You can assume 60% games played outside, but if the real percentage is 80% then the total number of games will be underestimated by a factor 2.
Another problem is that some players almost never use OGS. If they play 40 games on OGS and 4000 games elsewhere, with your 60% assumption you would infer that they needed only 100 games to reach 1d.
To avoid that you might want to discard outliers, like 5% of the accounts with the fewest games and/or accounts that have have long periods of inactivity. But then you will miss players that really improve fast.
2 Likes
Why is so important the number of games?
Is that really a proxy for rank improvement?
I’d choose time instead.
As an example, I checked my rank progression on EGD database against those of other Italian players. It was astonishing (for me) to see that I improved few kyus in the same time (years or even months) that our stronger players reached shodan.
Very few games are tracked on EGD since there are only tournament games, but the bit of information was very interesting.
We could do a similar analysis on OGS: quickest vs slowest progressions.
7 Likes
I’d love a 2023 version
9 Likes
How did you get the data to make this?
1 Like
@Allerleirauh , What about histogram like this? :
“number of games” 2000 may mean all users with at least 2000 ranked games
and we calculate average rank of these users
2 Likes
2023
https://www.reddit.com/r/baduk/comments/14sgtqw/active_players_on_ogs_today_i_wish_there_was_a/
(by trung12ly)
10 Likes
What’s the definition of “active”? 8k feels low 
I guess only those who was here on 7th July are counted
3 Likes
In that case that’s pretty good 
1 Like
Oh wow the rank graph is backwards that threw me for a minute 
2 Likes
please post the exact percentiles
why not do the last 6 months of those with a rating deviation of less than 100 like the original graph did.
OGS isn’t wildly out of line with other online servers, though it was slightly ungenerous before things got reworked in 2021:
Note that in order to make the ranks graph smoothly, I’ve opted to set 1D as 0, 2D as -1, 3D as -2, etc.
It’s less that the ratio is “bad” depending on what you mean by “bad” and more that the threshold for ranking to Dan is rigorous, and broadly that is something the server has in common with its peers. People familiar with the current comparisons will have to weigh in on whether or not OGS could still stand some rank inflation to be in the middle of the pack—the devs said they brought Dans more in line with the average—but it looks like most servers end up with a similar distribution to the graph above, if not a slightly more lax cutoff
2 Likes
Can we have ranks down to 30k?
Thanks to the big spike at 25k.
2 Likes
anoek looked at this and found that handicap stones weren’t significantly effecting the result of the game below 25k. The system will still pair you with the closest available rating to your own.
1 Like
Very interesting. I noticed in the first post where it shows accurate percentiles like 60k and 15d.
I thought it was based on ogs rank, but I guess OP was making up their own rank criteria?
1 Like
The conversion formula between rating and rank works to those ranks. OGS artificially limits ranks to 9d and 25k for posterity and functionality. It’s not a technical limitation but rather a design choice.
3 Likes
What equation are you using to find the difference in ranks?
It’s an exponential function, right?
2 Likes
My in-real-life experience is that a ranking system can’t possibly be accurate enough for handicap stones below 25k, but handicap stones still make sense for players who play together more than once.
However, they should be chosen based on previous results of the two opponents together, by simply adding or removing one stone at every win or loss.
For the weaker player, this is also an encouraging way to have achievements and measure their progress as they start playing at 9 stones and then remove more and more stones.
3 Likes