Author: Russell Reagan
Date: 21:44:35 01/02/03
Go up one level in this thread
On January 02, 2003 at 12:05:31, Steve Maughan wrote: >Has anyone documented all the factors that affect a program's branching factor? >I notice that Tiger 15 and Shredder 7 seem to be approaching BF = 2, which IMO >is amazing. Ed also makes reference to a BF of 2 on his webpage. I remember >that a BF of 6 was the norm 15 years ago. Then hash tables came in and if fell >to 4. Then null move and it fell to 3 and now we seem to be at 2. The common >thread to all of these elements is forward pruning i.e. hash tables cut off >whole branches and null move the same. So is BF simply a function of >selectivity and the ability of a program to forward prune or are there other >elements? > >Your thoughts please, > >Steve At this point, it's probably reasonable to say that the main factor driving down the effective branching factor is forward pruning (of various sorts, including extentions, futility pruning, and the obvious things that come to mind when you think of forward pruning, like null-move). However, it wouldn't be valid to say that "effective branching factor is determined by the forward pruning methods," because as you pointed out, there are other things that came before forward pruning. Going from EBF of ~36 to ~6 is a HUGE step. I am guessing that the move from EBF of ~3 to ~2 is due to smart search techniques like Ed has been describing recently. Add in a lot of knowledge so your program can correctly determine when to reduce a search or do more accurate aspiration searches, and that kind of stuff. Those kind of semi-selective search decisions that aren't really based on some kind of "rule" or "trick" like null-move is, but things that are just common sense improvements in the framework of alpha-beta. Of course, those are common sense improvements that take an awful lot of work to discover...
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.