Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Fail-low pruning

Author: Bo Persson

Date: 15:07:27 01/17/06

Go up one level in this thread


On January 17, 2006 at 17:55:41, Tommi Rimpiläinen wrote:

>On January 17, 2006 at 17:26:59, Dann Corbit wrote:
>

>>These methods are theoretically unsound, but work well in practice.
>>
>>A method is said to be unsound if a search to depth K fails to find the best
>>move at depth K.  Alpha-Beta will not fail to find it.  It will find exactly the
>>same move as a full minimax search would.
>>
>>Other sorts of reductions like null move do not provide the same guarantee.
>>
>>So "technically" they are unsound.  But they work very well in practice.
>>
>
>Unless we're dealing with a mate prover, the search results may always be
>disputed. That's why, I think, people should not lay too much emphasis on
>theoretical soundness. What sould be pursued instead, is that the evaluation
>reflects the real value of the root node as propably as possible. But this does
>not mean that the unsound techniques should not be justified by some vangue
>reasoning.
>

Yes, part of the "work well in practice" is that a null move search doesn't
search to depth K, but to K+1 or K+2. So, even if it is "technically unsound",
it still works well, on average.

I find it amusing that chess programming is very much about being good "on
average", the opposite of real-time programming (where you have to minimize the
maximum, to guarantee a response time).


As others have pointed out, a level K+2 search without null move is even better,
but how do you achive that?  :-)


Bo Persson



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.