Author: Stephen A. Boak
Date: 18:55:01 10/27/99
Go up one level in this thread
On October 27, 1999 at 11:55:49, Amir Ban wrote:
>You can do as you please, but if you are too selective in your opponents, you
>can't claim your rating to be realistic.
>
>As I said, I play crafty partly because there's no one else there of roughly
>equal rating.
>
>I don't understand this about trying to increase my rating. When someone with a
>high rating thinks that playing him is exploitation, he's really saying that his
>rating is inflated. If the rating is realistic, it will hold.
<snipped>
>Amir
You make a one-sided point, but ignore the other side of the coin.
If *both* ratings are realistic, *both* ratings will hold up after playing
each other. If *either* rating is unrealistic, then the ratings *may* not hold
up after playing each other. I say 'may' because it depends on whether two
unrealistic 'in the same relative way'.
And whether a rating of any single program is 'realistic' depends on whether
it plays a group of opponents with ratings that are *on the whole* realistic
('on the whole realistic' means some specific opponents may be unrealistically
high and some may be unrealistically low, but overall the ratings are
realistic).
Any program that plays a subset of all possible opponents *may* have an
unrealistic rating, depending on whether that subset has *unrealistic* ratings,
on the whole.
Any program with a realistic rating that plays a subset of all possible
opponents when the subset is unrealistically rated overall, will tend to obtain
an unrealistic rating also.
Theoretically, any two programs with realistic ratings, whether relatively
close or relatively far apart, will on the average maintain their realistic
ratings, and their relative rating distance, even it they are 600 pts apart, or
400 pts apart, or 200 pts apart, or 0 pts apart.
This is not true, however, if they play each other at 'unrealistic' times
(i.e. when the other has an inflated or deflated current rating). :)
I acknowledge, however, that program ratings may not be transitive--i.e it may
not be always true that if A beats B, and B beats C, that A must beat C (I'm not
talking about winning in single games, but in long matches). In a battle of any
two specific, realistically rated programs, their relative difference in ratings
may not hold up--because their realistic ratings and the relative rating
difference may not truly reflect the non-transitive results applicable to their
battle.
Bottom line--if programs don't play under controlled circumstances (and the
variables are not always easy to control, due to Hardware, Software,
Connectivity issues), then bragging rights regarding playing strength are
subject to critique no matter which specific program is being discussed. No
play lists as well as manually choosing which opponents to play and when, may
affect any program's rating. I say 'may' because...well, see above paragraphs.
In human chess tournaments, I never complain that my own rating is inflated or
deflated, i.e. unrealistic, even though I believe I sometimes play sandbaggers
who artificially keep their ratings low, so they could more readily win big
tournament dollar prizes. Such players must have lost points to other players
who tend to now be unrealistically 'overrated'. Maybe for every sandbagger I
lost to, I beat another sandbagger in another tournament where there was little
prize money at stake and they were 'trying to lose'.
My rating is based on playing many, many opponents of all types--high rated,
low rated, same rated. Some are overrated when I play them, some are
underrated--I will never truly know the difference, and from a topview, perhaps
their ratings are, on the whole, realistic, and perhaps then my own rating is
realistic. Hey, but who is perfect? :)
--Steve
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.