Author: Anthony Cozzie
Date: 08:43:44 06/26/03
Go up one level in this thread
On June 26, 2003 at 08:11:23, Albert Bertilsson wrote: >Hi Everybody! > >Previously I've used check-extensions, and now I added single-reply extensions. >This has given me some problems because it seems that the engine tries to extend >to much now! Some positions are just searched to very few plies. I guess it has >to do with one side only having one move to do (single-reply) and the other side >constantly checking, and this makes the search go deep immediatly and not >returning until searching to max depth or getting 3-fold repetition. > >I've solved this temporarily by only allowing to extend at most Depth plies, so >when searching to ply 6 maximum search depth should be 12 plies. Anybody have a >better idea? > >Forgive me for interrupting the other threads with this computer chess related >question. This is still the computer chess forum? > >/Regards Albert 1-Rep extensions can be very time consuming. After putzing around a little I decided to limit them to Depth=1 (horizon) nodes. I consider 1-rep a quiescence extension. Zappa will regularly spit out PVs > 2x brute force depth for forced positions. At one point it solved WAC213 (36 ply combination) with a 10 ply search (and I'm sure its possible to do it in fewer ply). It looks like you encountering the selectivity tradeoff. There really is no right answer: no extension is always warranted and no pruning decision is always safe. For example: Crafty/Fritz: null move pruning, some extensions, relatively low selectivity, lots of NPS Shredder: lots of pruning AND lots extensions, very high selectivity CM9000: little pruning, a ton of extensions Junior: lots of pruning Anthony
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.