Author: Tony Werten
Date: 00:04:38 04/16/01
Go up one level in this thread
On April 15, 2001 at 17:35:27, Alvaro Jose Povoa Cardoso wrote: >On April 15, 2001 at 14:39:12, Steve Maughan wrote: > >>I haven't done any tests but intuatively I'd go for 64 bit. The number of >>unique positions in checkers is not that much different to chess (maybe 10^80 as >>opposed to 10^120) and the number of positions searched per sec is higher than >>chess. Remember that a 64 bit signature is 2^32 time safer. >> >>Steve Maughan > >That must me the reason why I get 48 hash collisions on a 50 million node >search. (I check the hash move for legality). >I thought that if I have half the squares of chess and 4 different pieces >(instead of 12 for chess) I could get away with 32bit hash signatures. > >Alvaro Cardoso Could be, you are only using 1/6th of the chessnumbers so you should be able to make higher quality hashnumbers, but probably it's still not enough. The best way IMO is to use 64bit, use fe the right 18 bit for indexing, shift the hashnumber 18 to the rigth and store the right 32 bit (the signature). You are then using a 50 bit hashnumber but you only store 32 of them. cheers, Tony
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.