Author: Stuart Cracraft
Date: 16:28:08 06/17/99
Go up one level in this thread
On June 17, 1999 at 15:27:35, Michel Langeveld wrote: >Is it for evaluation a pawnstructure? >What kind of members does the hash struct has? >Is 32 bit hash enough for this table? > >Kind regards, > >Michel Langeveld Having pawn hash tables reduces pure-pawn evaluation (i.e. not taking into account any relation to other pieces) to a negligible part of overall evaluation-time and makes it possible to cram as much evaluation for pure-pawns into the evaluator without guilt. Also since pawn structure changes more slowly and all the pawns are the same type of piece, there are fewer entries for such a hash table and it changes more slowly, resulting in decreased memory requirements and much higher hit-ratio than the regular full position hash table. Also, keeping the hash table for pawns between moves (or even having an on-disk pawn-hashtable data file that is loaded in between games and appended to with new entries after a game) makes more sense. It makes sense for such an on-disk pawn hashtable data file to have an auxillary program for masters to "rate" various pawn structures numerically as desirable and undesirable for the program if they see something in the program's own pawn position hash table score as grossly incorrect for the program's style. Such rated positions by the human masters might be flagged as "unchangable" once loaded into memory. And so on. Experiment! Have fun with your pawn hash tables! --Stuart
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.