Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Branching factor of alpha-beta

Author: Robert Hyatt

Date: 10:00:15 08/23/01

Go up one level in this thread


On August 23, 2001 at 12:44:42, Sune Fischer wrote:

>>>I just upgraded my little engine from using a negamx to an alpha-beta algorithm.
>>
>>Negamax _is_ alpha/beta.  It is just re-formulated so that + scores are
>>always good for the side on move in the tree.  Normal alpha/beta has +=good
>>for odd plies, -=good for even plies, which makes the code messier.
>
>I thought that negamax was minimax, only "re-formulated".
>
>

It is.  When I first read your question, I never considered that you would
really be doing pure minimax.  I caught this a bit later (below) and kept
going.



>>Normal alpha/beta should give you an effective branching factor of about
>>sqrt(x) where x is the effective branching factor of pure minimax.  That
>>ought to be somewhere around 6-10 max.
>
>I thought that was a theoretical best and only if I have a perfect moveordering
>(which of cause I never do).

You can get _very_ close to theoretical best, using the normal move ordering
ideas like killers, captures, hash moves, history moves, PV moves, etc...



>
>>Most everyone uses negamax as the framework.  Whether you use simple
>>alpha/beta on top of that, or something more sophisticated (I use PVS for
>>example) doesn't change the look of the basic search that much.
>
>Okay thanks.



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.