyatt wrote: >On March 02, 2000 at 09:09:48, stuart taylor wrote: > >>Obviously a bug in stalemate evaluation S. Taylor > > >There is _no_ stalemate evaluation to have a bug. This is found by the >tree search, where there are no legal moves and the side on move is not in >check... > >The problem is that the rook checks will never end, and once the engine sees >that, a draw by repetition, or by the 50 move rule, or by stalemate, all look >_exactly_ the same. Kx was the quickest/easiest way to end the game. > >This is rare, but happens. I have (so far this past week or so) seen this >happen to crafty playing Chess Tiger, where crafty was 3 pawns up but avoided >trading rooks and ended up in this trap. And on another occasion it returned >the favor by doing the same thing to Tiger. It is a depth of search issue, >and not evaluation. You just have to see deep enough to realize that the >checks never end, The "human-approach" is to see that the wK has no squares to move to and this is an evaluation issue. It's probably not easy to implement into a program (if at all) but it would probably be a great step forward... Martin and then avoid stepping into that position, rather than >stepping into it and then finding out (later) that the checks never end.
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.