OGS has a new Glicko-2 based rating system!


#81

I tend to agree looking at your history.

What I’m seeing (and I’ve seen this on a few other profiles as well) is that you’re playing a lot of stronger opponents, which is great for getting better, but has resulted in a lot of consecutive losses. Glicko, for better and worse, tends to accelerate your rank change when you have a lot of consecutive wins or losses.

I’m going to have to think on this a bit. It’s come up a few times now and it’s a a very legitimate approach to getting better at the game so I don’t like that the system is punishing players who play a lot of stronger opponents.


#82

I’ve read the debate in the thread in regards to timeouts and their potential/actual effect on the ratings system. In the end I think that not counting timeout wins/losses in the same manner as “regular” wins/losses in an effort to avoid artificial rating changes is the wrong approach for a few reasons (some of which have already been mentioned):

  • Abuse - No matter what system you put in place there are always those people who value the number next to their name above accuracy. If you change it to some other variation, they’ll simply adapt and exploit the new system. Trying to outsmart these players is wasted effort.

  • Rating inflation/deflation - Any rating formula that doesn’t have a positive and a negative outcome will lead to the average rating for the site drifting up or down which is always a bad thing.

  • Unintended consequences - Tweaking a data set to alleviate one issue often leads to changes that affect the “normal” results in unintended ways (e.g. rating inertia or the opposite - larger than expected rating swings). I applaud the effort to understand the meta issues and to do the best thing possible to achieve accuracy. I am hopeful that there will always be discretion applied when considering changes. I love the fact that this is being done in a transparent manner and that feedback is considered.

Ratings are great, I eagerly awaited finding out what my “new” rating was going to be when the new system went online (yay, I’m went up 3 kyu!). That being said, this is primarily a site for playing games and learning the game. What the ratings provide for me is a guide to the strength of thousands of potential opponents/instructors. I am just learning the game so I am far away from the concerns of those who strive to achieve the highest ranks here. Those in that position would certainly view the ratings and the method of calculation much differently from me.

Keep up the good work.


#83

1492 ± 158 means that we’re 95% sure your real rating is somewhere between 1334 and 1650 (1492 - 158, and 1492 + 158)

That’s very interesting about the ranks having the fractional parts there. What browser / device are you using?


#84

ah i forgot, sorry. Opera (v46.0), Windows 8, PC


#85

hm…You are saying the same glicko-number is a diffrent ranks in diffrent gamingpools (such as boardsize and timesetting), right?

This makes it just more difficult to interpret, without help :flushed:


#86

Ah, that could very well be the case! :slight_smile: I mostly play correspondence tournaments and most of the time, I’m the lower ranked player in those games because of the tournaments I choose (automatic site wide, not explicitly beginner). Those losses accelerating the rank change sounds pretty logical to me.

When I started playing I really hated losing and it kept me from learning. When I play higher ranked players I really don’t mind losing as much as long as I’m learning. It’s a way for me to not value my own rank as much as I used to.

I’m sorry :stuck_out_tongue: Seems like a pretty complex thing to factor in the new rating system. Let me know if you need anything from me :slight_smile:


#87

850exp(rank0.032)

Because people seem interested in this formula, I made a table of how all the ranks translate to ratings:

30k	     850.0
29k	     877.6
28k	     906.2
27k	     935.6
26k	     966.1
25k	     997.5
24k	    1029.9
23k	    1063.4
22k	    1098.0
21k	    1133.7
20k	    1170.6
19k	    1208.6
18k	    1247.9
17k	    1288.5
16k	    1330.4
15k	    1373.7
14k	    1418.3
13k	    1464.4
12k	    1512.1
11k	    1561.2
10k	    1612.0
 9k	    1664.4
 8k	    1718.6
 7k	    1774.4
 6k	    1832.1
 5k	    1891.7
 4k	    1953.2
 3k	    2016.7
 2k	    2082.3
 1k	    2150.0
 1d	    2219.9
 2d	    2292.1
 3d	    2366.7
 4d	    2443.6
 5d	    2523.1
 6d	    2605.1
 7d	    2689.8
 8d	    2777.3
 9d	    2867.6
10d	    2960.9

#88

I’m a little confused.

Yesterday I was 11 kyu (overall). Then I became 9 kyu, but…

  • in the homepage I’m 9k (9.0 ± 1.5)
  • in the thumbnail of my games I’m 11k (vs16k)
  • in the game page I’m 10k (vs 21k)


#89

Regarding @cloudedthoughtz situation:

Is it possible that the old system gave a more fixed amount of points for beating new players, while the new system gives fewer points for beating new players? It seems that most of their wins are against lower ranked players and most of the losses are against much higher ranked players (at least 5-6 ranks).

It’s possible that as a mostly site tournament player, fewer low ranked players 1) stick around for a long time AND 2) have an established rank AND 3) continue to join site tournaments.

Perhaps one method to improve could be to periodically try add in some quick match correspondence games? Maybe this could find more players closer to your level who have established ranks who you could win against, allowing for progress to show better.


#90

My situation is the opposite. My rank jumped from 12k to 6k, but all of my past opponents have been DDK. I think I’ve always been playing (and defeating) weaker opponents, so now my ranking is possibly over inflated. But I guess that should get corrected once I start playing stronger players.


#91

Hello,

I have seen a 3d player becoming 7d and 5d players becoming 8d or 9d. So, “jumping” from 12 kyu to 6 kyu looks like a reasonable jump :slight_smile:

Arnaud


#92

I created custom game with 9k - 9k rank restriction but 7k accepted it


bug

Update: in next game 8k accepted


#93

Just a question about the new graph:
387 wins vs. weaker opponents
75 wins vs. stronger opponents
97 losses vs. weaker opponents
200 losses vs. stronger opponents

But on the old statistics:
Ranked games played: 886
Won: 552 Lost: 334 Draws: 0

By doing some hardcore maths, we can see that these numbers don’t match out. Does that mean that 127 of my ranked games have been against players who are exactly the same strength than i am, or is the new system somehow missing 127 of 886 games? Or is the old 886 wrong?


#94

I’ve heard automatch is using the old ranks not new ones, but I’m not sure.


#95

I have jumped from 4k to 1d. I think the next games will be very boring for my playing partners.^^’’ But the rank willl be corrected soon after a couple of lost games I think


#96

Except that many people like you who were 4k will now also be 1d, so it might not change anything…


#97

It’s just that the old ranks are still appearing in some places, yet to be cleaned up.


#98

@_KoBa That can be because some corr games are not taken into account because of that timeout rule anoek added (and probably will be modified again in the near future)


#99

Bug? I only played 2 games and win both but the chart says I lose 1 game to weaker opponent.


#100

… also, you only have one ranked game, not two!