Author: Dieter Buerssner
Date: 09:33:25 04/02/04
Go up one level in this thread
On April 02, 2004 at 10:45:52, Heiner Marxen wrote:
>On April 02, 2004 at 10:06:10, Bernd Nürnberger wrote:
>
>>On April 02, 2004 at 09:10:06, Heiner Marxen wrote:
>>
>>>On April 01, 2004 at 07:57:27, Bernd Nürnberger wrote:
>>>
>>>>On March 31, 2004 at 12:35:21, Renze Steenhuisen wrote:
>>>>
>>>>>>(1)
>>>>>>I recognized that a different move ordering delivers very alternating
>>>>>>results in my alpha beta search. (I implemented null move, pv search,
^^^^^^^^^
>>>>>>mvv/lva, some history moves, transposition table, quiescence search ...)
>>>>>>Is this a bug or is this a normal behaviour?
[...]
>>That's just a coincidence, another example from another position:
>>
>>10. 0:00.57 459625 0.22 Rxf4 Kg3 Rf7 Kxg2 Kb4 c5 Kb3 c4
>>
>>... and with a slightly different move ordering:
>>
>>10. 0:01.78 1694656 0.01 e3 Kg3 exf4 Rf5 Rb2 Kxf4 Ka4 Rg5
>>
>>That still seems somewhat strange to me ?!
>>
>>These are not made up strange examples. A different move ordering
>>changes everything very much in my engine and I am not glad about
>>this fact, whether it's natural or not, because I still cannot
>>understand it ..
>Without transposition table that is a bug, IMHO.
>With TT anything can happen.
I think, it can also happen with pruning/extensions that depend on the bounds of
the search - for example null move. Depending on move ordering, you might search
the same position with different bounds. Perhaps once null move fails high (but
a "real" search would have shown, that that fail high is wrong). Now you search
with a slightly higher beta the same position, and the null mvoe search may not
fail high again.
Perhaps Bernd can easily check this, by disabling null move and TTs.
Regards,
Dieter
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.