Author: Bruce Moreland
Date: 09:33:27 11/29/00
Go up one level in this thread
On November 29, 2000 at 08:23:56, Severi Salminen wrote: >Hi! > >Robert Hyatt said that when he tested the performance differences between SEE >and MVV/LVA he saw a 10% advantage for SEE. Is this 10% really worth it when we >consider the fact that SEE might miss some tactical shots, like pins and other >check involving sequenses? This might be also true if we use standing pat >cutoffs in qsearch, but generally. Has anyone pitted a version of his/her >program using SEE against MVV/LVA in a long match (more than 100 games at >least)? What were the results? > >Severi I saw a lot more than 10%. You get better move ordering, and you get pretty good forward pruning at the tips. I used a piece list (one for each color) that is sorted low to high. I have a pawn list but it isn't used in the SEE. What I do is count the pawns attacking the square by looking at the board squares that can attack the destination. Sometimes that's enough to tell you the result of the sequence. If it is more complicated I have to go through the piece list and process rays. That's expensive, although you might be able to add data structures that will help you. It's very messy and not a lot of fun. bruce
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.