Author: Robert Hyatt
Date: 05:32:48 11/02/98
Go up one level in this thread
On November 02, 1998 at 04:49:55, Ernst A. Heinz wrote: >On November 01, 1998 at 23:46:42, Robert Hyatt wrote: >> >>On November 01, 1998 at 13:47:18, Ernst A. Heinz wrote: >> >>>On November 01, 1998 at 12:56:12, Inmann Werner wrote: >>>> >>>>you are right. my percentage is about 50% in normal game. I have to think about >>>>it. >>>>On the one hand, if I make a hit, the evaluation is extremly fast. On the other >>>>hand, if I only hash the pawns, I get much more hits, but have more to do in >>>>much more difficult in evaluation, because good king standing and rook standing >>>>depends on pawn structure. >>>>When I have a little time, I try both and look whats really faster. >>> >>>Werner, >>> >>>All pieces that tend to move around *frequently* will greatly hurt the hit rate >>>of your Pawn hashing. Therefore, I recommend that you never include any major >>>or minor pieces therein. The Kings, however, are different "animals" which can >>>safely be hashed together with the Pawns as long as they must stay behind safe >>>shelters in the middlegame. "DarkThought" drops them from the Pawn hashing >>>*dynamically* as soon as King safety does not matter anymore. >>> >>>=Ernst= >> >> >>I did this in Cray Blitz, but believe that the current scheme I use in Crafty >>is far better. With the king included my hash hits were in the 90% range or >>so, while with it out, 99.9% is common. I looked at what I was doing, and >>it was mainly king safety related. And the king safety with the king at g1 >>is no better/worse than the king safety with the king at h1 or h2, if you only >>look at pawns. I now do normal pawn scoring, then king safety on *both* sides >>of the board, and store all this with just the pawn hash signature. When I >want >>to evaluate king safety, I look at which side the king is on and start with >that >>raw pawn structure score around the king... and then factor in other things >>(like is there a queen, how close are the pieces, are there open files with >>rooks on them around the king) and so forth.. >> >>I now get 99.9% hash hits which covers the king safety pawn structure... and >>there's no loss in accuracy of any kind... > >We must be doing something right then because we also get more than 99% >hit rates with the King factored in. :-) > >=Ernst= how many entries? IE I never paid attention to this hash number when we ran on a 16gb machine because we used such rediculously large hash tables. But if you can get such a large hit percentage with pawns+kings, then the original idea of pawns+kings+rooks ought to be pretty close, because even the king can walk a long way in a 12 ply search... however I generally run with 10mb pawn hash, which is something like 500K entries, and don't ever see this number under 99% (integer truncation)... I'd think it has to be bigger to stick at 99% with kings factored in...???
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.