Author: Uri Blass
Date: 02:45:00 12/19/03
Go up one level in this thread
On December 19, 2003 at 02:27:17, Steffen Jakob wrote: >On December 19, 2003 at 01:24:27, Russell Reagan wrote: > >>On December 19, 2003 at 01:00:31, Steffen Jakob wrote: >> >>>I repeat my posting from below because the ruffian thread pushed it very fast to >>>the bottom of the message list. :) >>> >>>What are you using for the hash key in your distributed perft implementation? >>>How do you make sure that there are no hash key collisions which are possible in >>>the usual zobrist key approach? Those collisions are too rare to influence the >>>playing strength of a chess engine but would make the result of your perft >>>project invalid. >> >>I remember Albert saying that he uses 128-bit hash keys, which is not >>theoretically sound, but should work in practice. Deiter also uses hash tables >>for this I think. Maybe he can tell us what he does. > >I like this distributed perft project very much (and contributed 4 solutions to >subproblems ;-) but the only reason why we are doing this is to get the *exact* >number of lines. Even if it is wrong by one line then the result is wrong and >the whole effort was rather useless. Even if the result is correct then we >cannot be sure about it. Therefore I would propose to run a validation without >hash tables. Can it be estimated how long this would take? > >Best wishes, >Steffen. I do not see a reason not to use hash tables when it is possible to use hash tables and be safe with 192 bytes. I guess that if there is an error it is not because of hash tables but because of other reasons. Uri
This page took 0.01 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.