Author: Robert Hyatt
Date: 08:08:02 04/07/04
Go up one level in this thread
On April 07, 2004 at 09:14:40, martin fierz wrote: >On April 07, 2004 at 08:56:26, James Swafford wrote: > >>On April 07, 2004 at 06:55:31, Andrew Williams wrote: >> >>>On April 07, 2004 at 06:49:59, Renze Steenhuisen wrote: >>> >>>> >>>>Hi all, >>>> >>>>could someone give me some numbers that are common with hashkey collisions? >>>>Because I guess my % is little too high... >>>> >>>>I'm getting like 0.03% [which is 1 every 3000, if I'm not mistaken] >>>> >>>>This is when using TT=32MB (haven't got the exact number of entries) >>>> >>>>If you think it is an error, any suggestions on where to start looking? >>>> >>>>Thanks! >>>> >>>> Renze >>> >>>One in 3000 seems very high. How many bits are there in your hashkey? >>> >>>Andrew >> >> >>Even though you said you're using Crafty's random num gen, >>I would start by doing some hamming-distance checks. >> >>For reference, my program gets: >>Checking minimum hamming distance between random keys: 14 bits >>Checking average hamming distance between random keys: 31 bits >> >>If your hamming distances are comparable, you can conclude >>your zobrist keys are ok, and go from there. >> >>-- >>James > >i never understood why people think hamming distance is a good measure for the >quality of random numbers. e.g. for 8-bit numbers i can produce a collision with >the numbers > >a = 11111000 >b = 11100011 >c = 00011011 > >because b^c = a. the mutual hamming distances all come out to 3-5 :-) > >cheers > martin It is about the chess tree. Burton Wendroff and Tony Warnock wrote a paper published in the JICCA years ago, which addressed this topic... They explained why this is important. Ideal hamming distance is 64, but there are only two 64-bit numbers with this property across the entire set...
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.