Right now, there is only one ranking that is used for matchmaking on all board sizes and time controls. I assume this is done for simplicity. However, I like to play both 13x13 and 19x19 games and my (relative) strength on the 13x13 seems to be higher. This has the result that when I play both sizes regularly my overall rank for 13x13 is too low and my 19x19 rank is too high. As a result I would lose almost all games on the large board and have too easy games on the small board. My solution to this was to stop playing 13x13 altogether, so that I can have at least sensible 19x19 games. This is not great.
Since the rankings for the individual sizes already exist, I don’t see the downsides of using those for matchmaking instead. The overall rank can then be the cosmetic one that is displayed next to your name.
many people have very outdated rank in some of these separated ranks.
better solution would be to allow to choose not overall rank. But only if it is not outdated.
Just to make it clear, all your ranks are about 8.7k.
Yes, now they are, because I stopped playing on the small board. In until February I had 40 wins vs 25 losses on the small board and 55 wins vs 77 losses on the large board. In my memory it felt like a more drastic difference, but that was the point I decided to focus on the large board.
Yes, I have noticed this phenomenon with players who have an inflated rank which is primarily based on playing blitz on 9x9 and in no way reflect their rank when they play on 19x19 and there is time to think about the moves.
I am one of those but, to be fair, is it really a problem?
It’s outdated precisely because I don’t play 13x13, so it does not matter. If I started to play 13x13, my rank would adjust and become correct again. Sure it means there will be a few incorrectly-matched games until it adjusts, but it seems a small price to pay.
I thought of playing more 9x9 and 13x13 but under the current system I would tend to create a new account dedicated to it, precisely because I do not want this to affect my regular rank. This means I would be playing a few incorrectly-matched games anyway, but with the inconvenience of switching accounts…
As another example, I have a friend who is decent at 9x9 but sucks at 19x19. He should rank down on 19x19, but he doesn’t because winning in 9x9 keeps him up. That’s annoying.
The existence of these statistic datas doesn’t imply that it would be an easy task to rebuild the whole system.
Actually, no. It used to be that different ranks were used for different boards and time controls, but testing was done and it was found that the ranking system was more accurate with them all combined.
The perceived differences weren’t a bother for glicko.
Hmm, I guess the ranks are highly correlated and more data increases the accuracy. It’s hard to define the accuracy of a ranking system, but I assume accurate means that most players win and lose about 50% of their games. Still, that could be an artifact; If I play half the games on a small board and win all of them and the other half on a large board and lose all of them, then the ranking would still be accurate in some sense but it would not make for good auto matches. So it would be accurate but not precise. A ranking system is precise, if the winning chances for each game are given only by the ranking system and independent of other data (such as e.g. board size, who plays which color or time control).
Furthermore regarding the lack of data or outdated ranks, the Glicko system has an uncertainty on the ranking. The overall rank could act as a fall back if the uncertainty is larger than some threshold maybe.
Accurate in this instance means how frequently the rating system correctly predicts the outcome of the game.
A totally ignorant rating system predicts correctly about 50% of the games. A good rating system predicts correctly about 55% of the games according to Remi Coulom’s WHR paper (but I suppose this is only the case when most matches are not hugely skewed).
It may be true for most users. But let’s not forget about exceptions. Currently if skills of someone are unusually different on 9x9 and 19x19, then the best thing they can do is to create additional account. But why force them to do so complicated thing? Similar functionality may be integrated in single account for those who decided to choose so.
Simple solution: create different accounts for different board sizes.
It is a simple solution, but one that mostly highlights the underlying issue.
Considering how much work would be involved to modify the site, this simple solution seems adequate.
Yeah I feel like it’s possible to have the best of both worlds here:
- 9x9 rating confidence high? Use 9x9 rating
- Else? Use overall rating
Or use weights for a more gradual transition.
Perceptually though, I don’t think any of these changes will make much of a difference. As we noticed, OPs 9x9 rank is very close to overall. I just checked my own and it’s also similar, despite feeling stronger on 9x9.
It would be interesting if someone can find an account with a significant ratings difference, and see how bad their win-loss ratio actually is.
Have a look at my early days
