OGS has a new Glicko-2 based rating system!


#122

It’s a rough guide to see where a player is stronger or weaker. Nothing more.


#123

I don’t actually understand the quote from anoek - I don’t understand what it is getting at.

The suggestion is:

Clearly it can be done, because it already is :slight_smile: :slight_smile:

There seems to be good concensus that although they are “strictly informational”, it is great information that we like to have.

Literally the only suggestion is one of presentation: present them as k/d, which is what we all think in anyhow.

Otherwise this table is the one place where we suddenly have to convert in our head to get information out of it.

Which we are happy to do, because we love that information. Just … why not make it easy for ourselves?

GaJ


#124

@GreenAsJade They are on different scales means that e.g. 10 kyu rank on overall is not the same as 10 kyu rank on any of the breakouts. So, if OGS showed you that information, you’d be fooling yourself, because they’re not equivalent.


#125

But this is about how you use the information. And that same argument applies if you go and compare 1770 to 1800 in a different section you are fooling yourself.

It is exactly the same as comparing 10k to 11k in a different section.

The only suggestion here is about presentation.

It is easier to “read” and “appreciate” 10k and 11k than 1770 and 1801.

Furthermore, although they are different scales, they still have a sensible proportionalitly, so 10k is clearly someone quite good and 20k is still a beginner. I have no actual idea what 1700 means at all. So people are going to do this conversion anyhow with these numbers.

Therefore it is only about whether the server does it or we do.

GaJ


#126

Rather than thinking rank maybe just think stronger or weaker.


#127

I’m not clear why people keep suggesting ways to deal with the way that it is right now, when the improvement suggested is clearly better.

Sure, there are all sorts of mental tricks one can do to correctly interpret the glicko results in the breakdown table.

The question is why inflict that on ourselves, when a more readily absorbable presentation is available?


#128

Because the people who know what they’re talking about say that what you’re suggesting will mislead people.


#129

Ah well, let’s debate that then.

I don’t think its been shown how it will mislead people any more to compare glicko rating between breakdown groups versus comparing kyu ratings between them.

As we saw: the first thing that everyone asks when they see the glicko rating is “what does this mean in kyu”.

We’re kidding ourselves if we think that generally the use of that table will be in glicko terms - people will be converting in their heads.

Furthermore, the use that they will make of the table is the same way whether they are k’s or glickos. If it is invalid to compare X and Y, it doesnt matter the units that X and Y are expressed in.

GaJ


#130

Solid argument. If it is indeed true that the numbers can’t be compared, maybe what would be more useful is possibly some charts showing the ratio of games played for each time control and board size?


#131

Anyways, I think what @GreenAsJade is saying is also valid. People are used to the kyu/dan ranks for simplicity.

Aside from the ranks, I also like the ratings to be readily available. I’m sure the developers will find a middle way. :wink:

When I view my opponent’s profile, I usually focus on the overall rank and rating, no matter what speed or board size I play.


#132

Since the traditional k/d ranks might not make sense due to the different ratings pools, maybe a different approach is needed.

Perhaps OGS could come up with a similar metric that takes advantage of Glicko’s strengths. Glicko is a relative rating scale. With the Glicko math, you can calculate chance of winning an even game based on rating differences, right?

Could we have a tier system where people in Tier 2 have a 75% chance to beat people in Tier 1? Then people in tier 3 have a 75% chance to beat people in tier 2 and so on.

For each of time/size sections in the table, you could sort rank all the players on OGS based on their 95% confidence interval rating and start the bottom people at Tier 1, then bucket everyone by score above the base.

There could be a separate site wide page that shows the table with an average/median Tier level for each dan/kyu rank.

Some features:

  • It’s different enough from the traditional k/d ranks, but the purpose is very similar. So it would distinguish itself from the overall k/d rank without appearing as a confusing duplicate. The overall k/d will appear more prominently and clear all doubts that this is your OGS rank.
  • Players can compete in whatever categories they prefer and still have a metric that spurs competition and still shows progress.
  • It would serve to highlight the incredible depth of go.
  • There’s no top limit for competitors, this could attract strong players seeking a high score. (maybe not, but we can be optimistic)

The 75% is arbitrary, could pick any percentage that makes sense.

This is longer than I thought it would be. Mods feel free to split it into a different topic if I went too far.


#133

It’s because kyu/dan rankings have extra meaning imbued in them. Glicko is all about relative rating between the players in the ratings pool. K/D rankings have a long history and have powerful meanings due to that history.

Kyu/dan rankings have extra meaning because people will compare them to the other kyu/dan rankings in the chart, k/d rankings of other servers, k/d ratings of other organizations. People will complain and make a lot of work for moderators/devs to respond.


#134

makes sense, i read the post and took that a face value, my slippage is more than just +/-1, and i don’t know if there are others who have slipped as far, or if they even noticed.


#135

I like your idea


#136

That is a good argument.

:slight_smile:

I’m not entirely convinced that the benefits of just being able to use the chart by easily reading it don’t outweigh the possible downsides mentioned.

But if the moderators/devs said “this is the reason why those are glicko numbers” I can totally respect that.

GaJ


#137

i get that, i purposely chose to go against the higher ranks, as going against those who are even with me, some were stronger, most weaker, and i thought the ranking operated more like chess, the higher rank you beat, the more points you get.


#138

That is true, however, since you peaked at 17k you have since lost 84 games and only won 20. So even though the wins mean more and the losses mean less… the losses still add up :wink:


#139

i get that too, i’m not saying that i didn’t have a part in this. i just see even LESS incentive to play thos at my own level


#140

That’s fine… you’re free to play whoever you want. It just comes down to whether you care more about growing as a player or growing your rank. Playing only people of a higher rank than you may help you grow, but it’s likely your rank will drop a bit before going up since it is purely based on the game result and doesn’t analyse how you played.

If all you want is for your rank to go up, then play weak opponents. If you always win, your rank will always go up. If you always lose, your rank will always go down… but if you have a balance of wins and losses the system will have a better chance of aligning with your true strength by comparing who you beat and who you don’t.


#141

i’m greedy, i want both, i want my rank to go up because i’m able to beat stronger players