Author: Vasik Rajlich
Date: 04:05:52 04/05/04
Go up one level in this thread
On April 04, 2004 at 07:50:40, Ricardo Gibert wrote: >On April 03, 2004 at 18:55:49, Albert Silver wrote: > >>>>I can give an example >>>> >>>>A good evaluation can evaluate that black has 0 chances to win here(I think that >>>>it is a draw but assume for the discussion that there is no knowledge to be sure >>>>about it) >>>> >>>>[D]5rk1/6pp/8/8/8/8/5PPP/5RK1 w - - 0 1 >>> >>>Interesting position, NOT even the top program like Shredder8 can evaluate it as >>>draw >>> >>>Analysis by Shredder 8: >>>1.f4 Ra8 2.Kf2 Kf7 3.g4 Ra3 4.Rb1 h6 5.Kg2 g6 6.h3 Ra2+ 7.Kg3 Ra3+ 8.Kh4 Ra4 >>> ± (1.24) Depth: 26/49 01:19:11 1383446kN >>> >>>(Pichard, MyTown 02.04.2004) >> >>It's hardly a straightforward draw, and black's defense is a difficult one. >> >> Albert > >You must be thinking of the 4 vs 3 ending. The 3 vs 2 ending is not difficult. Actually, it doesn't really matter if it's straightforward or difficult. The question is, what is the expected result. 3 vs 2, the expected result, given a reasonable level for the black player (including enough time), is right around 50%. I'd be happy to have Rybka evaluate this as +0.00 (or +0.01). However, there are some interesting theoretical issues. In the middlegame, general heuristics are quite ok - you have some advantage of magnitude X. The opening position is clearly drawn, yet it's clearly appropriate to give white a plus score. In the endgame, you'd like to be more precise, regarding whether the position is drawn or winning. For example, it's hard to talk about having an advantage in a king & pawn ending. It's either winning, or it's drawn. Middle scores just show that you don't know yet. The question is, where do these meet? For example, how do you score something like R+B v R, without tablebases? You really can't give scores in a vacuum - scores always imply a certain type of opponent. Practically, it doesn't matter until you start to evaluate endgames precisely, which not even Shredder, Fritz and Junior do. Cheers, Vas
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.