Author: Robert Hyatt
Date: 11:10:36 07/10/02
Go up one level in this thread
On July 10, 2002 at 14:05:02, Robert Hyatt wrote: I have been studying this quite a bit and have made a couple of observations that partially explain what is going on: 1. I use PVS. This means that > 99.9% of all nodes are searched with a null-window and either fail high or fail low. Which means almost all hash entries are fail-high or fail-low scores, not exact scores. If a move gets a false match and it causes a fail high or fail low, then widening the window makes that false match entry useless and the score ends up being unaffacted. 2. I was only checking to see if the best move or score for the best move changed. This is going to be hard to cause in a PVS search. It seems that the search might first become more unstable, with fail highs or fail lows (particularly on the null-window searches) that don't show up in the output at all. I probably need to measure more than just a wrong move or wrong score to see the effects of false matches better. 3. If there is someone _not_ using PVS, they could provide good information by doing this same test to see how a non-PVS search would behave. I might disable PVS in Crafty to see what that does as well.
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.