Author: Volker Böhm
Date: 09:21:34 08/30/04
Go up one level in this thread
Hi, depends on your chess engine. I use nearly "allways" but exclude pices that are pinned to king. This is quite fast for me to compute. SEE has errors. The more correct algorithm for computing the value of a capture move is quiescense. SEE tries to predict it´s value. It may improve your SEE if you can reduce the amount of errors cheaply. The errors that occurs most often are: 1. The attacking piece is one of two hanging pieces. Both of them cannot be saved in one move. Thus capture something! 2. A defending/attacking piece is a pinned piece (pinned to king, pinned to queen, pinned to undefended piece) 3. A piece is defending more than one piece that can be captured. For move ordering the errors are not relevant. For foreward-pruning in q-Search it is a larger problem. I had many problems to implement a foreward-pruning in q-Search with SEE that increases the strength of Spike (our chess engine). The standard rule prune if SEE < 0 does not work for Spike. Greetings Volker
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.