OGS has a new Glicko-2 based rating system! [2017]

Reading through this list I see that a lot of correspondence only players are seeing big changes downwards in the ranking system. I only play correspondence games, and have dropped 2-3 ranks under this system.

I am not really worried as i mostly play site tournaments and ladders, but i am curious if the weightings on different speeds have been changed to create the overall rank?

I was under the impression that that system protected the loser from losing multiple points but the winners still had their individual victories?

1 Like

Nope, while that totally appeals to my sense of fairness as well, it’s actually really harmful to the rating system and causes a lot of inflation problems (I did try that out while we were analyzing the new system.)

5 Likes

Hm. I haven’t specifically looked at the breakdown of what the average rank change was across speed groups was (that’s kind of a tricky question to ask since lots of people play a mix…) I’ll look into it.

(P.S. There isn’t any weighting difference between the different speeds.)

3 Likes

Hmm well if there arent any weightings, then i guess my new rank is just that. It is not unbelievable to be honest

It might very well be, but it is an interesting observation that you’ve made that correspondence players are seeing more of a downward change. I’m not sure if it’s true or not, but spot checking just the people that I play with, it seems possible, so worth some investigation.

5 Likes

@anoek I play mostly corr, and i was myself surprised dropping a rank with this update

@anoek rose 2 ranks… sorry to destroy the pattern.

2 Likes

My rank also increased by two stones and I almost exclusively play correspondence games.

The new rank is in-line with the rank I use at Go clubs and tournaments, so I was pleased to see the change, but of course it is early days!

Mine was kept as 11k. I just noticed something weird though. As an 11k I played a 10k player, I won, and the graph shows I won against a weaker opponent, instead of against a stronger one.



Plus there’s some weird behavior for weaker players when deciding the rank. My friend never played a ranked game and was at 25k (although I’ve always told him he’s probably 18k-20k), but it put him at 13k: https://online-go.com/player/173604/
Is it considering unranked games in that case?

Edit: I also noticed it’s the same for the rest of the graph. Lots of wins against players of higher or similar (in kyu terms) level, and it says I have 0 wins vs stronger opponents in that period… Example: https://online-go.com/game/6449674 This was clearly a stronger player (13k me vs 11k him below the picture, 16k me vs 15k him in the chat) but it says I have 0 wins vs stronger opponents in this period.

My rank did not change at all. My impression is that the strange rank changes have to do with using old games to make changes now. How strong do old results impact current ratings? If I have played many games, does that mean that the results of my former opponents are more important than my current games?

One more question: How can I download my old ELO data before it is deleted?

Your game history was “replayed” entirely, so there is no difference in terms of impact on your rating.

You can use the API to get this: Player Detail Full – Django REST framework for your specific data

6 Likes

When creating custom game in rank restriction there is option ( 9 Dan + ), but option ( 21 kyu - ) doesn’t exist.
So its impossible to give 16 stones handicap to random beginner. I wish to be able to create such game sometimes.

2 Likes
  1. I am still a little confused as to whether the broken out scores are in the new glicko-2 or still the old elo ratings. If it is the latter, then if possible they should be updated to reflect the same system (IMHO).

  2. I think it makes sense to keep the breakdown of numbers permanently since they certainly do make sense. It is very relevant to me to know how my blitz rank compares with my correspondence rank, and my 9x9 (horrible) corresponds with my 13x13 and 19x19 scores. This is inspiring me to play games with more formats and time variations.

  3. I agree about the desirability of presenting numbers transformed to the kyu/dan rankings instead. I appreciate the fact that the assignation of kyu/dan rankings to glicko-2 scores is a guess, and potentially subject to change, but given that there is a current formula, why not just have a toggle for the ratings so users can choose whether to see the rankings as glicko-2 or kyu/dan?

1 Like

I think there’s a bug.

Different people are reporting that they can restrict different ranges.

Unless, of course, the “bug” is a feature, which is that you can only restrict up to 9 stones in either direction?

Unfortunately this is the exact thing that the developers are worried about: that you think that you can compare in this way.

The breakdowns are on “different scales”, which means you can’t compare them to each other.

You can only compare your 9x9 blitz rank with your opponent’s 9x9 blitz rank. You can’t (meaningfully, as the mathemeticians would understand it) compare your 9x9 blitz rank with your 9x9 correspondence rank.

Ironically, though, this limitation is true whether they are displayed in glicko or k/d.

4 Likes

It seems like there’s no “timeout” status in correspondence games anymore.
Before, if you timed out in 10 games in a row without playing a single move in between, just the first one would count. This was to prevent a negative effect, if people couldn’t play for whatever reason over an extended period of time.
With the new rating system, every game counts. This could have to do with the fact that correspondence players are downgraded on average.

That doesn’t bother me. I’d still rather know my score for a particular board size/clock precisely because they are so different.

We no longer have the T but I can assure you the functionality is the same. Except that with the new system the winner doesn’t get points now either.