Author: Alessandro Damiani
Date: 02:18:21 01/01/00
Go up one level in this thread
On December 31, 1999 at 16:50:42, Dan Andersson wrote: >Yes and no! The implemetation in the paper does. But you could easily find some >other function of the board to substitute for it, I did. The first one only >extended if the king was in check (easy if you use bitboards of any form), a >function of the number of replies to a check and king safety. But if you use ETC >the search space shrinks (giving a saving of nodes searched) as a function of a >hopefully better AB window and the penalty for generating moves is not so great. >But then its imperative that you implement a limit to how near the leaves you >use ETC. > >Regards Dan Andersson Thoughts: Conspiracy theory is based on the static evaluation. Taking into account only some terms of the static evaluation we are able to bias the search and therefore define on what selectivity is based. Important terms are: material, kingsafety, passed pawns. We could stage the search like this: - layer 1: full static evaluation (improve everything) - layer 2: material, kingsafety and passed pawns (improve important things) ABC is doing a one-ply search to decide whether to terminate the search. It looks one ply into the future. Dan, do you use scores from quiescence search like described in the paper? Perhaps SEE is good here. Alessandro
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.