Author: David Dahlem
Date: 06:18:20 09/05/04
Go up one level in this thread
On September 03, 2004 at 21:10:03, Robert Hyatt wrote: >On September 03, 2004 at 18:49:27, Alessandro Scotti wrote: > >>I was trying to teach my engine this ending, and so far I haven't found a >>solution I really like. I more or less know how the ending should be played, and >>my attempts mainly focused on moving the lone king towards the "bishop corner" >>and the other king in a sort of opposition. While I was tuning the parameters, I >>noticed how slow is progress in some positions. In this one: >> >>[D]8/8/8/3k4/8/8/8/3BN2K w - - 0 1 >> >>GNUChess takes 44 moves to mate, I gave up even earlier as pieces seemed to move >>without a logic. >>So I have two questions now, the most important: how should this ending be >>played, exactly? Would it be useful (i.e. lead to a faster mate) to have say a >>two-stage code where we first gain the center of the board and then push the >>other king to the corner? >>And the second one: using the EGTB for this ending, would it be possible to >>analyze the paths that lead to mate and automatically generate good coefficients >>to help an algorithm (or actually almost piece-square) based solution? > > >All you need is an evaluation that will do the following: > >1. drive the losing king to the edge of the board, using all three pieces to do >so so that all three pieces are "close to the action." > >2. drive the losing king to the corner attacked by the bishop. > >A 3-4 ply search is enough to do this, although not "optimally"... What about when your engine is on the losing side of a kbnk ending? Does Crafty have code to try to avoid the kbnk mate, if possible until the 50-move rule? Regards Dave
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.