2021 Rating and rank adjustments

[reverser]

Thanks to all the OGS staff!!! You guys are awesome!

[FritzS]

All you need is one point of reference. Everything else would follow from game results and the probability of winning/losing (which you can derive from the rank difference).
It would be great to take for example a historic player from the past (like Go Seigen) who is no longer playing active games but where we have plenty of game results and from there to calculate all ratings/ranks. Of all players of the world for all time?!
That would be cool!
We would just have to agree what rank we give to Go Seigen because if we would change his rank, all other players would shift up and down in ranks with him.

[FritzS]

If this would be true (“every game pretty much a 50/50 outcome probability”) then you wouldn’t need the 25k limit in the first place, because the rating system would give such players about the same ratings anyway.

[manGOban]

Nice! Now becoming a SDK player isn’t so far fetched anymore. And that bugfix really smoothes the rank curve! Could we maybe get a slider/buttons to be able see the last 1 week/1 month/6 months/1 Year/ Total or our rank graph?

[Vsotvep]

@manGOban

Could we maybe get a slider/buttons to be able see the last 1 week/1 month/6 months/1 Year/ Total or our rank graph?

We already have that, just drag and select a range in the lower graph:

[AkiraToya325]

For some reason, I got the update but now it suddenly shows everything like it was before. Anyone else have the same problem?

Nvm it’s good now

[meili_yinhua]

@Chinitsu

@mekriff

The issue is that any floor depends ultimately on the ceiling

Seems like a falsy statement to me. Skill-based rating isn’t based on “ceiling”, because we don’t know the “skill cap”. We either have an arbitrary 0 point or set 0 as “knows nothing”.

So basically the ceiling ultimately depends on the floor.

but we do know the rank ceiling: and that is 9p for professionals and either 9d or 7d (depending on philosophy) for amateurs. Some servers add in 10d, but in the current philosophy of go there will not be a higher rank to achieve. Now, this is not the “pure” ceiling of the “perfect player”, but just the highest humanity has gone so far, and generally is expected to get stronger over time as the top players (and the average player) learns more and expands our knowledge of go.

[utah2]

Yes, true, we can choose our own x-axis range for our rating graph, but the resulting y-axis range is way too wide for the data.

And the space allocated to show the rating is too smashed down to see much.

Furthermore, I personally don’t care about seeing the blue confidence-interval shading when I want see my recent improvement in rating (if I have any).

[utah2]

Also, the ratings graph should have discontinuities. Ratings should be constant between rated-game results, assuming the ratings are constant between game results. If my last game was a month ago, my rating doesn’t gradually change in a straight line during the month.

[Vsotvep]

The code is open source, feel free to improve it if you have a better idea.

[xenonoxide]

If it makes you feel any better, you’d have to “defend” your new rank against other dan players, and if you lose you won’t retain the dan rank. If you do retain it, then it means you’re strong enough for it.

Anyways, it’s ultimately all subjective

[square.defender]

@Vsotvep

We already have that, just drag and select a range

someone in help chat:

I notice that I have huge flatline in my ranking because I played only one game between 2017 and 2020. Is this something that can be cleaned, i.e. make my records start at a certain date?

So I got an idea: this “just drag and select a range” should have option to “save” it. So
reduced range automatically stays the same after page reload and others see it.

[utah2]

“Feel free time improve it if you have a better idea.”

It would seem be better to first see if the change is widely agreed.

[gennan]

@mekriff

but we do know the rank ceiling: and that is 9p for professionals and either 9d or 7d (depending on philosophy) for amateurs. Some servers add in 10d, but in the current philosophy of go there will not be a higher rank to achieve. Now, this is not the “pure” ceiling of the “perfect player”, but just the highest humanity has gone so far, and generally is expected to get stronger over time as the top players (and the average player) learns more and expands our knowledge of go.

I estimate that the perfect player is about 13d EGF, meaning that it can give about 3-4 stones handicap to the very top of human pros (at about 10d EGF).

[square.defender]

maybe 5 - if instead of playing perfect it knows perfectly psychology of opponent and uses trick moves that work only on current opponent.

[gennan]

Perhaps. An omniscient go “devil” who takes advantage of knowing the human’s fears and weaknesses may be able to give more handicap than a properly playing go “god”.

[DVbS78rkR7NVe]

In Fast Correspondence “?” ranked player is placed above anyone else, could be some bug Play Go at online-go.com! | OGS

[DeYoung]

How well do the new DDK numbers align with AGA and EGF, if we know? It’s mostly curiosity for me, but it would be nice to have some objective-ish idea of “how good am I in general?” Rather than just how good am I relative to the OGS population.

[BHydden]

higher than the current EGF ranks, but apparently EGF are going to release a new model soon which should quite closely match what OGS just switched to in the kyu ranks, only diverging in the mid to high dans where these ranks will be easier to achieve on OGS than EGF

[DeYoung]

Thanks - that makes sense. How well do EGF and AGA ranks line up? (Curious since I’m in the US)