Who do we need to speak with to get default Komi on smaller boards changed?

I started a thread 2 weeks or so ago with respect to why Komi should not be forced to 5.5 across all rulesets across all smaller boards and haven’t received any official response or good explanation on why this is this way.

  • I understand that not everyone agrees on the value of komi – this is perfectly reasonable. What I don’t understand is why 5.5 needs to be forced across all rulesets for competitive games. If you like 5.5, then play with ING rules, there is no need to force others into 5.5, we should be able to agree to disagree, especially for a server that takes different rulesets seriously like OGS.
  • Based on Sensei’s Library, the perfect komi using area scoring is 7 (based on human games as well as KataGo self play): Komi at Sensei's Library
  • Data from GoQuest shows that 7 komi does not produced lopsided winrates for white.

The only argument I’ve seen presented in previous threads is that score changes at increments of 2 for non-seki area scored games, so 5.5 to 7 makes no difference. This argument is deeply flawed because:

  • By the same logic, we should force 5.5 on all area scored games (including 19x19) because it doesn’t make a difference in non-seki games.
  • Odd point seki is real (attached example).

I don’t expect everyone to agree on the value of Komi, but thats the entire point of having different rulesets, we can agree to disagree. As I am fairly new to OGS, I would like to understand, who is the person that needs to be convinced in order to carry out such a change? It seems this issue keeps falling on deaf ears.

5 Likes

@anoek You’ve been mentioned before as someone worth pinging on this – do you have the authority / power to change default komi values on smaller boards to what they are on their 19x19 versions?

Not exactly what you are aiming at (I presume), but in challenge you can change the komi.

2 Likes

In your example, the ranked box isn’t ticked. I believe OP cares more about the komi value in ranked games than whether custom komi can be set in unranked games.

5 Likes

It is the case that only anoek has the power to make this change.

Changes of this nature definitely take time, good clear proposals and usually consensus in the forum [*].

I think that you have started with a good proposal, though there are some things you could improve to make your case better.

“Get Komi default changed on smaller boards”

is a very big and general statement/request.

If I understand it correctly, your actual suggestion is more nuanced than that. It is something along the lines of “change Komi for certain rulesets on certain board sizes”. It might improve the clarity of your suggestion if you elaborate on that more clearly.

Also, generally, a well thought-through proposal doesn’t start with “I demand to speak to the manager”.

An opening like that might be called the “Karen opening”, and is somewhat like a DDK opening at Tengen. Attracts attention, but not necessarily the kind of attention you want.

A better opening is powered less by passion and more by general appeal - if everyone in the forum agreed with your proposal as “obviously a good thing”, the only remaining hurdle would be whether it is easy or hard to implement.


[*] though consensus in the forum is not required if anoek thinks its a good idea :slight_smile: But it certainly can be a useful tool to persuade anoek :wink:

1 Like

I knew this, the issue is that it can’t be ranked. The issue is that 5.5 is being forced in all ranked rulesets.

I empathise with OP and don’t think he’s being a Karen. He made a previous thread detailing why komi should be higher. Lots of people, myself included, agreed. Nothing happened. So he made a new one asking who has the power to make the change.

Rather than more pointless forum arguing, why can’t anoek just change a few numbers in the BE source code? It’s obviously the right thing to do if you are a go player with domain knowledge.

8 Likes

For reference

and

since I believe I merged two threads discussing the same thing.

Presumably the proposal is to change to

  • 7.5 (or 7) for area rules, Chinese, New Zealand, AGA
  • 6.5 for Japanese and Korean
  • I have no idea for Ing rules.

I’m not sure everyone would necessarily agree with 7 over 7.5 although it is supposed to be standard for example for New Zealand rules.

Similarly though, one might want to agree on what should happen to the komi for handicaps? Should that be bumped by one or two or?

3 Likes

is a very big and general statement/request.

Sorry, to be 100% unambiguous, komi values should be changed to their default values on their rulesets (i.e. 6.5 for Japanese, 7.5 for Chinese, 7 for New Zealand, 5.5 for ING). The issue is that 5.5 is being forced on all rulesets and I have yet to see a good reason to do this.

An opening like that might be called the “Karen opening”, and is somewhat like a DDK opening at Tengen. Attracts attention, but not necessarily the kind of attention you want.

Tengen is great in 9x9. I’m not trying to come off as Kareny, I’m just genuinely curious who is the lever mover that can get this done because all attempts to request this previously have been ignored without providing an explanation on why 5.5 is forced on all rulesets. There have been numerous posts already in the forum supporting a move in the komi which have not received an explanation from the server.

Changes of this nature definitely take time, good clear proposals and usually consensus in the forum [*]

This seems like a stalling tactic and an excuse to avoid exploring the underlying arguments. It doesn’t take time to fix this kind of thing, it merely requires changing the setting so that the komi setting logic is the same regardless of board size. Posts asking for fixing default komi have been made time and time again and have received a lot of likes – the only response we received from another mod is that anoek has shown disinterest. If he just came in and said “we don’t care about small boards” that would be enough for me to stop making a fuss about it.

4 Likes

Similarly though, one might want to agree on what should happen to the komi for handicaps? Should that be bumped by one or two or?

The proposal is for komi values to be unaffected by board size. This means that the komi would carry over from whatever is their value in 19x19. That would be 7 for NZ, 7.5 for Chinese, 6.5 for Japanese and Korean.

I think for handicap games any scale that lowers komi in steps of 2 points would work.

3 Likes

I am ambivalent on this issue, but some facts should be pointed out regarding the data. The GoQuest data is useless. Unless they have changed the rules since I played there, the 9x9 games are blitz, so you are comparing apples to bricks. Also, GQ used to make (and I presume still does) peremptory AI decisions on games, precluding late losses by blunders. This contaminates the whole dataset: apples and bricks again.

In addition, I wonder about the validity of any dataset mixing correspondence and live games.

Komi calculations that I have seen have generally been based on high-end games and AI games (the ultimate high end). The relevance of this to low-end games (e.g., DDK) is questionable. Some time back (4 or 5 years) statistics were published in the Forums showing that White on OGS was favored at certain levels—at the current komi. I don’t remember the details, and it is possible that new statistics might show a different result.

Really, the only important statistics are those for OGS. If they show a significant imbalance in the win rates, then komi should be changed. If they do not, then komi should remain the same.

3 Likes

I think the answer is just that’s the way it is, and has been for a long time. (as in why it is “forced” on other rulesets.)

It predates katago and strong AI by quite a while, so it’s not some malicious attempt to shun any AI inspired evaluations, or to make sure that Black has a better winrate for some reason.

Similarly at the time it sounds like people had looked into it and thought it seemed reasonably fair.

Anyway,

Any particular reason for choosing steps of 2 points? I think the scale (difference in players rating) and the size of the steps are kind of important. That would include any handicap stones.

I think if you could link that data it might also help, assuming @anoek has time to look into it.

I personally don’t really mind what the komi is, maybe it would be good to make the komi the same on 9x9, 13x13 and 19x19 only dependent on the ruleset.

1 Like

Poor words. The reason for not changing the komi is not that someone is too lazy to change a couple of numbers in the back end and satisfy your wishes.

  • Change a rule set have a traumatic transition, since all current games and tournaments would still have to apply the old rules and there would be two concurrent rule sets for years.
  • Every time series statistic would also be disrupted.

This said, I do agree that komi should be changed to follow international rule sets, such as Japanese 6.5 and Chinese 7.5.

Making custom community decisions for different size boards and handicap is a much harder topic and I agree with anoek hesitation in doing anything (but I think that discuss it is interesting).
Regarding the arguments to support a community-chosen change in a rule set, I support @Conrad_Melville, it should majorly take into account OGS statistics. And I am very against arguments that use AI inferences, using AI to settle a rule is a pandora box. We are not making rules for AI perfect (or whatever it is) play, we are making rules for humans play.

And praize for @GreenAsJade for commenting on OP tone.

5 Likes

Thanks for your thoughts.

The GoQuest data is useless. Unless they have changed the rules since I played there, the 9x9 games are blitz, so you are comparing apples to bricks.

The right komi amount is right regardless of the speed of the game. There is no reason to think that a komi would be fair at a fast speed but unfair for correspondance games. GoQuest data is absolutely valid in terms of reviewing the fairness of 7 komi. I’ve played 4000+ games (1750 ELO) in GoQuest and have never been affected by a premptory AI decision, this shouldn’t affect the dataset.

Really, the only important statistics are those for OGS. If they show a significant imbalance in the win rates, then komi should be changed.

My argument is that we shouldn’t need to do a statistical analysis to determine the precise komi, but rather that there is no case for forcing 5.5 across all rulesets. One wouldn’t force 5.5 komi across all rulesets in 19x19, the same should be said of smaller board formats. The burden of evidence for why there is a necessity to do this should be on the developers rather than the people protesting against this. It is likely that the swing in winrate from 5.5 to 7.5 will be less than 2% in either 9x9 or 13x13.

5 Likes

This seems wrong. The 2 points increment means 5.5 is (almost) same as 6.5. But in a game with integer scores, fractional komi is not really possible. In go, the 0.5 part (or 0.1 or 0.9) is just a deceptive encoding of “which side wins in case of draw”. So 5.5 is very different from 7 (where 7 is 7 while 5.5 means komi 5 with W winning ties).

This sounds reasonable (apart from ING which is 7.5 iirc). Komi for various rulesets were chosen for a reason. At the very least reason for territory scoring to reward B in B+7 games and area scoring to reward W in the same case.

1 Like

Any particular reason for choosing steps of 2 points? I think the scale (difference in players rating) and the size of the steps are kind of important. That would include any handicap stones.

I actually went with this because this is what it seems its based off currently. I don’t have strong opinions about this system because I detest handicap games as a player and I never use it.

I am partial/sympathetic that updating this specific aspect of how handicap games are carried out in OGS is something that would make the komi update process more tedious as it would require a different adjustment per ruleset potentially.

That’s not the question. The question is whether the dataset is valid. Let’s say all the games in the data were decided by blunders (because it is blitz). The random blunders were roughly evenly distributed, resulting in a roughly even win rate. The even win rate in that example would be the result of random blunders, and would say nothing one way of the other about whether the komi was correct. Datasets need to be legitimately comparable to be compared.

4 Likes

Datasets need to be legitimately comparable to be compared.

We are going off on a strawman / tangent. We can look and argue about any dataset forever, the fact is, no dataset will yield a 100% conclusive determination of what the perfect komi is. We have to accept that this is debatable and that it likely is between 5.5 and 7.5.

No one would argue that Komi is debatable, so why force 5.5 across all competitive rulesets? Lets embrace the fact that different rulesets exist for smaller boards instead of arbitrarily imposing 5.5 on all of them.

4 Likes

:rofl:

3 Likes

I play mostly 9x9 games, when I’m not too busy with daily life. I’ve posted here several times before that I definitely feel that I have a better chance at winning when I play white with the too-large komi given here than when I play black. It is my belief, based on the mathematically chaotic nature of smaller boards, that komi on 9x9 ought to be smaller than komi on 13x13, which in turn should be slightly smaller than the komi on 19x19. Just my opinion.

I usually play with color, komi, and handicap set to automatic, and I usually must be the one to challenge, since most players here are better than my lowly rank of 10k to 8k. Since text contains no emotional cues, let me state that none of this is a complaint. I enjoy the OGS software. I do not enjoy being attacked for stating my opinions, as I have been in this forum in the past.

1 Like