Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Symbolic: First pattern instance generation

Author: Steven Edwards

Date: 00:19:22 03/28/04

Go up one level in this thread


On March 27, 2004 at 19:19:56, Roger D Davis wrote:
>On March 27, 2004 at 18:42:04, Steven Edwards wrote:
>
>>I've now moved into the pattern instance territory.  There's only a few basic
>>boot strap pattern recogizers so far.

>hey, for those of us that "don't LISP"...what is a bootstrap pattern recognizer?

A pattern recognizer (in the above context) is a function that attempts to
produce pattern instances (i.e, filled in templates) for a particular pattern.

A bootstrap pattern recognizer is a pattern recognizer that does not rely upon
any previous instances having been already generated.  Bootstrap patterns are
named such as they must be run before any pattern that needs pre-existing
instances (of other patterns) for operation.

A bootstrap pattern in Symbolic is roughly the same as a primitive pattern in
Paradise.

>Does that mean, for example, that Symbolic will explicitly attempt, at every
>ply, to search for possible knight forks?

No.  Special recognizers applied immediately after a node expansion may, upon
matching, eliminate any further consideration of the subnode.  Examples:
tablebase hit, mate in one hit, and only one move hit.

Checking for possible knight forks is done only after some combination of
earlier recognizers determines that it would be worthwhile.  This idea is used
throughout the whole pattern matching code with the intent to minimize overall
calculation, just as a human would.

>How will this be superior to current methods?

With the goal of minimizing the number of search nodes generated, Symbolic
should come out far ahead of any traditional A/B searcher.  With the goal of
having a comparable playing strength, it is far too early to tell.

>Keep going, dude...gotta admire for you doing this.

Thanks.  Things should become more interesting in a few months when I start
adding real chess knowledge.  (Symbolic needs a working plan generator first.)
I like to think that a lot of the knowledge part is already done: it's in the
form of chess books like _Winning Chess_ by Chernev.  Texts like these show how
to recognize patterns and construct plans that can guide a search several ply
deep with a near unity branching factor.  My main job then becomes one of
translation: English to Lisp.



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.