Author: Bas Hamstra
Date: 20:42:28 01/11/01
Go up one level in this thread
On January 11, 2001 at 06:16:47, Severi Salminen wrote: >On January 10, 2001 at 20:45:40, Bas Hamstra wrote: > >>The fail high and then low problem is normal if you use nullmove and aspiration >>window. Vincent (and me too BTW) doesn't use aspiration window and therefore >>doesn't see it. Nullmove causes search inconsistencies. >> >>Bas. > >Thank you very much! If this is the same in reversed order (first the search >fails low and then after infinite window search the true score is within the >original window) I can sleep well again :) And focus on improving other areas. > >Severi It used to bother me a lot too. Then Bruce Moreland explained that search inconsistencies (which this is about) are unavoidable if you use hashtables and pruning based on alpha/beta values. To quote him: "If you use a hashtable, you have dirt in you hair. If you do nullmove, you are buried in mud". Pretty clear :-) If you think about it, it's logical. Nullmove makes errors. With different windows different branches will be pruned resulting in different errors. Enough to affect the rootscore. I saw it all the time. Bruce too. When I hacked the ExChess code to see if it had the same behaviour a year ago the answer was: yes. If you turn nullmove and hash off the problem should not occur at all, or else you have a bug. I don't use aspiration anymore, as soon as I can measure it matters a lot I will put it back in. At this point I am not convinced, though most do use it. Ciao! Bas.
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.