Author: Don Dailey
Date: 18:09:47 12/08/98
Go up one level in this thread
On December 08, 1998 at 19:41:45, Will Singleton wrote: > >> >>.... There is a >>slight enhancement involving the window you use on the re-search >>but your description is not buggy, and your problem is elsewhere. >> > >What's the slight enhancement? > >Will When the zero width window fails high, the score returned is actually a lower bound. Therefore the re-search does not have to be done with window (alpha, beta) but can be done with the tighter window, (score_returned_from_fail_hi, beta) Having said that, you must always take care that lazy evaluation and any margin crap in your program returns the right assumptions, or you can get strange behavior. For instance, if you do lazy evaluation, let's say your program assumes that the positional component of the evaluation never can be greater than 1.5 pawns in either direction. If you do a material evaluation and can take a fail hi cutoff based on this assumption (material - 1.5 >= beta) then the score you must return is material - 1.5, not just material even though either case is a beta cutoff. OTHERWISE, this re-search stuff in the previous paragraph can go wrong. - Don
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.