Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Measured min-max branching factor

Author: James Swafford

Date: 10:07:40 10/17/02

Go up one level in this thread


On October 17, 2002 at 11:32:47, Tim Foden wrote:

>On October 17, 2002 at 07:58:00, James Swafford wrote:
>
>>
>>I did this test because I've always heard "35 or 36" as the
>>accepted bf when doing a min-max search, and Vincent has been
>>claiming that it's more like 40.
>>
>>Test 1: WAC1-100 (10 sec/mv) ==> avg bf = 34.29
>>Test 2: ECM98    (10 sec/mv) ==> avg bf = 36.85
>>
>>So, I'll stick with "35 or 36".
>>
>>Ok, a few details.  No qsearch, no null, no pruning of any kind,
>                                          ^^^^^^^^^^^^^^^^^^^^^^
>>no hashing.  Alpha was set to -inf and beta to +inf upon entry
>>to the search.
>
>Alpha-beta is a pruning algorithm.  If you are using it you should get a bf of
>roughly sqrt(36).  I assume that in fact you weren't using Alpha-Beta in your
>tests either.

Yes, I know very well what a-b is. :)  I mean no pruning of _any kind_;
pure minimax.

>
>Or do you mean set to -inf..+inf at every search _node_?

Yes.

This is nothing new of course.  I decided to measure full width bf
because Vincent was claiming it should be 40.  Knuth estimated
38, but 35 is the accepted standard these days.

Basically I wanted to see for myself.

--
James


>
>Cheers, Tim.
>
>>I measure bf after each completed iteration simply by doing:
>>  bf[iteration] = nodes[iteration] / nodes[iteration-1];
>>
>>After each problem, I compute the average of the bf's for each
>>completed iteration.  After the entire suite, I use the average
>>of the averages as the final bf.
>>
>>--
>>James



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.