Author: Paul Clarke
Date: 06:50:40 07/01/04
Go up one level in this thread
On June 30, 2004 at 20:17:49, William Bryant wrote: >>On June 30, 2004 at 19:50:23, William Bryant wrote: >> >>>I am looking for suggestion or references for setting the Delta value in the >>>QSearch used to filter captures. Those captures that don't exceed the Delta >>>value are pruned even though they may be winning captures. [snip] >My old program used the followng: > >long Delta = alpha - DeltaWindow - (MATERIAL); >gen_caps(Delta); > >where DeltaWindow is an amount of tolerance and MATERIAL is the material balance >at this point Could you adapt the idea Ed Schroeder describes at http://members.home.nl/matador/chess840.htm#LAZY%20EVAL ? He does full evaluations at pre-frontier (*) nodes and uses the difference between the full and quick evals at those nodes to compute the safety margin for lazy evaluation at the leaf nodes. It seems reasonable that something similar could work for your delta window, as the difference between full and quick eval gives you an idea of how important positional factors are in the current position. (*) I mean the nodes at the last-but-one level of the normal search - hopefully I'm using this term correctly.
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.