Author: Howard Exner
Date: 16:24:10 10/25/98
Go up one level in this thread
On October 25, 1998 at 10:23:51, Amir Ban wrote: >On October 25, 1998 at 05:43:29, blass uri wrote: > >> > >>It is not enough and I think it should use negative depth for moves that leads >>to a pawn ending. >> > >This is already done > > >>2) There is an evaluation problem. >>Junior should evaluate the pawn ending as winning for black and avoid Kh2 >>because the black king is closer to the white pawns than the white king. >>Junior evaluates it as better for white because white has a passed pawn. >> >>I hope this problem will be solved in the next version. >> >>Uri >> >> >>> results with other programs? >>>(I should note that Junior5 avoids 30. Kh2 at depth=16, but IMO it is too slow.) >>> > >Transition to a pawn endgame is an old Junior problem, but I'm improving all the >time. The problem is that you can and should search to a much greater depth than >you planned to originally. There are several programs that will do better than >Junior, but quite a few who will do worse. I've seen this in many programs also where a situation arises that a simple swap down to pawn ending (or the avoidance of a swap down) is like a blind spot for computers. Once the two ply or whatever length ply of piece exchanges is over the programs immediately wake up to the correct evaluation. One recent example is from the Rebel 9 vs Hiarcs game from the KKup2. Rebel thinks it is winning bigtime by exchanging the Rook for the Knight because it will win a pawn and go into a KP vs K ending. The ending though is an extremely basic draw. Another example is from the Aviod move suite I shared here about a month ago. Hiarcs missed a simple (at least simple for humans) check with a Queen that would force a Queen exchange and an easily won endgame for itself. It's that "Savant" thing with programs where they play so hellishly strong but at times slip into dunce mode. Recent examples that come to mind
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.