I think the 13k default rank is doing harm [Closed]

But then I’m excluding 13ks, that’s not a bad game for me or for them!

Look me in the eye and tell me that once you’d checked that “no provisional players ty” box, you’d actually remember and uncheck it on some idle tuesday.

Hmm - I thought you guys were discussing having this option even after humble rank.

Since humble rank seems squarely on the table, there seems little point in debating some other proposal to avoid 13k beginner matches.

That would be solved by having to check that box in the create game dialog every time…

Hurray for defaults!

I only have the ranked games, therefor there are no numbers for canceled games.

I consider a player as veteran if the deviation is smaller than 150

First ranked game against veteran September

won lost
scored 448 1477
resigned 1287 3075
timeout 258 701
canceled / annulled ? ?

last year:

won lost
scored 4885 12900
resigned 12256 24522
timeout 2038 5963
canceled / annulled ? ?
1 Like

Last year 50% of new players were weaker then 16k.
In September 50% were weaker then 18k.

For more info take a look at the linked topic:

1 Like

It’s interesting to see the stable ~25-30% winrate regardless of type of outcome.

For fun, I picked a fairly stable 13k and went through 400 games, cancel rate seems to be around 5%. Necessarily,the cancel-due-to-newbie rate is going to be even lower. But even if we assume 5% for that, with 2000 new users per week and <10000 games between them, that’s 500 games at most.

Of course the relevant percentage would also be affected by whether the cancelee is actually a newbie. Regardless, unless my sample was biased or my coffee isn’t working, it’s unlikely that a new account sees more than 1 cancellation per 4 games.

Now, provided that the cancellation is handled courteously, that shouldn’t be a big problem.

Alrighty an initial cut of the humble rating/rank is up on the beta site for anyone curious and wanting to test things out!

5 Likes

Hm. Not much going on there. And somehow I doubt correspondence games are the best choice for a betatest server. ;D

The feature is now live, we’ll see how things go!

11 Likes

:partying_face:

Thank you for all your hard work on OGS.

4 Likes

I had just 2 new player (1150±350) accepting my open challenge with rank restriction min rank 18k, max 8k.
1150 ≈ 20.6k if I’m right.
Is it intended, that provisional player can accept my games even if their rank is not within the rank restriction?

Screenshot_20181017-093046

What’s weirder is that this guy started at 1150, lost his only game, his 19x/overall properly dropped to 789 (31k) but his overall/overall rating increased to 1192 (18.4k), variance dropped to 253.

@anoek

Somewhat unrelated: can we have an automatic message to the new account’s opponent with something like “Your opponent cannot chat, because the account has not been verified.”? I think it would ease some tension with newbies.

The unverified player could get a similar automatic message in chat saying “You can only use chat after verifying your account by linking it to an email (Top left menu > Settings), logging out and logging back in.” Or does that already exist? :stuck_out_tongue:

5 Likes

Not weird at all. My overall/overall rating is ≈1500 at the moment, while my 19x19 ratings fluctuate between 1000 and 1200.

With humble rating, his rating is 1500±350 but is shown as (and used for matchmaking?) is 1150±350.

After the game, his overall rating droped to 1315±253 (= 1171±253 humble rating).

Since my 19x19 ratings are much lower than my overall rating, his rating loss there is much higher for 19x19. (loss against much weaker opponent)
My 19x19/overall rating before was 1019. Therefor the rating drop from 1500±350 → 1056±319 (781±319 HR).
Same for 19x19/corr (my rating 949): 1500±350 → 1020±327 (726±327).

(Btw: Why is my calculation of the new humble rating off by 20 points for overall and 10, 5 points for 19x19s? I literally copied the conversion formula)

1 Like

I still don’t get why his overall/overall rating should increase from the loss. It’s his only game.

Edit: Ah, I see. So the decrease in variance alone was enough to counter the rating point loss and turn it into a net gain. That’s just weird.

1 Like

Oooer. Did we end up with a concept of humble rating?

Surely it was supposed to be humble rank, derived from your glicko rating (the only one you have)?

Concept:

“Your humble rank, for matchmaking, is the lower bound of the rank derived from your glicko rating”

It’s going to be confusing as heck if we now have two “glicko ratings”.

What’s more, if someone is 1500+/-350, they can’t be 1150+/350 as well. They are 1150 +700/-0.

I had imagined we would have one rank - the humble rank - which has no stated uncertainty, it just is the rank you are assigned, and it is derived from your glicko rating, by taking into account your uncertainty.

3 Likes

Yea that makes more sense to me as well.

I from the beginning thought that we’re humbling rating, after all rank is just a social construct cosmetics to keep kyu/dan tradition.

Hmm - there’s an extent to which this is just a technicality of which number is used in the matchmaking system. Since the limits we can set in matchmaking are rank-based, I’d thought that rank is what’s used thoughout, so the humble rank would be what’s needed to feed in.

FWIW, the original ticket raised by lemurk was titled " Introduce humble rank for display/matchmaking", and said

"One of the proposed solutions suggests introduction of ‘humble rank’ concept, which corresponds to lower confidence boundary of glicko-2 rating (i.e. rating - deviation) for display / matchmaking purposes. "

But, actually it doesn’t matter whether we use the humble rank or its reference point the humble rating for matchmaking, what matters is that we don’t confuse the issue by the way that we display it.

That would mean, at the very least, that if a person has a glicko rating of 1500 ±350, the humble rating or humble rank number should not show up with ±350. There is only one exact number used for matchmaking, that’s how it needs to appear. IMO.

I would also expect that a person’s actual glicko rating with its uncertainty is what shows on the profile in the graph. The matchmaking value (humble value) is a line that would start at the bottom of the uncertainty envelope and work its way up to the glicko rating depending on how that formula works.

2 Likes

I’ve been using Many Faces of Go for years and I still like it. You can get a trial version on a 9x9 board free. Playing with people is much better, but I still play a lot with Many Faces of Go. For one thing, I can start a game with Many Faces of Go any time and not worry about hurting my opponent’s feelings if something or someone interrupts me and I want to stop playing in the middle of a game.

I think it’s very helpful for learning go too, but don’t just play a computer player. They don’t play exactly like humans, so you can develop bad habits that work against the computer program but not against people. You also get more variety playing people, and, you know, you are paying fellow human beings.

1 Like