Author: Dan Newman
Date: 15:54:23 09/16/00
Go up one level in this thread
On September 16, 2000 at 18:32:26, Larry Griffiths wrote: >On September 16, 2000 at 17:04:23, Dan Newman wrote: > >> >>I once tried to measure the error rate and got no errors at all with 64-bit >>hashcodes over several hours of testing. I was able to measure the error rate >>for 32-bit hashcodes--that was about 1 false match/second (at perhaps 100k >>probes/s). I think someone came up with an estimate of (very approximantely) >>one error/day with a 64-bit hashcode at 100 knps--or was it 1 Mnps? Anyway, >>the error rate is very low and can mostly be ignored. I do try to make sure >>that such an error won't crash my program though... >> >>-Dan. > >Thanks for the info Dan! > >My piece position table is 12x64. I use random numbers currently eliminating >duplicates or zero keys in the position table. Maybe I should try to figure >some error rates like this. Great idea. > >Seems like there was a reason to keep track of castleing status in the >hash-table but I dont remember why. Might have to read my "How computers play >chess" book again. > > >Larry. you'll probably want to fold as much of the state as is reasonable into the hashcode. That includes castling status and EP target too. The only thing that generally isn't included is the game history, even though the history can easily affect the score via the 50 move and three position repetition rules. (Including those would remove much of the utility of the transposition table.) Anyway, the reason for putting in things like the castling status is that you can have two positions, one with and one without, which can easily have quite different scores if you were to search them. -Dan.
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.