Author: Serge Desmarais
Date: 02:19:33 08/31/98
Go up one level in this thread
On August 30, 1998 at 20:56:10, Robert Henry Durrett wrote:
>On August 30, 1998 at 17:10:31, Komputer Korner wrote:
>
><snip>
>
>>Tabiya positions are nothing more than unclear positions which have an average
>>evaluation of +/=/= or 56% score for white.
>>--
>>komputer Korner
>
>KK, you left out a couple of important "details."
>
>(1) The number of times the position has occurred in recent master & GM games is
>one of the defining properties. You forgot to mention that one. Noone would
>care about a position which occurred rarely in modern practice.
>
>(2) Similarly, noone would care about a position which did not serve as a
>critical juncture point in current chess theory.
>
>Tabiya positions occur often in current GM practice and are critical juncture
>points in current chess theory. These properties are practically the defining
>properties of tabiyas. Note that unimportant positions would not recur often in
>GM play.
>
>{P.S. Sometimes, however, a position is popular, and hence occurs often, for
>years and then someone finds a new idea and the position quits occurring. After
>some date, you just don't see it anymore. Current theory is the only thing that
>we should be concerned about here. This is one of the pitfalls of blindly using
>a large collection of GM games to define an opening book.}
>
>Bob D.
Also, you forgot another element : fashion! Yes, the opening choices depend
on what the few top players play. When a champion of the world starts playing
something, you see a lot of weaker players playing the same. Then, the champion
changes and so the popular openings!
As for your pitfall, of course any computer's opening book must contain weak
or stupid moves, or even blunders! (the more positions you include, the more
chances there are to have a few stupid moves). So what? With the autolearning,
the program will just lose one game with it (it could also play it and wins!),
just like the grandmaster who did it before. If he lost with the move, the
statistics would be negative, so the program would not play it! In an opening
book containing over 1 million moves/positions, one cannot look at them all one
by one, though it is an option in Rebel who VALIDATES its own book that way. Of
course, it must also misjudge some positions, thus avoiding good gambits in the
future.
Serge Desmarais
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.