Author: Larry Griffiths
Date: 09:34:55 09/17/00
Go up one level in this thread
On September 17, 2000 at 03:54:23, Johan Melin wrote: >On September 17, 2000 at 02:46:34, Larry Griffiths wrote: > >>I tried using my compilers rand() function but it was only generating >>random numbers from 0 to 0x7fff. >> >>I ran 78 positions from the BWTC suite >>and got over 10,000 mis-matches using my 64-bit hash keys. >>I looked at my smallboards and they usually differ by a piece missing or in a >>different position. My Piece table does not have zero or duplicate entries >>and the random numbers in the piece table look good. >>I built hashkeys using the two different smallboards and they do indeed >>generate the same hashkeys. >> >>Maybe my hashkeys should be prime numbers? >> >>I will play around with random some more and see if it changes my mismatch >>rate. >> >>Thanks Bob. >> >>Larry. > >With 10,000 mismatches something must be broken. How good are your random >numbers? Try counting the bit-difference between all pairs of 64-bit keys. >The average should be around 32 and the minimum above 12. > >/Johan Melin Hi Johan, I built 64-bit keys with prime-numbers in each 4 bits (4 x 16=64 bits) and also with each nibble (4-bits) containing 2 bits on (using 3,5,6,9,10,12) and this seemed to cut the mismatches between my small boards in half. I also tried a 128-bit hash key but it did not reduce the mismatches by much. I think BWTC position 256 is one of the suite tests that really eats my lunch. This position causes 5961 mismatches in rapid fashion in my hashing. It looks like this... [D]r3kb2/ppp5/7r/3p4/4p2q/5n1P/PPPP1PP1/RNBQ1R1K b q - bm Qg4 Qxh3+; id "BWTC.0256";
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.