Author: Jay Scott
Date: 09:10:28 02/06/01
Go up one level in this thread
On February 05, 2001 at 02:27:12, Steffen Jakob wrote: >Normally Hossa has no chance to find the solution. Some time ago I worked on an >idea to detect and evaluate such blocked positions. The main idea is to not only >evaluate a position statically but also to consider how the evaluation >progresses from ply to ply. If you already seem to have such a big advantage >like black in Eduards test position than you "should" be able to make this >advantage bigger within "some" plies. If you are not able to do this, then this >is a strong indication for a blocked position. I have written a liitle bit about >this on Hossa's homepage (http://www.jakob.at/steffen/hossa.html). That seems like a difficult idea to implement correctly. It's gonna mess up the hash table, isn't it? I have proposed that playout analysis is a good way to detect fortress draws: http://satirist.org/learn-game/inspire/playout.html But it seems difficult to make good use of in a chess program, because playout analysis is essentially a selective search that spends most of its effort far from the root, which means that in most chess positions it is wasting most of its time. It's only in these unusual blocked positions that such distant information becomes vital for the evaluation. If there were a way to recognize when playout analysis would help and only do it then, it would be a win. But somehow I can't see such an expensive analysis as a feasible search extension. :-) Jay
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.