Author: Jeremiah Penery
Date: 18:56:02 05/31/05
Go up one level in this thread
On May 31, 2005 at 06:42:36, Vasik Rajlich wrote: >You could try to redefine EBF so that a good extension decreases it while a bad >extension increases it. > >It would be something along the lines of "depth of important variations". Maybe Then you have to define 'important variation', and try to extend moves only in those variations. That sounds like what SE tries to do, and it requires reduced-depth searches to determine whether a variation is 'important' or not. >it's impossible to do it formally. For example "depth of the PV" is going to be >related, but not quite correct. > >Anyway, if you could do this, it would measure true search efficiency. EBF as it >is currently defined means nothing - it tells you basically if you prefer to >formulate selectivity via extensions, or via pruning/reductions. The problem is that most 'normal' extensions (check, one reply, recapture, whatever) are busy extending the most idiotic lines out as far as they will go. Sure, one of those extensions is sometimes triggered in an 'important' position and it makes the program look like it searched deeper. But in reality, 99.99% of the extensions added are in completely nonsensical lines. It seems clear to me that extensions, by default, increase the branching factor. Unless you can somehow cause the extensions only to trigger in the 'important' nodes/lines, most of the effort of extending is wasted.
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.