Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: How can moveordering efficiency be measured?

Author: Vincent Diepeveen

Date: 11:07:37 01/01/02

Go up one level in this thread


On January 01, 2002 at 13:53:37, J. Wesley Cleveland wrote:

>On January 01, 2002 at 13:10:33, Vincent Diepeveen wrote:
>
>>On December 31, 2001 at 21:32:24, J. Wesley Cleveland wrote:
>>
>>>On December 31, 2001 at 05:28:54, Dann Corbit wrote:
>>>
>>>>On December 31, 2001 at 05:26:11, Michel Langeveld wrote:
>>>>>I want to check how efficient Nullmover is.
>>>>
>>>>By how often you fail high on your first choice.
>>>
>>>Note that this is somewhat deceptive, e.g. crafty gets about 90% failhighs
>>>first, but often visits twice as many nodes as the minimum. A more accurate way
>>>is to compare the total number of nodes visited to the minimun. The recursive
>>>algorithm to count this is relatively straightforward:
>>>
>>>if (failhigh node)
>>>  minimum_nodes = minimum_nodes(fail_high move) + 1
>>>else
>>>  minimum nodes = sum for all legal moves(1 + minimum_nodes(move))
>>
>>Problem to determine minimum is nullmove reduction factor.
>>It completely depends upon how you extend crazy tactical moves because
>>after tactical move your nullmove will nearly always fail.
>
>If a nullmove causes a cutoff, the nodes visited are part of the minimum.
>Hashing causes much larger problems, as a move which appears to be unnecessary
>could add hash table entries that cause cutoffs later. I think of this as a
>rough estimate more than an exact figure, but believe it is much more accurate
>than fail high first percentage.

You can never proof a minimum search tree for iteration n where n is big
with nullmove and hashtables.

That's impossible, it's even evaluation dependant.

theoretic minimum for 10 ply search with always evaluation = 0 and using
nullmove R=3 is pretty little.

like 2000 nodes or so ?






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.