Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: rebel 10~!! super strong on amd k62 500

Author: Ratko V Tomic

Date: 23:26:12 07/27/00

Go up one level in this thread


> Why 5-6 bits per ply?  Just enough to represent an
> appoximate evaluation of the position?

No, not the evaluation but how many bits are needed to
specify the ply. The evaluation of a move strength
is implied in the move choice itself. In principle a human
evaluation of a single ply could have vast amounts
of descriptions about the strength and weaknesses of a move
(or, say, about the time used or the player's behaviour
during the game). But all such analysis of a strength of
the move follows from the specification of the move. This
kind extra information is external to the game (it comes
from sources other than the game itself, and varies in quantity
depending on the source). In other words the quantity of
information contained in the game itself (and thus extractible
solely from the game) equals the minimum number of bits
needed to specify (reconstruct) uniquely that game.
Anything else comes from other sources. Your observation
in fact highlights that this additional/contextual information,
external to the game (where the game is placed in the context of
the existent chess theory), can also be useful in judging the
strength of the players, thus adding further to the effectivness
of the expert extractor compared to the rating calculator.

The crude estimate of 5 bits/ply assumes thus 32 choices
available per ply, all equally likely (which maximizes
the entropy production per ply). In a real game that is
obviously not the case (i.e. different choices have different
probabilities, unless the set of players includes random players),
thus the entropy production is lower than the 5 bits per ply,
perhaps around 2-3 bits on a more realistic set of players.





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.