2021 Rating and rank adjustments

Well, I’d like to see the quote from anoek, but I can state some general principles: It is now becoming more and more clear in stochastic inference that classical null-hypothesis significance testing (NHST) is questionable as decent science. NHST is likely not what anoek engaged in, and even if it was, it seems not relevant here.

I would let anoek’s good model play out both above and below 25k. If all 25 kyu players win half the time against other 25 kyu players, there is no reason to change anoek’s floor of 25k. All of these players will remain 25k.

On the other hand, if some 25k players win 1/4 their games against other “25k” players, they should be maybe 27k or similar. If some “25k” player win 1/10 of their games against the top 25k players, they are probably 29k or lower. I don’t know anoek’s exact general algorithm for handling this, but 25k is too high for the moment for players that generally lose against other 25k’s. I want all “25k” players to find players at a similar level, or an approximately equalizing handicap against stronger players.

In other words, there is no reason for anoek or anyone else to worry about formal statistical significance. What he already has in place for 15 kyus is great when compared to assuming that all “25k”'s have equal playing strength.

3 Likes

Do you have 25k friends or something? How do you know how it is down there?

[duplicated post removed]

S_Alexander asked,

“Do you have 25k friends or something? How do you know how it is down there?”

I’m trying to benefit all current 25k players. That’s my primary goal here. If you have a category of player that you think will lose out in any way in my proposal, please let us know.

2 Likes

Probably could be its own thread, but I also do not understand the 25k “floor” heh. I doubt it matters though since anyone who plays enough to get an accurate rank will inevitably be stronger than 25k…

1 Like

I’m not completely following all this but doesn’t this mean that 25k’s are also generally winning against 25k’s?

But also discussion here:

And here:

Although I suppose these have been over taken by events somewhat

5 Likes

Thank’s a lot !
Nice game !
Maybe it’s time to me to give money.
I’m a debt for all those beautiful things. I’m passed long and Great time in OGS.
long life for OGS

5 Likes

25 posts were split to a new topic: A separate ranking pool for children?

I like rank clipped to 9d and 25k because it’s pretty.

2 Likes

I have split part of the discussion from here into: A separate ranking pool for children?

As it seemed to deserve its own topic.

This thread is already massive enough, please try to stay on topic and consider creating a new thread if you have any related idea.

If I have moved or deteled something you think belongs here, I appologize, please simply repeat your point again.

Many thanks.

10 Likes

Does anyone have a good idea why ratings changed? Ranks are understandable. But I lost 150 rating points, maybe. Is it because ratings are more stable (fixed volatility bug), and now they don’t push each other apart as much, so they bunched up towards average a little?

1 Like

my suspicion is exactly that, a bit like cooling down a fluid: when all the particles/players are moving less quickly and bumping into each other less the volume contracts

3 Likes

Hi, this may no longer be an issue, but I thought I’ll ask to get a confirmation, just to be sure.
btw, This is not a big deal, no urgency whatsoever. I do hope this gets cleared up eventually.

on the first day when this adjustment came live, I “created” a custom game, and restricted my opponent’s ranking. say, I only wanted to play with one of the 12k players, I specified the min and max rankings in the set up. To my surprise, a 6k was able to accept it. I suppose there’s a logical explanation for that, but it’s a bit confusing. I haven’t tried it since, so I don’t know if a 6k can still accept a 12k game, I hope not.

could you please advise what is the proper way to set up these rank-restriction games under the new system? Thanks.

1 Like

But those other dan players will also have had their ranks similarly changed, so if they were 1d when you were 4k and you are now 1d, they will be (around) 5d, so they will still have to give you handicap stones and still find it (approximately) just as hard to beat you.

1 Like

I compared the European Go Database records for some well-known British Go Association players with their new OGS ranks and the OGS ranks look a bit inflated. However, I note BHydden’s answer regarding proposed changes at the EGF.

Andrew Simons/Uberdude:
https://www.europeangodatabase.eu/EGD/Player_Card.php?&key=14337147

(4d/9d)

Matt Marsh/marshmn:
https://www.europeangodatabase.eu/EGD/Player_Card.php?&key=15213671

(4k/3k)

Gerry Gavigan/gerrysw11
https://www.europeangodatabase.eu/EGD/Player_Card.php?&key=16333592

(11k/6k)

Sandy Taylor/afar
https://www.europeangodatabase.eu/EGD/Player_Card.php?&key=15086049

(2d/5d)

1 Like

Andrew didn’t play any rated game on that acount for 7 years and Sandy only played 6 rated games on that account in the past 7 years. So those two may not be the best examples.

After the coming EGD update, Matt’s EGF rating will have fluctuated around 2k in the last 3 years. His new OGS rating has fluctuated around 1k OGS in that same period.

After the coming EGD update, Gerry’s EGF rating will have fluctuated around 10k for the last 4 years. His new OGS rating has fluctuated around 6k in that same period.

Here is a different case where is used some estimate from play quality:

From this small number of anecdotal cases, the following rough comparison table between new EGF ranks and new OGS ranks can be made:

OGS  EGF
---  ---
 1k   2k 
 4k  ~6k
~5k   9k
 6k  10k 
 7k ~10k
 9k ~15k

This may seem “wrong” for players who are used to EGF ranks, but a similar comparison table between new OGS ranks and AGA ranks may show a different picture.

I think that the data acquired from the recent rank linking feature was used to fit the new OGS ranks somewhere between EGF ranks and AGA ranks.

5 Likes

Edited: The other ranks are derived as 1 rank = 1 handicap and all handicap games giving about the same winrate for black (for games where black+handicap is within one around white.

The alignment at 1d is visible in your comparison table. Why it diverges at lower ranks I don’t know.

3 Likes

Oh, I’m a little bit surpised by that, because 1 stone handicap is basically just komi advantage for black (worth only half a move).
So when there is a full rank difference, white has an advantage when black gets only 1 stone handicap. Around 1d, I’d expect a 57/43 winning chance for white.

I now remember that this came up before: 2020 Rating and rank tweaks and analysis - #97 by gennan. But it seems that OGS was not convinced by arguments made by me and @KillerDucky in that discussion.

3 Likes

Wild update. I got bumped 4 stones from 9k to 5k… My AGA rank based on tournament play has been around a 9k for a while now, so wonder if the new rating system is better calibrated for stronger players?

4 Likes

Yes this is a big oversight if what @flovo said is correct. Why did OGS decide to come up with a completely new standard for 1 rank difference?

If you want ranks to be separated by no-komi handicap then you should do it the IGS way with 4K and 4K+.

It shouldn’t be possible for every OGS rank to be 50/50 in no komi and also be accurate for 2stone-9stone handicap games.

3 Likes