Author: Robert Hyatt
Date: 12:31:17 09/11/99
Go up one level in this thread
On September 11, 1999 at 11:40:57, William Bryant wrote: >On September 11, 1999 at 11:26:45, William Bryant wrote: > >>On September 11, 1999 at 00:19:40, Robert Hyatt wrote: >> >>>That seems to be backward. If score > beta, the _current move_ you just >>>searched failed high. That move should be stored in the hash table to try it >>>first the next time you search this position. The problem we are discussing >>>is the _other_ case.. the one where you search all movs and still end up >>>with best == alpha, which means all moves were bad... >> >>Alright, >> on re-rereading, let me try this again. >> >>You completed a ply and found no new alpha (failed-low). >>alpha is already greater than a constant MateScore (Mate - some constant). >> >>You store this in the hash table as follows. >> >>if (alpha>MateScore) >> store the value as MateScore and the flag as hash_Lower to show that this >> is the lowest value this position could have. >> Because you failed low, there is not best move. >> >>if (alpha<-MateScore) >> storethis normally, with the flag hash_Upper to show that this is the >> highest value this position could have. >> Again, because you failed low, there is no best move. >> >>Closer? >> > >Actually, One more try-- it is the extremes on a fail low node that cause > problems when abs(alpha) > MateScore > >so, >if (alpha>MateScore) > store the value as MateScore and the flag as hash_Lower to show that this > is the lowest value this position could have. > Because you failed low, there is no best move. > >if (alpha<-MateScore) > store this with a score of -MateScore, > with the flag hash_Upper to show that this is the > highest value this position could have. > Again, because you failed low, there is no best move. > >Even Closer? > > >William >wbryant@ix.netcom.com That sounds like what I am doing, assuming you mean MATE-300 or -MATE+300 where you say MateScore and -MateScore...
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.