2019 Ratings Motion

[Javaness2]

Of course, the highlight of the European Calendar is the Annual General Meeting of the European Go Federation. Half the delegates have no idea what is going on, and the other half don't agree about what is going on. Something like that.

This year's agenda is here https://docs.google.com/document/d/1hR8 ... VV5dQ/edit
There seem to be only 2 motions, both of which are about the rating system.

Motion 1: Extend the ratings down from 20 kyu to 40 kyu.
Motion 2: We've seen the thread on this already here on LifeIn19x19, it's an investigation by Dave de Vos goratings.eu/Home/About

Motion 1 has been defeated on about 4 occasions already. I see no reason for that to change this time around, but I have a lot of sympathy for the idea. If 20 kyu to 15 kyu is a sea of inaccuracy, why not push that pollution down to 40 kyu? Is the fear that players are just somehow going to get stuck between 40k and 30k and never get promoted out of a region of dark instability? If that really happened, the world would be a crazy place. Sure it would be a throwaway rating, but children like to have ratings. What is the real harm in throwing a 34 kyu rank away?

Motion 2 probably has a good chance of passing, perhaps with a demand to validate very carefully the data first and a Christmas present for Aldo.

I'd love to see Motion 3 along the lines of "The EGF resolves to adopt a consistent approach towards promotion practices and rating practices."
Today we can have a 20 kyu going 6 wins 0 losses in a tournament. Depending on where he lives he can have (at least) 3 different treatments for his rating:

Standard: His rating changes only according to the algorithm. His next tournament rating will be his tournament exit rating.
Reset: He won too many games. After processing the results accordingly, with his rating at 20 kyu, we that for his next tournament rating he will be reset to 18 kyu.
Doctored Reset: He won too many games. We change his entry rating to 17 kyu and process the results accordingly. His next tournament rating will be his tournament exit rating.

Each of these 3 approaches will create different ratings for the players involved. In a rating system, it just seems really wrong that you can have these 3 different approaches under the guise of "but it is up to the national federations to decide what rank a player has." Do they want to have GoR or don't they?

[jlt]

40k seems a bit low, but many ranking systems like KGS have had 30k ranks for years without problem. Not only children, but also adult beginners, may prefer not to have a rating below absolute zero.

[EdLee]

a rating below absolute zero.

...but zero sits between 1k and 1d and is ( confusingly ) skipped.

[Javaness2]

In this context 1 kyu and 1 dan are labelled as ranks.

At the moment a 40 kyu will probably not have a zero rating. He'll most commonly have have a falsified rating. After entering at 40 kyu, he will be reported as having entered as a 20 kyu in the submitted ratings file for the tournament. In order to defend the integrity of the rating system this process of falsification is described as correct by various commentators. Think about that for a minute.

This confusion is the second example to consider in the context of Motion 3.

[jlt]

I think that according to the EGF rating system, 1d consists of ratings in the interval [2050,2150).

I don't know about AGA. Is it the interval [0,100) ?

Anyway, I don't mind being below freezing point as long as my temperature appears on the thermometer.

thermometer.jpg (11.14 KiB) Viewed 13778 times

[HermanHiddema]

IMO, the rating system should just not have a lower bound. It is artificial and unnecessary.

Mathematically, you could make a rating system where rating zero represents "perfect play" and higher rating are worse (e.g. 1000 = 1d, 2000 = 10k, 3000 = 20k, etc). But in practice that might confuse people (though higher = worse is currently already used for kyu ranks).

In practice, you could e.g. add 6000 points to all EGF ratings, remove the bottom, and That makes 9000 = 10d/9p, 8500 = 5d, 8000 = 1k, 7000 = 11k etc. with no bottom.

[gennan]

I'd love to see Motion 3 along the lines of "The EGF resolves to adopt a consistent approach towards promotion practices and rating practices."
Today we can have a 20 kyu going 6 wins 0 losses in a tournament. Depending on where he lives he can have (at least) 3 different treatments for his rating:

Standard: His rating changes only according to the algorithm. His next tournament rating will be his tournament exit rating.
Reset: He won too many games. After processing the results accordingly, with his rating at 20 kyu, we that for his next tournament rating he will be reset to 18 kyu.
Doctored Reset: He won too many games. We change his entry rating to 17 kyu and process the results accordingly. His next tournament rating will be his tournament exit rating.

Each of these 3 approaches will create different ratings for the players involved. In a rating system, it just seems really wrong that you can have these 3 different approaches under the guise of "but it is up to the national federations to decide what rank a player has." Do they want to have GoR or don't they?

I'm all for examining the rating system and possibly update it in such a way that it handles such cases automatically. For example, bayesian inference and / or machine learning might work quite well to make a robust system that even proposes promotions (and perhaps demotions).

But before we can start fiddling with it, we first need to establish if people are even prepared to consider updating the system. In general, people don't like change.

[Pio2001]

Hi,
I'd like to know how many players are ranked with the european system. French players are not. They are ranked by the french system, and all the ratings appearing in the european list for them are just a copy of their french rating (updated with the games of their last tournament).

If we remove players from countries having their own rating system, how many are left, that are truly rated by the european system ?

[Javaness2]

Hi,
I'd like to know how many players are ranked with the european system. French players are not. They are ranked by the french system, and all the ratings appearing in the european list for them are just a copy of their french rating (updated with the games of their last tournament).

If we remove players from countries having their own rating system, how many are left, that are truly rated by the european system ?

I think only Hungary and France have internal rating systems.

[gennan]

Hi,
I'd like to know how many players are ranked with the european system. French players are not. They are ranked by the french system, and all the ratings appearing in the european list for them are just a copy of their french rating (updated with the games of their last tournament).

If we remove players from countries having their own rating system, how many are left, that are truly rated by the european system ?

There are about 50.000 players in the European system and this includes players from countries that have their own rating system.

French rating list: https://ffg.jeudego.org/echelle/echelle ... trgroupe=0
European rating list of France: http://www.europeangodatabase.eu/EGD/cr ... dgob=false

These lists are different, so I don't think the French ratings are copied into the European list. But French tournament results are probably sent to the European system, so there should be a good agreement between these rating lists.

Belgian rating list: https://www.gofed.be/rating
European rating list of Belgium: http://www.europeangodatabase.eu/EGD/cr ... dgob=false

[jlt]

The rating calculated by the French federation (FFG) is analogous to the EGF algorithm, but there are three main differences:

• Ranks go down to 30k.
• If a player has won too many games during a tournament then the algorithm automatically resets his rating prior to the tournament.
• If a player has lost too many games during his first rated tournament, then the algorithm automatically resets his rating prior to that tournament.

The rank calculated according to the FFG appears in the European Go Database under the "declared rank", while another rank is calculated according to the GoR (Go Rating). Both ranks are usually close but there may be some differences, in one direction or another.

[Pio2001]

The rank calculated according to the FFG appears in the European Go Database under the "declared rank", while another rank is calculated according to the GoR (Go Rating). Both ranks are usually close but there may be some differences, in one direction or another.

Oh, I see now. Thanks.