Author: Dan Homan
Date: 11:18:13 08/14/98
Go up one level in this thread
On August 14, 1998 at 12:53:12, Inmann Werner wrote: >On August 14, 1998 at 10:33:05, Dan Homan wrote: > >>On August 14, 1998 at 10:08:23, Inmann Werner wrote: >> >>>Question about aspiration window. >>> >>>I use Aspiration alpha-beta. >>>For the first searched move on a ply in deeper depth, I set alpha=oldbestvalue-x >>>and beta=oldbestvalue+x >>>Nothing unnormal. >>> >>>Now I found some positions, where my program lowers the evaluation of this first >>>move of a ply (till now the best!) always with x. (see above) >>>If I change x, it is the same until x gets really high, and i get a "true" >>>value. >>> >>>The disturbing thing on it is, that one ply deeper, the program should recognice >>>how bad this move really is, but it does not make a fail low! (If I get a fail >>>low, I research with open window) > >----- > >>I am going to guess here. Your search cannot return a score less than >>alpha-x, so if you define a fail low as (score < alpha-x), you will never >>fail low (like you describe). If you define a fail low as (score <= alpha-x), >>then everything should be ok. >> >> - Dan >> > >Can i really not get a score less than alpha? > >When I will use score<=alpha-x, this would only be good for the search on the >first move!? All other moves are searched with a close window. So, i think, i Sorry, I meant use (score <= oldbestscore - x) as your fail low condition. - Dan >always would get a fail low. >Am I right? > >P.S: I looked at Minimax (some reference for me), and Minimax uses scorefor a fail low detection. Is this a bug in Minimax? > >Werner
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.