Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Trojan Horse (crafty)

Author: Robert Hyatt

Date: 17:49:22 07/21/04

Go up one level in this thread


On July 21, 2004 at 11:46:18, Matthew Hull wrote:

>On July 21, 2004 at 11:29:10, Robert Hyatt wrote:
>
>>On July 20, 2004 at 22:02:46, Derek Paquette wrote:
>>
>>>I saw bob talking briefly about this 'coding'
>>>what exactly is it?
>>>
>>>does it sacrifice a pawn if it thinks that the person its playing is just
>going >>for a draw?
>>>how exactly does it work
>>>
>>>thanks in advance,
>>
>>
>>The trojan horse attack is a generic position where the computer has castled
>>(usually king-side) and the human has not.  The human plays something like
>Ng5, >the computer responds with h6 to drive it away.  Rather than retreating,
>the >human plays h4 and if the computer plays hxg5, the human plays hxg5 and
>has a >terrific attack down the h-file.  The trojan horse code in crafty simply
>gives a >huge penalty for taking the piece and opening the file, to avoid the
>problem >completely.
>
>
>I think there are related positions where crafty is vulnerable to attack on the
>kingside, and not only with the trojan horse -- positions similar to the
>Falcon-Crafty game where blacks pieces are all on the queen side and white has
>bishops, knights and queen ready to sac and pillage on the kingside.
>
>You hardly ever see the commercial progams walking into setups like that.  But,
>I've not studied it enough to know to what extent the wide ICC book is to
>blame.


Crafty has some second-order and third-order eval terms to catch that, but I
won't begin to say it is well-tuned yet.  IE it certainly understands the idea
that the more pieces are involved in the attack, the more exponentially bad the
score gets. But it is certainly probable that it is poorly tuned at present.

And the book certainly can hurt.  IE for a couple of months I have played with a
_wide_ open book.bin file on ICC.  No books.bin/bookc.bin, so that Crafty
wouldn't play its "normal openings" and reveal much about recent tuning changes
for the WCCC.

Still there were reports of commercial programmers having 19.14 to use to test
book lines against while preparing openings during the tournament.  I'll address
that next year but I won't say how. :)




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.