Author: Dieter Buerssner
Date: 13:55:26 10/25/01
Go up one level in this thread
On October 25, 2001 at 15:39:19, Uri Blass wrote: [Much snipped] >Usually king safety test positions are positions when the sacrifice is correct >and this is the reason that I chose position when the sacrifice is wrong. I think, this is a very interesting point, that bothered me for some time. Many test positions start with a sacrifice. I think, it is easy to overtune an engine because of this. For example, I have seen that Yace gets sometimes much better results with a higher value for a pawn. But in normal games, I will see more bad trades than. Also rather aggressive search extensions can often help, while it might hurt in normal games. Test-suites like the one posted here by Jon Dart, with many "avoid move" problems might be better suited. Of course, this has other problems. The engine might choose another bad move, and this might not be easily detected. Giving a list of all good move will not be easy or debatable. But in practice probably many engines will have similar good looking but wrong ideas, and so the avoid move type of postions should work well. Of course, often the engine will be just lucky and solve just positions extremely fast, without detecting any problem with move to avoid. But with many positions, this should level out. Regards, Dieter
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.