Author: Dan Newman
Date: 17:31:16 02/13/01
Go up one level in this thread
On February 13, 2001 at 08:28:33, Andreas Herrmann wrote: >On February 13, 2001 at 07:37:41, Gian-Carlo Pascutto wrote: > >>On February 13, 2001 at 07:25:52, Andreas Herrmann wrote: >> >>>>It depends on whether you apply the extensions before probing into >>>>the hashtable or not. >>>> >>> >>>Means this, i can use both ways? >> >>As long as you keep things consistent. Same position at same >>point in tree should be same hit in hash. >> >>>If i extend before the hash probe, i have to store depth as depth+extension? >> >>Indeed. >> >>-- >>GCP > > >Is one of the above ways better than the other (depth with or without >extension)? > > >Thanks for your answers. I ask this because i have a bug if i use upper and >exact bounds. Sometimes my chess program spends the opponent a piece (a queen, a >rook or so) for nothing. >From the above ways i tryd in the past very often both ways, but the bug was in >both cases. I think i have to look to other reasons for that? > >I often ask me, could this be a hash kollision? But i think not, because this >happens in about 1 of 3 games. I only uses 32-bit hash keys because Delphi >cannot manage XOR with 64-bit values. But i think many other chess programs uses >only 32-bit hash keys and have not such a bug. If i play the same position in >what this happens i never got the wrong move in the pv. So i cannot reproduce >the error. > > >Andreas Do you use only 32-bits for the whole hash code, or is this just the key part so that you have 32+n bits for the hash code? (n is the number of table index bits.) I've measured the error rate for 32-bit hashcodes in my program, and it's about one error per second on a 500 MHz P3 or about one error per 100k probes into the transposition table... With 64-bits it's probably less than one error per day. -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.