Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: The firecracker crafty book

Author: Pete R.

Date: 17:05:30 07/05/00

Go up one level in this thread


On July 05, 2000 at 17:40:21, Dann Corbit wrote:

>>3) Does/will his Crafty use the "brainy" add-on, in essence being able to
>>grab a CAP score even if Crafty has gone out of book?
>
>Does not.  At 40/2, Dr Hyatt's SMP crafty can outsearch CAP, so I'm not sure it
>would be valuable at that time control.  Might be nice for blitz.

For the 40/2 example then his Crafty would in all likelihood be generating CAP
data, which could be saved after making a move if the depth is reasonable. Such
records could then be added back, and if such a position comes up in a blitz
game you're good to go.  Even in a long game Crafty may not want to waste time
searching the same position over again if the CAP record exists.  But the
reliance on CAP data could be adjusted based on the time control of a given
game. You might also want to adjust the reliance depending on whether the CAP
record is based on a single search of the given position, or is the result of
any backsolving.  If the CAP move and score was generated by backsolving, it
could be more reliable than a search at long time control.  Backsolved records
could be flagged as such to allow adjusting the reliance on the CAP data. Maybe
you could just change the depth to something large as a measure of
reliability(?) if a separate reliability field (values like GM-approved,
supercomputer run, etc) doesn't exist.

Anyway since his Crafty box is constantly living on ICC, it would be nice to
have it generate some CAP feedback that can be used to improve the book further.
 That could include very easy things like logging played positions that don't
have an existing CAP score, particularly the first half-dozen or so moves out of
book.  This would extend the CAP book in a focused way. Also, for any game that
Crafty loses, the pgn file of that game (or EPD list) should mark the last
position that had a CAP record.  This will help pinpoint critical positions to
be pounded, perhaps by supercomputer or GM analysis. Just tossing out some
ideas.




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.