Author: Robert Hyatt
Date: 11:47:48 09/20/99
Go up one level in this thread
On September 20, 1999 at 12:14:10, Bruce Cleaver wrote: >Here's an idea: many programs include a pawn-only hash table, and the hit rates >on it are quite high. Since the bishop evaluation in any postion is strongly >coupled to the type of pawn structure (bishop + many pawns on same color is >generally a negative, for example), you might save computation time by putting >these together. The pawn-only tables would then be a subset of the pawn-bishop >hash tables (those pawn-bishop positions without any bishops). > >Other pieces aren't nearly so coupled to the pawns (rook, knight, etc.). > >No doubt it has its drawbacks as well. Perhaps the space required would greatly >increase, leaving less room for the full hash table, thus actually decreasing >performance. Still, I'd like to hear feedback.... > >Thanks, > >Bruce I used to do a king-safety hash table, which is a bit easier since there is only one king vs two bishops per side. I typically saw the hit rate drop to 70% rather than the usual 99% for pawns only. I think bishops would make the memory required to avoid overwrites so big, it would overall hurt performance. I stopped the king safety stuff, and just take the pawn shelter on both sides of the board and store it in the pawn hash entry after computing it. Then, when I evaluate king safety, I just pick the shelter score for the right side of the board. I get way more hits, and it is faster overall. Pawns are tied to other pieces as well.. ie weak pawns are tied to rooks being present, etc. It is probably easier to hash the conditions, but apply the scores when the pieces are evaluated...
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.