[Freeciv] ELO and ranking

[Arnstein Lindgard <a-l@xxxxxxx>]

This in response to Horn Gabor's ranking system efforts.

Per Mathisen wrote:
> Problem: Best way to reach the top is to only fight weaker players.
>
> This is not the kind of behaviour we want to encourage. I do not know ELO,
> but is it not possible to penalise playing too many consequtive games
> against only weaker players?

Mark Metson wrote:
> I guess if the top people are absent so you cannot actually fight them 
> maybe some kind of challenge+timeout could be applied so if you reach the
> rank just below them and issue a challenge to them and they do not answer
> the challenge within some reasonable period of time they default their 
> title or something like that...

Horn Gabor wrote:
Quite a lot!

Mike Kaufmann wrote:
> no. all games should be ranked. if someone wants to not be ranked, he
> should play as a guest.

Jacky Mallett wrote:
> Most players are fairly cognisant of who plays from which ip, which in


Hirisov, thanks for pointing me to this thread. I'm not sure if I can
help with your formula, but one thing should be pointed out here.

The ELO system of the international chess federation is based on
local _tournaments_, set up with the Monrad Swiss pairing system or
something similar. The local chess club honcho sets up the rounds,
matching opponents. (Honcho can also participate, since that system
is completely deterministic.) Games played outside of tourneys are
not normally _ranked_ with the central authority FIDE or USCF.

I don't think ELO, as such, copes with the idea of players picking
fights wantonly.

If you stick to rankings without enforced matchmaking by system, you
may have to apply even more creative mathematics to adopt ELO to
Freeciv, keeping the above in mind! You probably deserve an award if
you figure it out :)

What I'm thinking of all this, looking around the Internet, is that
there seems to be a certain supply of people who like running web
pages. For example there is this new Polish fansite that wants to set
up long-term games and so on. Considering the superior contents of
Civfanatics and Apolyton, it's not likely to compete very well.
However, if someone somehow got the idea of making a Freeciv tourney
site, I think that would be a success.

Then you'd get a better regulated system for producing serious
rankings.

If you take the time to look at the following system,
I'll demonstrate how it can:

- Encourage varied matches.
- Be socially facilitating.
- Regulate ELO more sensibly.

Suppose 20 players sign up for a tournament, and everyone gets to
play 7 duels. The first round is set up at random. Winners are then
matched against winners, and vica versa. You may only meet one
particular opponent once. Each game scores you 1, 0.5, or 0 points.
At the end of tourney, you also get a secondary score which I  call
"Quality points". This is the sum of all your opponents' points. When
two players have the same score, Quality is used to determine tourney
rank. (You have achieved the same score, but with tougher
opposition).

Player Bob's biggest ambition is to score higher than his arch-rival
Bill, both mediocre players. Bob wins the first game, and is then
informed that his next opponent is Hirisov. Death and drek! I can't
possibly beat him, and feel very unlucky. But there is consolation;

A) Because I'll loose, my next game will be an easier opponent.
B) Because Hirisov's likely to win lots of games, my Quality will
increase.

Meanwhile, Hirisov is thinking "Yum, easy points", knowing that only
0.5 points may decide the tourney winner.

So in the third round, Hirsov meets Warlock. If you're "lucky", Bob
will even want to watch the game, cheering frantically for Hirisov
because he wants more Quality points. Damned barbarians! Unfair
autodip!

I think this system puts more fun into matches that would
otherwise be pointless (and therefore never played), and thus
regulate the matchmaking (both ways) for games that are used to
calculate your ELO. Long-turn Freeciv? Arrgh!

Google.. turns out Mr. Elo came from Hungary and this page popped out
which discusses precisely this "inclusiveness" vs. "effectiveness" of
pairing systems.

http://www.chessbase.com/newsdetail.asp?newsid=260

I understand some are hoarding points by newbie-bashing, while I find
myself only playing duels against one other guy. You can't make
effective statistics out of that, you definately need some rotation.

Remember Elo's system is mathematics based on "logistical
distribution" I think they call it. It deals with expected results
from playing a variety of opponents, and you don't get to pick fights
unless you're the World Champion or something.

And why does it always take so long to set up those team-games
anyway? Someone just has to administrate things and give incentive..

Administrating a Monrad-like tourney could be as easy as asking
players to submit results and filling out some tables on a website.
You just have to make sure all games of round 1 are finished before
you start round 2.


Arnstein,
sorry for not coding lately.