Comments/Ratings for a Single Item

Just saw Gary's Comment. I usually play no more than five casual games at a time, concentrating on them as if they were tournament games. Hence my (misleading) high rating.

Very nice. Interesting. I may also help to pair players, finding fair
partners and also more challenging ones. Of course, the wide variety of
variants makes this rating less 'firm' than chess ratings. 

I think that Gary's suggestion to allow a game to be rated or not would
be a nice addition. This could be a choice made at the start of the game,
allong with time controls, etc. This would also allow some players to
develop rated 'niches', such as Shogi, or Shatranj-variants, while
playing other games 'just for fun.' Some may just want to be rated
generalists. Others may not want to be rated at all.

Fergus, please revise the 'DRAW EFFECT' in your algorithm, I suspect
there is a bug. 
Generally speaking, The method used IS NOT good, it is important the
rating of the contendor to make changes in the RATINGS after the game,
this is usual in Chess and other games. ELO is one of the best measures,
and it has proven to be really indicative, as it has been confirmed by
experience. You can see how, in some international tournaments, players of
different countries that have never played before against others in the
Tournament neither in the same other players environment or usual
contendors, can play more or less according to the forces indicated by ELO
rating. A good way to do it in our case can be assign, initially,
'preliminar ELO´s' according to the results in POINTS over NUMBER OF
GAMES, using any reasonable function. After that, use ELO method over the
history of the last 365 days, and it can be PONDERED by a factor, say 0.5,
if the game is NOT a Tournament game. In this way, we can obtain a more
reasonable measure. My Chess ELO has oscilated, over my past 20 years
history, between 1800 and 1950. I don´t expect this is necessarily my
rating in different CV´s, but I doubt it sounds reasonable enormous
deviations from these values, and if you observe my results against some
players better ranked with your method, you can think that something may
be wrong, and by these reasons I have some reasonable doubts about the
method used and its goodness for the purposes.

The idea of retro-active ratings is a displeasing 'after-the-fact' concept
to me.  The reason has to do with coffee-house style care free games versus
tournament rated games.  When I played chess at the Edgewater Park
Invitational, for example, I only lost 1 game, I drew against a 2000 USCF
rated player and only lost to a 2300 USCF rated player.  The 6 games
involved 10 hours of playing.  Now, I also play at a coffee-shop.  Wild
fun games with just 15 minutes on the clock.  These are not rated and we
can try bold ideas that we would likely avoid in a rated game.  By
comparison it as as if someone comes up and says, 'Hey, we've just rated
your coffee-house games and fudged them into your USCF rating.'  In regard
to the 1500+ rating I saw by my name here, by comparison: Best USCF was
1943, Rating in MSN real-time play broke above 2000, rating against Chess
computers reached 2110.  Current Gothic Chess is about 1975.
In regard to Chess Variants I would like to see the following system,
which seems fair:  All players start at 1600 (or 1500 if preferred). Rated
games do not begin until New games are started, say to go in effect Jan.
15th (or some other date).  In-progress games which may have been for fun
or simply experimental, do not count.  As example: When I played a
fun-game of Maxima with Roberto, he gave me lessons during the game.  Very
helpful lessons.  Would that have happened in a rated game?

Gary is right, many games people have played have been coffe-games, test
games or fun games in which result was not important, but it can be used
for a 'first' number, as follows. This is s pseudo-ELO idea, and it can
be good for us:
At first, we need a 'preliminar' measure, but it is going to be modified
after the first calculation, and it is going to be an adjusted measure with
time, once all people is concious about how it works. The first number is
A= 1000+ (Points/Number fo games)*1500, for everyone. Points is calculated
as usual, 1 for win, 0.5 for draw. Number of games refers to
LOGS in the last 365 days. After that, we can run an algorithm, sequenced
by time, and in each game the rating is modified as follows: If player A
has a highest rating than B, and he wins, his rating is modified with the
added change rate= K*(2500 - Rating of player A)/2500. K is a factor,
usually 1, but it can be modified in tournaments, to, say, 2 or 3. This
change rate is also substracted of the player´s B rating. If Player B
wins, his rating is modified with the added change rate= K*(Rating of
player A - Rating of player B)/C. C is a number that must indicate how
fast we need reflect the 'force change' in a player. I suggest C=100,
and K is the 'Tournament factor'. This rate change is substracted from
player´s A rating. In case of draws, the last rule applies, but with rate
change divided by two. An unrated player is considered, at first, with a
rating of 1000, unless he gives some evidence of another rating, and it
can be used, translated to our scale. Try some examples, and you can see
that this is a very reasonable measure.

The following website discusses rating calculations. It may be of use in
the CV Ratings project.  The site includes example calculations.

http://www.chess-express.com/explain.htm

(please, bring the option for 'unrated games' back as soon as possible) btw, i agree, the intention was good, top points for that, i just don't know why you didn't talk to the players/members beforehand, which would of been nice, or am i wrong, did i somehow miss this conversation?

Roberto, please carefully read the first paragraph on this page and the
warning above the 'Game Courier Ratings for *' table. You will see that
we are in agreement on the point you raised, but it is not relevant to an
evaluation of the method used to rate players here.

Christine, I will refer you to an earlier comment of mine dated
2006-01-08. We are in agreement that game ratings should be reserved for
games intended to be rated. But, for the time being, I am using what data
I have, which are the logs of previously unrated games, to test the method
I am designing to rate players.

What I wish to focus on right now is whether this method is an accurate
enough measure of relative playing strengths between players. In this
respect, it is looking good to me, but there might be ways to tweak it.

Christine,

Although it would seem unfair that you are rated lower than any of the
players you have defeated, further examination of the evidence reveals
that it is fair. Your three opponents have all played several more games
than you have, and they all have much stronger track records against their
opponents than you have against yours. As long as you have played few
games, this method works with the assumption that your performance was
probably due to luck. As you play more games, as your opponents have, it
begins to place more trust in your performances against other players.
Like Elo, the method used here is self-correcting. As you play more and
more games, it is better able to come up with a more accurate estimate of
your relative playing ability. Since you have played only a few games, it
doesn't yet have enough information to make a good estimate.