Author: Pete Galati
Date: 14:31:57 04/10/00
Go up one level in this thread
On April 10, 2000 at 15:54:49, Robert Hyatt wrote: >On April 10, 2000 at 11:01:43, Colin Frayn wrote: > >>I was just wondering how Crafty handles draws in its evaluation routine. I've >>been playing several 10 minute blitz games (10 min/side) between Crafty and >>ColChess and I managed to win the following two games, though I noticed that >>near the end Crafty was often giving the position as drawn to many ply (like 20 >>or more) as it thought it could draw by repetition when behind. ColChess plays >>along a few times, then realises what's about to happen and plays something >>else, whereby Crafty's analysis goes very negative again. >> >>Surely Crafty must know that ColChess will try to avoid a draw when ahead? >> >>Also, does Crafty have a contempt setting? I would have resigned both of these >>games a lot earlier :) >> >>Cheers, >>Col > >The only thing that will cause the above is drastic differences in the >evaluation. Crafty thinks that the opponent will either repeat or get a worse >position... so it assumes it will repeat. > >to make it resign sooner, try "resign=3" which resigns 5 moves after getting >to be a piece down. > I assume that that's actually somewhat flexible, I mean if you set Crafty to resign at 3 (3 pawns?) then Crafty goes down to scores of -3 or less in the search (must be the final search depth that Crafty reaches before making it's move) that if at the 4th move, if Crafty has somehow climbed out of the whole enough to interpret it's score as being -1, then the resign is off. In other words, 5 consecutive searches at -3 would get Crafty to resign. Or, does Crafty score each position currently on the board and resign by that (5 consecutive positions below resign number) but that method would have to take away Crafty's ability to be able to see if it would have a way to dig it's way out of a bad spot. Pete
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.