Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Move Ordering

Author: Robert Hyatt

Date: 21:57:08 12/26/02

Go up one level in this thread


On December 26, 2002 at 23:15:41, Miguel A. Ballicora wrote:

>On December 26, 2002 at 10:55:55, Robert Hyatt wrote:
>
>>On December 26, 2002 at 01:34:46, Miguel A. Ballicora wrote:
>>
>>>On December 25, 2002 at 15:18:29, Robert Hyatt wrote:
>>>
>>>>On December 25, 2002 at 10:46:17, Dieter Buerssner wrote:
>>>>
>>>>>On December 24, 2002 at 23:05:09, Robert Hyatt wrote:
>>>>>
>>>>>>[...] Why don't you go read Knuth/Moore's paper on
>>>>>>alpha beta.  There you will find that move ordering does _not_ affect the
>>>>>>final score, only the size of the tree.  Something every senion-level computer
>>>>>>science student should know.
>>>>>
>>>>>I think, in most modern chess programs, move ordering can affect the final
>>>>>score. Reasons can be extensions/pruning/hash tables.
>>>>>
>>>>>Regards,
>>>>>Dieter
>>>>
>>>>If move ordering affects extensions or pruning, _something_ is broken.  As
>>>>that violates the basic premise of alpha/beta...
>>>
>>>I am really surprised by this statement. Any pruning or extension that depends
>>>on alpha will be affected by the move ordering. With different move ordering,
>>>the same position might face different bounds, hence, different extensions could
>>>be triggered.
>>>
>>That is the point.  Do you want to find something by serendipity?  I don't.
>>I want consistent behavior every time.  And if something is dependent on move
>>ordering, it is dependent on luck.  IE do you overwrite a position that
>>eliminates a hash move which eliminates an extensions?
>>
>>_not_ a good design, IMHO.
>
>Don't you have any pruning/razoring/extension method that depends on the value
>of alpha? Hard to believe that any modern program has not any of those.
>
>Miguel
>

I don't do any razoring or any extensions based on alpha, so no.  Pruning
in the q-search is done, but that is all.


>
>>
>>
>>
>>>>
>>>>Hashing _can_ cause quirks, but it actually is more important to search _worse_
>>>>moves first and then graft those search results on to better searches.  That is
>>>>how we solve fine 70 faster than we should.  If the tree were ordered
>>>>perfectly it takes 26 plies, period...
>>>
>>>IMHO, faster in plies does not mean faster in nodes (time). With terrible move
>>>ordering you could solve a position in less plies but it could be in more nodes
>>>than with good move ordering.
>>
>>
>>I am talking time to solution.  Fine 70 takes 26 plies with perfect move
>>ordering.  Most solve it at a considerably shallower depth, and there are no
>>extensions whatsoever to help, so it is simply an artifact of hashing.  And
>>the only way hashing can help is as I explained earlier...
>>
>>
>>
>>
>>>
>>>Miguel



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.