Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Pattern Matching -- Avoiding Hand-Tuning

Author: Don Dailey

Date: 14:58:43 01/23/98

Go up one level in this thread


We have  "Method of analogy" code in Cilkchess.   We don't think of it
as
pattern matching although I guess it is.   Basically we determine with a
very
complex algorithm that a search will come out the same from more than
1 position.  For instance if you move a rook from a1 to b1 behind a wall
of
pawns there is a very little chance the score will vary significantly
given
a shallow search of 2 or 3 ply.   I think a lot of micro's use something
like
this but I am out of touch.  The code will prove the positions are
"tactically
identical" in the context of a shallow search of specified depth.

- Don

On January 22, 1998 at 16:14:53, Stuart Cracraft wrote:

>On January 22, 1998 at 14:39:12, Bruce Moreland wrote:
>
>>
>>On January 22, 1998 at 14:13:40, Stuart Cracraft wrote:
>>
>>>In other words, retain the tactical cruncher aspect of
>>>the program. But when two moves are approximately equal
>>>tactically, based on the score, use the pattern matching
>>>to break the tie.
>>
>>Lots of guys have suggested climbing this Everest, or a siimilar one.
>>
>>1) Find the mountain.
>>2) Find a sherpa.
>>3) Climb the mountain.
>>
>>Everyone seems to stop somewhere before step 3 is started.
>>
>>Some problems are:
>>
>>1) Identifying which patterns might exist in general.
>>2) Figuring out that one (or which one) might apply in this position.
>>
>>These are both really really hard, aren't they?
>>
>>bruce
>
>You forgot step 2a). "Get the oxygen canisters".
>
>That's what makes this kind of pattern-matching interesting.
>The Russian Kaissa team had an idea of the "method of analogies".
>I wonder if that's been used more generally (or researched.)
>
>--Stuart







This page took 0.04 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.