Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: re: Dann Corbit's database

Author: Mark Taylor

Date: 05:48:29 11/23/98

Go up one level in this thread


On November 20, 1998 at 20:25:13, syed wrote:

>Assuming one did make a collection of all chess games ever played in the modern
>era (this century), would it be possible to figure out the best move ever made
>for a given position in the middle game, for a given opening? Could this
>possibly be achieved a series of SQL select queries?
>
>syed

I added "self-learning" to my program, and the problem with moves from games is
that you do not know if the move(s) are any good - ***even if they have been
played by a grandmaster***. You can't even assume that the moves played by the
winning side are the best moves.

All grandmasters make mistakes occasionally - (for example) both grandmasters in
a game might miss a drawing line near the end of a game, with one grandmaster
winning (in error).  In this case the moves of the winning player cannot be
counted as "good" (nor can the moves of the losing player be considered "bad").

I designed my algorithm so that knowledge was only gained when a losing position
was reached (this must be an exact value, not an evaluation routine heuristic
estimate)... I re-searched (backwards in the tree) the sequence of moves leading
to the position and flagged those positions where a losing position could not be
avoided. Finally, the final position and all the prior flagged positions were
given a value of "LOST". I decided that when the program won **the quality of
the knowledge was inferior**, as it might have been because the opponent made a
mistake, and decided not to store it, although I might be mistaken in this.

Mark.



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.