Author: Eric Oldre
Date: 10:21:38 07/19/04
I've been using the technique described in Bruce Moreland's site. http://www.seanet.com/~brucemo/topics/pv.htm This seems to work pretty well except in some situations. When the trans table lookup returns an exact value that is between alpha and beta, then I don't know what the moves are that follow the current node, and I get PVs in my thinking output that look like this. 1 time nodes mov1 2 time nodes mov1 mov2 3 time nodes mov1 mov2 mov3 4 time nodes mov1 mov2 mov3 mov4 5 time nodes mov1 mov2 mov3 6 time nodes mov1 mov2 mov3 mov4 mov5 mov6 7 time nodes mov1 mov2 mov3 8 time nodes mov1 mov2 mov3 mov4 mov5 mov6 mov7 If I change the code to only allow trans table cutoffs of alpha or beta, then the PV is constructed correctly. It seems like this might be indicating an issue with my trans table implementation becase it seems odd that it would have a deeper value in the trans table for a position that is in the PV. Has anyone else had a issue with this? If so, what was the issue? or did you discover that it wasn't an issue at all? I have checked that my zobrist keys are being updated correctly by computing them from scratch at each node, so I don't think that is the problem. Eric Oldre PS... any update on when the CCC Search Engine will be online again?
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.