I think the 13k default rank is doing harm


#41

Note that you can’t measure “drift away from” the nominal rank, because the technical rank will still be what Glicko thinks.

I think that uncertainty is perfect for this. Glicko tells us when the technical rank is believable - just use that.

From a quick look at some beginners’ graphs, it seems to me that by the time it has come down from +/- 6.6 (the starting value) to +/- 3.5 the value itself is well in the ballpark.

So I’d just say "switch from qualifying rank to real rank when uncertainty < 3.5k.

GaJ


#42

I believe that uncertainty is definitely a useful tool, but what I was suggesting is that a 23k? Who plays like once a month should have a maximum amount they can rise (so if you win one, lose one, and win one, add the total change for the two wins) before losing the ? or for a 4d? a maximum they can drop before losing it, with the general idea of if the rating drifts from the nominal rank too much (when only counting the times it drifts away) they lose the ? rank

In case it’s unclear why this matters is because uncertainty rises when you don’t play.


#43

I think if, as I believe GaJ is suggesting, we tie the uncertainty ? directly into the uncertainty number (say, at 3.5k or worse) then it wouldn’t matter how long they’d been away since uncertainty naturally accrues over time and thus if they were away too long they would automatically get their ? back, even if they had previously played enough games to earn a confident ranking.


#44

Does that mean that when a person is away long enough they can estimate their rating again?

But what I was referring to is the case where uncertainty doesn’t drop, and thus they are forced to play at their previously nominated rank, unless you mean to say they can change their guess during the time they have the ?


#45

I wouldn’t imagine that if a person is away for a while they have to re-estimate.

In fact, in some ways it might be better not to conflate uncertainty due to being away with qualifying rank.

If you have been here for a while, then it’s pretty sorted whether you are TPK, DDK or SDK - the number of people who are away for a while and totally change in this respect would be insignificant. So I can’t really see why we need to treat your higher uncertainty after being away in any special way.

That being said if we do want to, then you wouldn’t re-estimate, you’d just take their “previous” rank as a new qualifying rank.

GaJ


#46

The thing about uncertainty rising with time away is part of the Glicko and Glicko-2 systems… It’s a way to account for any rise or drop in ratings when not active on a specific site.


#47

Right - I’m not saying “don’t make uncertainty rise”.

For sure that should still happen.

I’m just saying I’m not yet seeing the relevance of it.

The problem with the 13k starting rank is not just “the person has high uncertainty”. The problem is “they are out of their league”.

If we explore going away for a while: suppose you played steadily and become 13k +/- 2 (lets say). Then you went away for a while so your uncertainty rose and you you became 13k ± 6 (lets say).

Now … have you really forgotten everything? It seems very unlikely. A 13k who hasn’t played for 5 years will still know corner centre sides and basic shape. So they are at least 19k. So they can use the rank pairing settings in the get-a-game system to get a match with someone at that level. It’s not like they are 25k and compltely out of their league witih a wrong rating.

Now that I’ve typed this, I can see my suggestion of using the previous rank as a qualifying rank is incorrect.

In fact, if you hav been away so long that your established rank is totally wrong, then you should just start a new account, and select a new qualifying rank. :slight_smile: Like this person did.

(And isn’t it fascinating that in that thread we had to point out to them that they would not immediately get the effect they wanted from doing this: they wanted to stop being paired with people 15-16k, because they were no longer that good. But ironically in the current system making a new account makes this problem worse!)

GaJ


#48

Sure! I can get behind this now.


#49

I bet you could start someone’s ELO at 1500 to satisfy Glicko, but have a second Glicko-style rating with lower uncertainty starting at their estimated rank. The first ranking would be used to compute how their opponent’s ratings changed when playing them, while the second would be used for calculating their kyu-rating (with a “?”) and used for automatch purposes. The second rating would be adjusted as per the usual Glicko rules, but wouldn’t affect anyone else’s ratings. The “?” would be removed once these two ratings converged to the same rank, which would ideally happen pretty quickly, given the difference in uncertainty between them. For convergence, you’d have to treat a win and a loss differently, based on whether the secondary rating was above or below the primary. In general, you’d have to have a higher uncertainty in the direction of the primary and a lower uncertainty away from that direction.

Basically, you’d start as a 25k?, for example, but with an effective Glicko rating of 1500. If you lost against another 25k, with a Glicko rating around 0 (I think?), your primary Glicko rating would drop to 500 or so, but your secondary rating would drop from a starting point of 0 to -100, or so, and you’d become a 26k?. If you won your next game, against another 25k, your primary would go up to 600 or so, while your secondary would move up by a higher amount, maybe increasing by 200 to a score of 100, with a corresponding rank of 24k?. So long as the two ratings moved substantially closer with every game, you’d converge within half a dozen games, and your displayed rank would have a high likelihood of representing your actual skill.

This way, the global rating pool would still work the same way, but new players could play people of similar skill right off the bat, instead of getting smashed for a few games, or alternately smashing the opposition if they were a strong player making a new account.


#50

Why not just create “unraked” accounts. You play 6-10 games with no rank hit or gain to your opponent while your gets adjsuted. After those games you have fairly accurate ranks.


#51

This is a very similar proposal to “give people a qualifyig rank till their rank settles”.

At a basic level, it would appear more desirable not to make newcomers go through two accounts, if possible.

If you started with an unranked account, how do you get games with your unranked account? Getting games requires a rank. Or do you mean that an unranked account can play against whatever rank they like?


#52

I think that is basically what I proposed. With the slight twist that the qualifying rank is glickoed rather than fixed. That adds a fair bit of complexity. Is there a significant benefit?


#53

It’s definitely based on your proposal, which seemed good, but I think could be even better. Your post had gotten kind of buried, so I decided to just post something new. The main difference is to have a floating qualifying rank, rather than a fixed one, and use that for match making.

The benefit of having a glickoed qualifying rank would be that it would be a “best indicator” of someone’s rank, based on all available data, and therefore would ideally give the best matches. Someone that gave an inaccurate assessment of their rank, initially, would play more people close to their actual rank, instead of at whatever their fixed provisional rank was. This way, someone that’s actually a 5k, but accidentally hits beginner, will start to get 10k or 5k matches pretty quickly, based on their adjusted provisional rank, rather than winning a ton of 25k matches until the system decides that their rank isn’t uncertain anymore. At the same time, and actual 25k can start playing at their rank right away.


#54

I think the simplest idea would be to just let new users estimate their strength and choose if they want to start from, say, 25k, 13k or 1k, attaching an uncertainty of 350 to their ELO regardless.


#55

Last time we discussed something, I badly misjudged whether you were across it or not Sarah_Lisa. So I’m at big risk of doing that again … but … what you wrote is what I had originally asked for. It was shot down (in this thread, and others) because “Glicko must have new people starting in the middle of the range”.

Apologies if you already knew that, it’s just that your suggestion doesn’t seem to take that into account.


#56

If that really is the case (although I kind of doubt it, would be nice to get official word on this) then I think the current system is probably preferable to some complicated alternative with secondary ranks. How many games does a beginner with ELO 1500 and 350 uncertainty have to lose to get beaten down to 25k, anyway? Personally, I got there within a day.


#57

Supposedly it really is the case. I’m not across the maths.

If you think the current system is preferrable, what you’re saying is that all the problems reported in this thread and others aren’t real or aren’t worth fixing. I’m curious why you think that.

A “secondary rank” does sound yucky, but I’m not sure what it means.

The current proposal is a “qualifying rank” which is well defined and seems not too awkward (because from the user’s perspective they only have one rank at any given time, and from their perspective it is basically the right value).

GaJ


#58

Never said anything about two accounts. Users would just have no rank for first few games. They would only be barred from rank restricted challanges, I am ok with people starting at 13k too. What I think is wrong is that opponents take rank hit/gain from playing such players. First few games should be like KGS - only one side adjusts rank


#59

It seems worth revisitng the list of problems that the current system causes, which make it worthwhile seriously considering a fix:

  1. Certainly (not just theoretically) beginners who join will get games that are not just un-even, but are out of their league
  2. Along with that, demonstrably, their process of getting a first game will be marred by people cancelling out on them
  3. and SDKs/experienced DDKs will be faced with either playing out a game against a noob that they weren’t expecting, or rudely cancelling out.
  4. and experienced players will also have the experience of games being cancelled out on them by people who think they might be noobs (once again, demonstrably: it happened to me).
  5. Beginners can’t even set up games with other beginners if they want to, because their rank is so far away from 25k that they aren’t even allowed to by the system!

The “make the account unranked” proposal solves points 1 and 5 only, as far as I can see.

It makes point 3 and 4 considerably worse, because in that system a newcomer could challenge whomever they like so we’d see people “kicking the tyres” of Dans.

I believe that the “qualifying rank” proposal addresses all of the above points.

GaJ


#60

This is the case… I can’t be assed tracking it down right this minute as it’s late at night here but I’ve seen anoek say that players who have a ? for their rank do not affect the rank of their opponent.