Author: Richard Pijl
Date: 00:45:08 11/16/04
Go up one level in this thread
On November 15, 2004 at 15:57:23, Gerd Isenberg wrote: >Hi, > >i use the difference of weighted pawnbitboards ( 5*whitepawnBB - 3*blackpawnBB ) >mod 65077 as an index for a 64K entry pawnhash-table. Most often my pawnhash >hit-rate is quite ok, > 95% or even 99%. But it seems that a lot of pawn levers >decrease my hit-rate due to collisions. I guess in the positions you posted below the pawnhash hitrate decreases because there are a lot of pawnmoves possible, changing the configuration all the time. > >With "maximum lever" positions my pawnhash hit-rate is only about 75% after a 30 >second search on my amd64 2.2GHz box (~80% without eval hashing). > >I guess a zobrist-key pawn-hashing scheme is better here. I'm using zobrist-key hashing and my hits are even lower, 1135652 hits in 5375655 probes in a 7-ply search (first position), which seems ok in this type of position (18Mb pawn hash, 256k entries). Hmm, perhaps I should tell that I also hash the king positions with it :-) In normal middlegames my pawnhash-hitrate is above 95%. >Curious about your pawnhash hit-rates with one of the positions below. > >Btw. how many levers has each position, 8 or 14? What's in a name ... Richard. > >Thanks, >Gerd > >[D]8/k7/8/pppppppp/PPPPPPPP/8/8/K7 w - - 21.1% >[D]8/1k6/8/pppppppp/PPPPPPPP/8/8/1K6 w - - 19.8% >[D]8/2k5/8/pppppppp/PPPPPPPP/8/8/2K5 w - - 19.3% >[D]8/3k4/8/pppppppp/PPPPPPPP/8/8/3K4 w - - 18.6%
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.