Author: Robert Hyatt
Date: 11:09:34 08/28/98
Go up one level in this thread
On August 28, 1998 at 11:08:20, Robert Henry Durrett wrote: >On August 28, 1998 at 09:59:15, Robert Hyatt wrote: > ><snip> >>I hate to throw a damper on this, but it won't work, period. The first reason >>is that if the tactical engine rejects the positional engine's best move, the >>positional engine won't have a "second-best" move to try, because alpha/beta >>doesn't produce that. To obtain it, you have to do a full re-search, but with >>the rejected best move eliminated from consideration. >> >>To make this work, you take a *huge* performance hit, and will play far worse >>than if you do what we are doing now... ><snip> > >I guess what really matters is how often the "positional" engine would have to >be restarted "to do a full re-search, but with the rejected best move eliminated >from consideration." If that did not have to happen often, then the objection >may not be valid. If it happens once, it costs you a factor of at least 2X, because the positional engine then has to do another search... and the tactical engine then has to search this to a deeper depth tactically... Note that they can't search in parallel, because the tactical searcher has no idea which move the positional search will propose as best, and the tactical searcher will only be able to accept or refute that move *after* the positional engine has proposed it. Otherwise, the tactical searcher will simply find a "best move" (which may win nothing) and then only prove that all other moves are <= to that move... which helps *zero* with refuting the move the positional engine proposes as best, unless they happen to match. In short, it doesn't work as expected. And you get better results with a single engine, because otherwise you have to divide the 3 minutes per move up into "x seconds for a positional search" followed by "y seconds for a tactical confirmation" and also allow time for a refutation and positional/tactical test on a different proposed move...
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.