Author: Robert Hyatt
Date: 07:25:53 08/27/05
Go up one level in this thread
On August 27, 2005 at 05:15:57, Tord Romstad wrote: >On August 26, 2005 at 23:11:18, Robert Hyatt wrote: > >>Most don't do history like that. Usually it is just a 12 bit index <from><to>. >>So it doesn't differentiate between a king move from f6 go g7, and a >>bishop/queen move from f6 to g7... > >There is no reason you have to use the same history table for move ordering >and reduction. I have two tables. The table used for move ordering is similar >to what you describe. It is indexed by the from and to squares, and the >entries are incremented by the remaining depth when a move causes a beta >cutoff. The table used for reductions is indexed by the moving piece type >and the destination square, and for each move I count the number of times >it has been tried and the number of times it has produced a cutoff. > >Tord Although for a different reason, testing (for me) years ago didn't produce good results with that kind of history indexing. For example Qc4 now means 20+ queen moves, from all squares connected to c4.... It seems to contain _less_ information than using <from><to>. This was for normal history move ordering, however, not for reductions. I even tried <piece><from><to> but it just made the table bigger with no gain in performance/accuracy...
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.