Proposal: Make the display of decimal ranks more directly understandable

[jlt]

System D is analogous to the FFG rating system, in which 1k corresponds to ratings between 0 and -100, and 1d to ratings between 0 and 100.

[gennan]

I don’t really dare to add system D to the poll. I’m afraid it might lose the results sofar.

So perhaps people can cast votes for D by posting it?

[Samraku]

I’m like 99% certain it would, so good call.

[benjito]

Yeah Discourse doesn’t really support poll editing. And it would suffer from bias anyway, since most people will not come back and change their vote

[jlt]

I won’t vote for system D because it seems strange to say at the same time “I’m 2k” and also “I’m 1.1k”. Which is why I prefer system A.

On the other hand, I’d prefer yet another option: OGS stops displaying the ratings that are used for the Glicko system, which is not intuitive for humans, calls “Rating” the expression Rating = ln(rating)/525 x 23.15 so that 1 dans have a Rating between 30 and 31, etc. and displays the “Rating” instead of the “rating”.

[gennan]

@jlt

System D is analogous to the FFG rating system

It looks like the Belgian Go Federation uses the same system (Belgian Go Federation rating system | Belgian Go Federation).

[gennan]

@jlt

I won’t vote for system D because it seems strange to say at the same time “I’m 2k” and also “I’m 1.1k”. Which is why I prefer system A.

I also consider that the strongest argument for system A.

I’d say that a big advantage of system D is more easy calculation of handicap across the dan/kyu boundary. But that advantage isn’t quite enough to pick system D as my favourite.
I feel system D does have a good consistency, but the same can be said for systems A and B. I feel that only system C is lacking a bit in consistency.

@jlt

calls “Rating” the expression Rating = ln(rating)/525 x 23.15 so that 1 dans have a Rating between 30 and 31, etc. and displays the “Rating” instead of the “rating”.

If you then multiply those ratings numbers by 100, you get again something that looks similar to Elo ratings or EGF ratings.

[pokk97]

@gennan

Yes, shelving the matter until the next rating system update could be an option.

Perhaps this question is worth another poll?

I’m not aware if there’s a rating system update planned but if that’s indeed happening within the next year or so it would surely avoid some drama and save time to bundle rating changes in there

[Samraku]

@jlt

On the other hand, I’d prefer yet another option: OGS stops displaying the ratings that are used for the Glicko system, which is not intuitive for humans, calls “Rating” the expression Rating = ln(rating)/525 x 23.15 so that 1 dans have a Rating between 30 and 31, etc. and displays the “Rating” instead of the “rating”.

No, absolutely not, please noone consider this. The rating is extremely valuable whenever you want more precision. For example, in the command-line program I was writing a while back to calculate reverse komi, it allowed the ratings for one or both players to be given instead of k/d ranks, and it would use ln(rating)/525 x 23.15 internally to convert to more precise reverse komi than was possible with k/d ranks. It also provides some limited comparability with elo systems in other games (yes, they’re not “directly comparable”, but while anyone would acknowledge that 1800 on lichess is different than 1800 on c.c, one would also never suppose that their lichess rating gave no indication of where one might expect them to end up at on c.c

[jlt]

@Samraku

nternally to convert to

Maybe you misunderstood. I’d like OGS to display ln(rating/525) x 23.15 INSTEAD of rating. Both provide the same information since one can be calculated from the other.

[Samraku]

Ah. At that point I’d just multiple it by 2 and define 9d as 0 to make it align with the class system (each class is half a stone). Otherwise it’s just redundant with the k/d detail. And I’d still prefer to have the Glicko-2 number, as that is more intuitive trans-gamely

[gennan]

@benjito

Yeah Discourse doesn’t really support poll editing. And it would suffer from bias anyway, since most people will not come back and change their vote

Perhaps we can repoll in a different format, by eliminating systems that make the least sense to people, and allowing up to 2 options.

Please consider these alternative systems to display decimal ranks:

System A:

System B:

System C:

System D:

Which of those systems make the least sense to you

• System A
• System B
• System C
• System D

[benjito]

Given those diagrams, it becomes clear that system A is problematic. 1.0k = 1.0d

On a side note, I do wish these polls were not anonymous. I can appreciate the desire to keep the pressure off voting, but it does make it hard to tell which options are popular multiple choices vs single choices.

[mafidufa]

Huh? You could say that about every system. Every rank touches every other rank adjacent to it in all systems, as expected. I’m not sure what is special about system A that makes you say that.

[gennan]

@benjito

Given those diagrams, it becomes clear that system A is problematic. 1.0k = 1.0d

Confusion occurs in all systems at the kyu/dan boundary (at rating 1918.4917143963… on OGS)
In all of the systems, ratings slightly above will be displayed as a dan rank and ratings slightly below will be displayed as a kyu rank.

In system A, 1.0d is the lowest decimal dan rank, and 1.0k is the highest decimal kyu rank, so I think it does make sense within that system that they are in fact the same rating. The weakest 1d is equally strong as the strongest 1k.

I don’t find it more odd than for example system C where a 1.0d to 0.0k transition occurs at the boundary.

The ambiguity is smallest for system D, but even there some ambiguity exist between 0.0d and 0.0k at the dan/kyu boundary.

[shinuito]

@gennan

System B:

I think I’m understanding your point, vs how we would interpret what OGS is currently show.

System B seems like it makes more sense to me. It’s not emphasising such a randomly sharp transition between ranks (in this presentation), but in some sense that sharp transition also comes about anyway, where you transition between ranks, maybe the decimals make it much more obvious.

(Edit: basically what was said above)

@mafidufa

You could say that about every system. Every rank touches every other rank adjacent to it in all systems, as expected.

Was it pandanet that does things like 3k and 3k+, which maybe could be more compatible with a system B type display, to indicate where you’re above or below the threshold?

[gennan]

@shinuito

Was it pandanet that does things like 3k and 3k+, which maybe could be more compatible with a system B type display, to indicate where you’re above or below the threshold?

Yes Pandanet uses ranks like 3k and 3k+. But I think that’s more like the class system where each rank is divided into two subranks. Their handicap system also supports half-rank steps.
Here is says that 3d+ means 3.5d. But it’s not clear if this 3.5d is the lower bound, the upper bound or the middle of the 3d+ range.

[jlt]

Pandanet’s display is a bit like KGS, but with half ranks instead. My rank graph is like this.

[gennan]

On KGS, the 1k label would be placed at the lower boundary of the 1k range (system C).

It does seem like on Pandanet the 1k rank label is in the middle of the 1k half-rank range (centered between the promotion line and the demotion line), so I would characterise that graph more as a system B solution, except applied to half-ranks instead of full ranks.

At least, I assume you are currently labeled as a 1k on Pandanet? Or are you 2k+ there?

Perhaps the Pandanet system could be visualised like this:

This is sort of in between system B and system C.

System B:

System C:

[Groin]

We could use plus (+) and minus (-) so 3 classes in all.