OGS has a new Glicko-2 based rating system!

Correspondence games between two players that rank difference of more than 2 and end in a timeout will no longer be rated.

What about games finished before the switch?

Also, this seems like a source of abuse of the system, since people can deliberately time out lost games if the rank difference is large enough.

4 Likes

I like the breakdown by speed and board size, but there doesn’t seem to be any relationship between the ratings? My overall rating is higher than any of the specific ratings, and my highest specific rating is in blitz 13x13 which I barely play and don’t really recall being any good at. I guess it’s just having a hard time giving a rating there because there are so few games? But it feels like all of these are on different scales, which makes it hard to interpret them.

I have a friend in my list who has played almost 300 ranked games and his rating is listed as N. I’m wondering how that could be as I assume that means there isn’t enough data to produce a rating for that person yet?

This was a “bug” in a sense, I was using old information for a cutoff value. If you refresh and view their profile again it should be fixed

3 Likes

And so it is. Thanks for clearing that up.

This is true, they are on different scales. This is also the reason why there are no ranks associated with the broken out ratings. To be honest it’s an open question as to whether we should keep the breakouts or not, at this point they are strictly informational.

3 Likes

I’m all for strictly informational, I’m just not sure what information I’m supposed to get out of them.

Edit: I guess if, say, you’re getting stronger at correspondence but stagnating in blitz it’ll show you that. It would help a lot if you could assign even some informal kyu ranks to those for comparison purposes, though.

2 Likes

I think at this point they are just adding to the confusion. I suspect they will be removed but that’s just my speculation.

1 Like

This was applied retroactively as it resulted in a notable improvement in the performance of the rating system.

The potential for abuse is certainly possibility and it’s something I’m monitoring, though in practice I don’t think it’s as big of a risk as one might think. But we’ll find out :slight_smile:

5 Likes

If I create game with rank restriction 10k - 9k
From which exactly glicko rating to which opponent need to be to accept it?

Hi,

I found one user 10d, is it expected?

Arnaud

1 Like

Hello, thanks for any improvement and the time you spend into this.
I’m a bit worrying, if someone is going to lose, he can ran out of time and the game is canceled, according to your rule:
Correspondence games between two players that rank difference of more than 2 and end in a timeout will no longer be rated.

1 Like

The “Overall” rating is the only one that is used, it’s what drives your rank, which is what is applicable when you talk about a rank restriction.

1 Like

Yeah, I was going to feel it out. There are a few 10d+'s floating around, we’ll see how things perform.

Yeah, it doesn’t even have to be abuse. I’m not sure, but I suspect I time out more in positions where I’m losing, just because I tend to play faster in easy positions. I have no intent to abuse the system, but if the timeouts are not distributed evenly, it will skew the results.

7 Likes

“10k” is 9.0k - 10.0k or 9.5k - 10.5k ?
Is it 1600 - 1700 for example ?
I don’t understand what I see on graph.

Well, I’m a paper tiger now. Hear me roar.

Maybe when new system stabilize we should make a survey of these ranks and update sensei’s library rank comparison page.

8 Likes

If the new system continues to hold up over time will the old one eventually be removed?

1 Like

Congratulations! Looks like this was a lot of work. Excited that now people will be matched up more accurately!

5 Likes

10k is 9.0k to 10.0k . The ratings to ranking math is not as pretty as it is under the EGF model, it’s 850*exp(rank*0.032) where 30k is rank 0. For instance, 9k is 1664.4 and 10k is 1612.0.

4 Likes