Better stats with one color

The difference is probably not big. If the biggest move on a board is 7 points, then a 10k should be able to find at least a 5 point move on average, so I guess that if komi for a perfect player is 7 points, then komi for a 10k should be at least 5 points.

Anecdotal evidence: I said I played 508 even games on IGS. My average rank was 7k, and I can take indifferently black or white. Komi on IGS is 5.5.

1 Like

@Uberdude @david265 You want data? I give you data! My analysis of nearly 2000 games from a DGS ranked 9x9 tournament of mostly kyu players (a few Dan games) with Japanese rules and 6.5 komi. Population win rate is 50% black, 50% white within stastical error. This is in agreement with KataGo analysis for 9x9 assuming strong play.

https://www.dragongoserver.net/tournaments/view_tournament.php?tid=105

So if @david265 wins more as white on 9x9 with Japanese 6.5 Komi, that says more about his personal style of play than it does about Komi. If that was with the OGS 9x9 5.5 Komi, then it says even more about style than Komi as reduced Komi would favour black.

6 Likes

Hehe interesting data. (Follow the link on L19).

So time to consider other reasons as the komi?

Handicap games being included in the stats. That’s expected to make white win more.

OGS assigning white to stronger player by default in even games, is that still a rule, or in the historical data?

AFAIK this is still standard practice, based on the historical idea that the strong player takes white, even if that now seems unfair it has not been changed to my knowledge.

1 Like

Understanding katagos opinion on 9x9 is definitely confusing me (the difference of winrates with area vs territory rules, with integer or non integer Komi)

There’s another thread which is not unrelated to this current discussion in that it’s discussing small board komi (incl 9x9) for New Zealand rules

Thank you for this analysis! It does seem to disprove my assumption. Thanks to everyone for their patience with this thread. I guess it’s just me for some reason.

3 Likes

~ 2 years ago, “Automatic” (the default color assignment method) was coin flip if both players within 1 rank distance. If the rank discrepancy was higher, the stronger always got white.

I don’t expect anything changed since.

6 Likes

That would explain a lot

Thanks; so that, coupled with the other OGS behaviour (which I don’t like) of games being even rather than handicap by default, will result in even games with white being the >1 rank stronger player, so of course white will win more.

2 Likes

Interesting :thinking: I hadn’t really thought about this so I don’t really have a strong opinion either way. What would you say are the pros and cons of each one as the default?

I suspect quite straightforwardly that even games between players of different strengths is just more likely to be boring or annoying for both players

1 Like

Doesn’t the auto match produce handicap games?

Cons:

  1. Handicap games are a great feature of Go to enable two players of widely different rank to enjoy a game that is a challenge for both
  2. It sets an expectation for beginners who learn the culture of Go on OGS that handicap games are abnormal, and that widely different rank players normally play even games. This is not true in the wider world. I think it’s bad to have trained a decade of Western beginners to not like handicap games. If a 20k from OGS turns up at a real-life go club and tells the 5k they want to play even then the 5k may think the 20k is being an arrogant fool and the game would be boring. This is also why I don’t like analysis on in live game by default, it’s setting a bad expectation and culture.
  3. Even by default means there are fewer handicap games played. Handicap games are very useful for internal calibration/consistency of a ranking system. The definition of a 5k vs a 10k is there is that when they play a 5 stone game they should each win about half (actually white has half stone advantage, rating system can take this into account, KGS does). If instead they play even games and you expect the 5k to win 93% or whatever you get much lower quality data. I suspect this even by default rule contributes to the problems in OGS ratings.
  4. Even by default means that you have a lot of games with expected win % far from 50%. This means that non-skill based game results like error submitting move bug or timeouts from lag will distort the rating system. So if I as a dan player play a 10k even I’m expected to win practically 100%, but if I have a shitty connection and lag out in 20% of games then that 10k gets a massive rating boost and me a massive rating drop. Leading to lower quality ratings.

Pros:

6 Likes

Good answer and good reasons :+1: I’m convinced :beer:

2 Likes

Pros:

:joy::joy::joy: Got a deep laugh at that.

I do agree with most of it.
But i like to watch 13x13 on Goquest. Games played without handicap between high ranked with around 2-3 stones difference of levels.
Most of time the result is logical and i still find very, very interesting how the stronger takes control. Sometimes the weaker wins because well he played good and the stronger not that good, it happens too.
When the levels are more close, i enjoy their games too but sometimes it can become a bit boring, like on a soccer field when the two teams are about the same level and nothing happen.
I am not sure i will have such same interest if handicap was used. Maybe handicap is less instructive between high dan? Besides, a 2-3 handicap is quite different from a higher one. I agree thay letting a 5k play a 3d on even is quite ridiculous (can still happen sometimes to feel the power. If it’s not someone who can’t see what’s happening, it’s usually a very quick story).

The dichotomy between casual games and tournaments in real life is really great. There is a time for practice, friendly games where we use handicap, where everyone has his level to get fair games and sometimes progress too. And then there are tournaments where we put aside these and decide to see and prove which level we are, with a diversity of opponents we don’t have usually. The online world is pretty different on this side, players mix the both worlds and it seems not realistic to think we could reproduce the same organization. Not rated games never have been popular.

Note: i would rather watch them on OGS ofc but that still belong to the world of dreams

2 Likes

I fully agree. For these reasons, I think even games should be unranked when the gap between the players is bigger than about 5 ranks.

Such games can still be useful for teaching games about the opening (which games you typically would not finish). So it should be possible to play such even games, but not ranked.

1 Like

The cons mentioned by @Uberdude concern players with a large skill difference. I think that even games between players at most 3 ranks apart are fine.

2 Likes

Compatible opinions - and both work with handicap being the default assumption

2 Likes

Plus, refusing handicap is a bad habit so we should encourage good behaviour as noted by @uberdude above.

Funny story about that here:

https://senseis.xmp.net/?RefuseToTakeHandicap

1 Like