EGF Rating System Commission Report 2020

gennan · **#21**

An upper limit to grades was not discussed in the commission, but the new formulae imply an upper limit of 13d/21p EGF, assumed (roughly) to correspond to perfect play. A theoretical entity of this level would be able to give 5 or 6 stones handicap to a top European player (~8d/4p EGF) and 3 or 4 stones handicap to a World champion (~10d/11p EGF). With this assumption, top AI level is assumed to be about 12d/18p EGF.

Oberlappen · **#22**

I think there should be also thought making a new rating class, for tournaments with top players. The class A to C is not bad for normal tournaments I think (I don´t mention class D cause it´s only online), but in my opinion the system should be completed, wit han class S (for Special) or however someone would call.
We have some tournaments for top players, with way more time per move, then in average tournaments. The typical class A tournament, got 1 hour + canadian Byo Yomi (for example the typical 15 stones in 5 minutes), but there tournaments way more time to think, 90 min. + Byo Yomi or even two hours + Byo Yomi and in some even the Byo Yomi got a quite exceeded time, compared to other tournaments.
When we see top players, we can see that it´s quite a difference there, if they play just 1 hour with Byo Yomi or 90 min + Yomi or even more. In the longer games the stronger player reigns more suprior. In the more average timed tournaments, the outsiders do more often beat the favourites on high Dan levels, then in games with more time.

So it does matter for really strong players, if they have jsut the normal class A tournament time or the much bigger time from a Championship. Here really shines the strength of the good players and the outsiders has to do more, to get the win, cause the favourite here can mobilize more ressources with more time. I think an class S rating should be establieshed with tournaments who have at least 90 min and + can. Byo Yomi 15 stones in 5 min. (or the fitting pendant in the Fischer Time). Class S should then count 125% on the rating, instead of the 100% from class A tournament I suppose.
In the end, I think this would be a good solution. And it would be nearly exclusive for top players (only exception is the EGC, where more players got so much time).

betterlife · **#23**

gennan wrote:

An upper limit to grades was not discussed in the commission, but the new formulae imply an upper limit of 13d/21p EGF, assumed (roughly) to correspond to perfect play. A theoretical entity of this level would be able to give 5 or 6 stones handicap to a top European player (~8d/4p EGF) and 3 or 4 stones handicap to a World champion (~10d/11p EGF). With this assumption, top AI level is assumed to be about 12d/18p EGF.

Nice so we should extend to 12d if the robots are allowed to join!! Currently the strongest amateur player rank is 8d. The system can handle 9d but it's accutally not correctly implemented so atm an 9d would be downgraded to 8d by an egd rating manager.
If a player is entered as new then he gets these initial ratings:

Code:

5d   2500
6d   2600
7d   2700
8d   2800
9d   2670 <= this seems wrong
1p   2700
2p   2730
3p   2760
4p   2790
5p   2820
6p   2850
7p   2880
8p   2910
9p   2940

gennan · **#24**

It does seem reasonable to me to have 8d as the maximum Entry Grade for human amateur newcomers to the rating system.

Perhaps it should even be 7d? Most of the current players with ratings around 2800 have started at an Entry Grade of 7d and worked their way up by winning a lot of games against other top players. 8d is not an amateur grade that is easily aquired in China or Korea (I think you need to be something like top 5 amateur nation wide, which is quite formidable). Only in Japan it seems to occur more frequently, but Japanese dan grades are significantly weaker than EGF dan grades. 8d in Japan could be anything from 5d EGF to 7d EGF.

So I think it should not even be possible to accept 9d amateur as a first Entry Grade. It should only apply to human amateurs that can consistently beat quite strong pros (around 8p or somewhere between #200 and #400 on https://www.goratings.org/en/?). I don't know if such amateurs really exist. Does any of you know a name of such a player, who is still alive and actively playing in tournaments? I think he would be pretty famous (like Nie Wei Ping around 1981). In the exceedingly rare case where such an amateur player participates in an EGF tournament, it would justify manual intervention by the EGD manager to set his first Entry Grade to 9d amateur.

For pros it is a different matter. I think they should just declare their official pro Grade and the rating system should initialize their rating to the EGF rating corresponding to that pro Grade, like @betterlife's list (which is already the case, I think).

Javaness2 · **#25**

At the minute, professional players have their rating reset when they are promoted. Resets seem very desirable for improving players, but I don't immediately see why they are wanted for professional players. Is there any plan to change this part of the system?

gennan · **#26**

This was not discussed in the commission, but pro grades are not earned lightly, so personally I feel that resetting seems quite harmless to do in the few cases where the EGF rating is not quite as high as the corresponding pro grade rating. In fact, resetting for professional grade promotions may even be considered a good opportunity to calibrate top EGF ratings.

jlt · **#27**

Many tournaments are being canceled due to Covid-19. This will worsen the deflation phenomenon. I am afraid that if the situation continues like this for some time, ranks will become meaningless.

gennan · **#28**

How does tournament cancelling worsen deflation?

jlt · **#29**

If many people improve but don't play in tournaments, once they come back to competing they will become sandbaggers.

gennan · **#30**

It has always been the case that (on average) players improve between tournaments and this would inflate ratings if nothing is done to compensate for improvement. The EGF rating system tries to compensate by allowing rating resets (when promoting 2 ranks or more) and by giving a small rating bonus for each tournament game played. The correct size of the bonus is a bit of a guess, because the actual average improvement is hard to measure and it probably varies over time and over different regions. With statistical analysis, we found that the current bonus is probably too small, so it will be increased when the EGF rating system will be updated. But it's still a fairly crude mechanism.

During the corona crisis, I don't know if tournament players (on average) are improving at the normal rate (whatever that is) or better. Some tournament players will study and play a lot online and improve, but many others will improve less or even decline, because they have less motivation to study when clubs and tournaments are closed, they don't like to play online and/or they don't learn as much from playing online.

But if some players know that they are really improving a lot during the corona crisis (because they have great success playing online), they can just apply a rating reset in the 1st tournament they play afterwards and that will fix the problem. Unless ofcourse self-promotion is prohibited by the national go association of the player. Maybe such go associations might create a temporary policy to consider promotion requests with some online evidence given by the applicant?

jlt · **#31**

If (say) 50% of people improve by 1 stone and 50% don't improve, the former won't have their rating reset because 1 stone difference is not a clear evidence of improvement, so the whole rating system will be deflated by one half stone.

gennan · **#32**

I think it's very optimistic to assume that 50% of tournament players improve by a stone during the corona crisis (even while they can only play online). A large majority of tournament players is at a plateau (for years) and that won't suddenly change during the corona crisis.

gennan · **#33**

It is possible to fix this by changing the rating bonus mechanism. Now it works by giving a bonus for every tournament game. If this is changed to give a bonus over time (regardles if you played tournament games during that time), then it would be possible to compensate for average player improvement even during a pandemic.

jlt · **#34**

Some people are in favor of developing online tournaments and rate them with a full coefficient, but this solution is not ideal because of cheating or disconnection risks.

jlt · **#35**

gennan wrote:

I think it's very optimistic to assume that 50% of tournament players improve by a stone during the corona crisis (even while they can only play online). A large majority of tournament players is at a plateau (for years) and that won't suddenly change during the corona crisis.

Looking more closely at the EGF rating system, I think you are right. Around 5 kyu, the expected gain after one match is con*epsilon/2 which is a bit less than 0.4 rating point. If we estimate that 60000 rated games are played each year, and that 6000 people are EGF rated, then each rated player plays about 10 rated games/year, meaning that the system assumes that a player gains on average 0.04 stone/year.

The rating system does not necessarily correspond to reality, and my calculation is very rough, but after 1 year or a bit more of corona crisis, the drift will probably be less than 0.1 stone, which looks fine and could easily be corrected by adjusting the parameter epsilon at least for a while.

(I was imagining that many sandbaggers would appear all around but this probably won't happen.)

gennan · **#36**

I think those estimations are roughly correct.

In the upcoming changes to the rating system, the bonus will be increased for kyu players (it will be about 1 rating point around 3k, 5 rating points around 15k and 8 rating points around 30k, which is about 2 to 3 times the current value). We think this larger bonus is closer to the real average player improvement per tournament game.

Plugging that increased bonus into your rough estimations for the average tournament player improvement per year, the predicted values are still fairly small (0.1 stone per year around 3k, 0.5 stone around 15k and 0.8 stone around 30k).

gennan · **#37**

The announced EGF rating system update has taken place now:

Lorenz Trippel (AKA betterlife on L19) wrote:

(email sent to EGF Board and EGF Members on 5-4-2021)

Dear all,

in the recent years it showed that the calculation of the european go ratings needs a revision. It could be observed that there was a deflation in the ratings and consequently also in the european go ranks. It's natural that the ranks in Europe are bit higher than the ones in the United States or Japan but this gap even increased by the inherent deflation of the system itself. Some players did some research on this and finally Dave de Vos put up a webpage on goratings.eu to show the effect and propose a new formula. This new calculation algorithm has now been implemented. For stronger dan players there is often almost no difference, for the big bulk of players around 1 dan there is a change of bit less than 1 rank upwards, so for example a 1 kyu with 1992 rating points (GoR) has now 2068 GoR and can play as 1 dan, a difference of about 60 to 100 points. For a single or double digit kyus this change can make up to 140 rating points, it depends bit how many tournaments / games he has played.

We also expanded the rank floor from 20 kyu to 30 kyu. The intention for this change is to give beginners an easier start into the world of rated games as the usual case till now was that a beginner not only lost most of his games but also stayed for a long time at 20 kyu.

It's now more than a year that me and Julien Corcessin took over the european go database. At the moment it's bit more quiet than usual but still there is a lot of daily work validating the tournaments results and doing small improvements as the ones you see now. We still have a lot of work in lifting up the whole system to current web standards. If you are knowledged in the areas of PHP programming, databases, web design or simply want to help update the documentation or help validating tournaments please don't hesitate to contact us! There is also room for new projects like e.g. having an integrated web application for doing the tournament pairings instead of entering them in an offline java program.

Please, if you know some programmers let them know that we are very interested to have more people in our team!

I would like to say thanks to all members of the rating commission making this upgrade possible and a special thanks to Dave for all the work he has done.

Cheers,
Lorenz (EGF Secretary and EGD rating manager)

https://europeangodatabase.eu/EGD/

PS: Until the AGM in 2022 online tournaments can be entered as class C if they are well organized with proper anti cheating measures and national online championships can be entered as class B.

Adin · **#38**

I've noticed an important change, points gain/loss seem to be much lower than before. For example some years ago I had a nice tournament where i gained 109 points rising from 2k to 1k. But in the new system I only gained 59 points.

It was mentioned before in this thread that on average a player has roughly 10 rated games/year, which means roughly 2 tournaments. Now that it's become more difficult to gain/lose points and with players having 2 tournaments/year on average, doesn't it mean that the gap between real strength and EGF rating risks getting bigger? Essentially having a lot of underrated players (due to slowed down rating increase not being able to match improvement) and also having more overrated players.

jlt · **#39**

Edit: I misread your post, ignore the italicized part.

=== begin comment to misread post ===
The problem of the EGD rating not keeping up with the player's progress during the corona crisis was discussed a few messages earlier:

https://lifein19x19.com/viewtopic.php?p=260066#p260066

In a nutshell, the majority of players are not improving anymore, and those who are improving quickly can ask their rating to be reset.

On the other hand, you can't say there are no tournaments anymore. We recently had two Corona cups, the End Polio Tournament, the Paris Tournament and many others.

=== end comment to misread post ===

In the new rating system I managed to gain 49 points with the "End Polio" Tournament which is class C. So I think the rating can keep up with your progress even with a couple of tournaments/year.

gennan · **#40**

Marcel Grünauer wrote:

The formula at https://europeangodatabase.eu/EGD/EGF_r ... php#System does not mention handicaps. I looked around that page and this thread but couldn't find anything.

How are the rating changes affected if the player gives or takes a handicap?

Handling of handicap stayed mostly the same: handicap increases the rating going into the expected winrate formula by 100 * (H - 0.5), so it assumes handicap elevates the level of the handicap reciever by the size of the handicap.

EGF Rating System Commission Report 2020

Who is online