I noticed my rank drop unexpectedly far after losing a 6-stone 13x13 game as white. Looking at rank_utils, it appears that the ratings calculator does not take board size into account when adjusting ratings after a handicap game.
Using the rating calculator, I confirmed that my rank adjustment in this case (0.3k) would be expected for a 6-stone handicap on a 19x19 board.
However, clearly, stones in small board are worth a lot more than in a 19x19.
Is this a known issue?
What should the adjustment be? I think a reasonable starting point would be:
- For 13x13, the handicap should be multiplied by two (1 stone is treated as-if 2 stones).
- For 9x9, the handicap should be multiplied by four (1 stone is treated as-if 4 stones).
Those multipliers are based on a 19x19 being ~2x bigger than a 13x13 and ~4x bigger than a 9x9.
Interestingly, I just realized I’ve been (unintentionally, but, effectively) sandbagging, since I play ranked handicap games on small boards with friends. We adjust the handicap every time one of us wins twice in a row. Then I take my artificially depressed rank and play against others. Now I understand why I seem to be winning a little more than my fair share of those games.
(Until the ratings calculation is updated, I’ll just play these small board handicap games unranked – it’s fine – but generally I’d prefer to participate in the ratings system…)
I’d like for this to happen as well. I’ve lost some motivation to play 19x19 because my 9x9 games have pushed my rank up to the point where I get slaughtered in the former.
But this has been discussed before on the forums. Apparently the overall rank appears to better reflect someone’s skill than the board rank, so that is what’s used to match people.
To be clear, I’m not discussing the multiple versions of ratings—which I think is what you’re referring to—just the computation of how much the rating should shift after a game result.
E.g., it’s treated as “surprising” that white would lose a 6-stone game against someone 9k weaker on a 13x13, resulting in a big rating shift for both. But, it should be treated as “unsurprising”, with a small rating shift, because 6-stones on a 13x13 should be even-ish for a 12k weaker opponent.
I agree, 6 stones on 13x13 should not factor into rank the same way 6 stones on 19x19 does.
AFAIK, the autohandicap does factor in size to the handicap assignment, so I’d imagine the same mapping can be applied. Although that system has room for improvement as well, some related discussion in this thread about komi on small boards.
I thought I remembered that… in which case, fully agree. It should be the same mapping for both. Even if there’s room for improvement, they should match each other and change together.
Here what it’s probably supposed to be:
So it should have been 4 or 5 stones, maybe that’s just not so different from 6?
(I don’t know how the server uses this in reverse to convert 13x13 stones and komi into a 19x19 equivalent for rank adjustments.)