OGS has a new Glicko-2 based rating system!


#203

Thats why having similar results in both glicko and elo is more reliable than just glicko/elo alone.

Actually I believe glicko have lot more potential compared to elo.

By “validation” (not perfect term, but cant find better one) I mean elo telling “same story” like glicko for vast majority of normal (non-sandbagger, …) players - if this happens vast majority of my concerns will disapper.


#204

Is the OGS team planning to do something to actually reduce the ranks of high dans, and not just hide the problem by showing “9d+”? Having an unrealistic mapping from rating to rank affects handicap games. For example, I think that a six-stone game between FrancisCol and me would be fair, but if the system thinks that he should give me 10 stones, then he won’t be interested in playing a ranked handicap game against me.

This could be fixed by changing the function 850exp(rank0.032). Maybe the mistake was that this function was derived by giving the same weight to each player, which gave a great fit for kyu players at the expense of high dan players?


#205

I’m like a squeaky wheel here, but I swear: the bulk of this confusion is because the breakdown table is in glicko and not K/d.

If you had a profile page where everything was expressed in K/D, and the glicko number underneath is just an implementation detail, this whole “this is too hard to understand” would go away

The price would be that people would compare numbers from different scales in a way that mathematically is not valid.

To this I say “so what?”

I’ve been in the minority in thinking this is the way to go, but that doesn’t mean I’m not right :wink:

GaJ


#206

Yes, from a usability perspective a mistake has been made. Ratings should be rendered in kyu/dan terms because that is what Go players understand.

This illustrates the problem:

OGS is made by engineers. Such a justification makes eminent sense to an engineer. To a non-engineer, however, this is screaming out that even justifications of the rating table describe how to translate ratings into ranks. No-one cares about the ratings!

So, there’s a simple solution:

  1. Present the table in kyu/dan terms, not glicko rating terms.
  2. Provide a toggle, so that the glicko data can optionally be revealed.

I’d recommend going further. So, when displaying the kyu/dan ranks, drop the ± stuff. Again, no-one cares. There’s an intuitive way of showing the deviation without asking users to do math:

  • 8K” is an established rank.
  • ~ 8K” is an approximate rank where the deviation is quite large.
  • ?K” is a provisional rank where the deviation is too large to trust.

Life would be better. Everyone would be happy and the glicko data would still be available for anyone interested examine.


#207

In far future there will be far better and more complex system than Glicko-2 and next to each user rank there will be source code of it : )


#208

In the far future people will have ratings for everything(getting a bit dystopian here) including how well we live our lives, how good of a citizen we are, how trustworthy, and these will keep us all under control of the government, shunning away anyone who thinks of speaking out in order to improve are rankings. Of course this is already happening with the new plan china is planing to implement, but obviously there is not a 100% chance that this grim future is ours.


#209

I certainly agree with @Farraway, especially with the idea of the toggle for those who do like numbers (me)


#210

@Farraway +1

I don’t understand the complex new system :frowning:


#211

Yes, having both ranks would be nice, just for reference. I think that OGS has incredible UI, but the new ranking system and how they are displayed is really confusing. I know that it isn’t really that hard to understand, but I personally like the older graph better. It just made more sense. I think OGS should keep the older graph, but adjust it to show the Glicko-2 ranks. Also, it would be handy to have kyu/dan ranks on the side of the graph too, not only because it is easier to read, but also because this is how we ranked people in go for thousands of years(correct me if I am wrong about that). So it is sort of a tradition that all go players know.


#212

The irony is that it isn’t that complicated.

You have one number: your rank.

You have one line on the graph: how the rank changes with time.

The rest is all “presentation”.


#213

I think we need a more compact graph for people with small screens. Maybe take the old normal/full size thing and show the overall rating and small graph, and then the full size and selectors when you click a link.


#214

Yeah - I think it is more "rewarding when the vertical scale is bigger too :slight_smile: Right now any changes are very subdued because its a long and flat graph.


#215

I think when discussing a rating system the only significant question is “how readily can I assess the likelihood that I will beat another player in the same system?”

The rating system does not and should not care how your rank or its changes make you “feel”


#216

Nah, rubbish :slight_smile:

If your statement were true, the graph would not be needed at all

The ranking system helps me track my growth as a player.

So the other significant question is “how have I improved over time”?


#217

Yes but the scale doesn’t change the message. You could zoom the old graph in or out as much as you wanted to make it look flat or sharp… it’s the rate of change that determines progress regardless of how zoomed in or out the axis is.


#218

And you actually have both, I think the bottom graph does an excellent job of showing just your growth over time whereas the top graph has a stronger emphasis on your position relative to the group.


#219

I disagree. The scale definitely changes the visual impression of the message.

A long horizontal scale de-emphasises vertical changes.

But on a per-game basis, vertical changes are what we care about.

Therefore this:

is a lot more helpful than this:

… but my current glicko graph looks like the latter:

The proposal above would make it look more like the former.

GaJ


#220

If you want a more localised graph you can select a time frame from the bottom graph and the top graph will reflect that localised position. As you can see in my screen grab your progression over the last month and a half is much more obvious (as you wanted)


#221

I don’t want a more localised graph. That’s worse.

I want to see the whole timescale with a compressed horizontal access and an expanded vertical access, so the flow of my progress from start to end is more obvious.


#222

In 6 months you’ve gone up 1.5 ranks… that improvement is commendable, but I find the new graph much more humbling and a “reality check” as when you consider that the journey from 25k - 7d involves 32 rank jumps, a change in either direction of 1.5 ranks should not be given as much attention as it was with the old graph.

I understand the nice endorphin hit you get from seeing your graph climb (as you can see by comparing my old graph to my new graph) but consider for a moment the players that drop a rank or two… in the old graph that looks like a spiralling collapse where as the new graph shows clearly that, while they have dropped some, it’s hardly a “devastation”