Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Trivial alfa-beta question

Author: h.g.muller

Date: 12:11:37 02/19/06

Go up one level in this thread


On February 19, 2006 at 09:34:34, Robert Hyatt wrote:

>Yes.  general formula is this:
>
>nodes = branching_factor ^ ceil(D/2) + branching_factor ^ floor(D/2)
>
>For even depths, that is simply
>
>nodes = 2 * branching_factor ^ (D/2)
>
>while for odd depths it becomes
>
>nodes = branching_factor ^ (D/2) + branching_factor ^ (D/2 + 1)

It seems you shifted D by 1/2 in the odd case: floor(D/2) equals (D/2 - 1/2) for
odd D. So the expression there is

nodes = branching_factor ^ (D/2-1/2) + branching_factor ^ (D/2 + 1/2).

For big branching_factor B the prefactor of B^(D/2) is thus either 2 or sqrt(B).
That makes a geometric average of sqrt(2*sqrt(B)), or 3.55 for B=40.

But these formulae are for counting only end leaves, if we also count internal
nodes the (leading) pre-factors are 3 and sqrt(B), for a geometric average of
sqrt(3*sqrt(B)) = 4.35 (again for B=40). So the number 5 seems a reasonable
prefactor if 'positions' refers to all nodes in the tree, and the move ordering
is perfect.

For imperfect move ordering the exponent goes up, rather than the pre-factor.



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.