Author: Gerd Isenberg
Date: 10:56:25 04/14/04
Go up one level in this thread
On April 14, 2004 at 10:03:04, Alvaro Jose Povoa Cardoso wrote: >Hi, >I've been playing with the following history heurist variant: >Along with the PieceType-From-To formulation I added 8 bits that correspond to >the rightmost 8 bits of the hashkey. >So the new index to the history table is as follows: >HashKey-PieceType-From-To >HashKey is only 8 bits as said before. The idea is to try to match as much as >possible the moves that failed high with the 'true' posições where that happened >since in the classic hitory heuristic aproach we have little or nothing that >says that those history moves found at the current node where previously stored >in the exact same position as the current one. In other words history moves >don't mean they came from the same position. It is exactly this I'm trying to >avoid. >However when I tryed this idea it failed miserably with much higher node counts. >My question is why did this failed? >Is this a wrong idea from the very beginning? > >Best regards, >Alvaro Cardoso Hi Alvaro, Funny idea, but are you aware how huge those tables become? 256*PieceType-From-To, even if you pack PieceType-From-To to 4000 and not 6*64*64 (not considering color). If you mean 8 bits from zobrist keys, there is not much similarity if those bits are equal, except the same position, so you may have a kind of an additional killermove shared by 256 positions with huge memory resources. May be your idea works better with hashing some material properties eg. the parity of all major pieces for each side, or better some bits less, eg. only two with #allpieces % 4. Regards, Gerd
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.