Author: Gian-Carlo Pascutto
Date: 08:38:44 02/13/00
Hi all, Just for fun, I decided to convert my program to AB-conspiracy search, to see how I'd work out. The results were not very encouraging... I have several questions: first, does anyone here use, used, or tested with AB-conspiracy search ? If so, what were your results ? second, the only source of information about ABC-search that I have is the McAllester/Yuret paper of 1993. It is informative, but I am still wondering about several things. For example, I am still wondering whether the constant they use to vary from conspiracy depth to ply depth (page 14) has any effect on the actual tree shape. Also, what would be a good value for the singular margin ? A pawn ? Half a pawn ? Even less ? third, they recommend to use a quiescence search as the static evaluator. But the ABC-algorithm can do that kind of search automatically due to the way it works. So why would one use a seperate qsearch function ? My program seems to have a lot of trouble to archieve even very small conspiracy depths, resulting in very weak tactical play. This seems to improve as the move ordering gets better, which is mentioned in the article, but what about those cases were your ordering can't be possibly right ? (discovering tactical shots) Its frustrating to see the ABC search look at a position that's mate-in-three for half an hour and never have it realize what the winning move is just because it happened to look at the wrong move first. Even with good move ordering the results are terrible. I can beat the darned thing. Uh! Also, the gains of using a smaller window seem minimal. I know there are several strong programs using a variant of conspiracy-number search, so I am curious as to what advacements they use. Is there any (preferrably online) source to new developments in this domain? In short, if anybody has any experience with this kind of search, I'd very much like to hear about it... -- GCP
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.