Life In 19x19

The komi value that achieves winning ratios by Black and White that are as close to each other as possible can be either an integer or a half integer.

In 9x9 it seems the right value (with area scoring at least) is 7. Computer tournaments use that value because 7.5 turns out to be a bit too much.

Of course, with an integer komi, ties are possible. So I was wondering if using a simple tie-breaker like first passer wins ties or Button Go would keep win ratios basically the same or if one of the colors would be inherently favored.

I once wrote this proposed solution: http://senseis.xmp.net/?ResolvingJigoByKoThreats

HermanHiddema wrote:I once wrote this proposed solution: http://senseis.xmp.net/?ResolvingJigoByKoThreats

Very nice!

EDIT: An inconvenient of both rules is that they force players to capture dead groups, I think. A player who lost the pass fight can force game resumption by disagreeing on the status of their dead groups so that their opponent is forced to capture them and pass last (or not pass first).

Theoretically, the perfect komi is an integer, and any fractional value is not 100% fair.

That's true, but it's perfectly possible that a fractional value works best in practice. The score difference in a game played perfectly needn't correspond to the komi value that brings the most statistical balance in games played imperfectly.

luigi wrote:EDIT: An inconvenient of both rules is that they force players to capture dead groups, I think. A player who lost the pass fight can force game resumption by disagreeing on the status of their dead groups so that their opponent is forced to capture them and pass last (or not pass first).

Of course, this can be solved by cancelling only the last pass in the event of a resumption. This means that whoever passed last plays first after a resumption.

Button Go achieves something similar as taking the button doesn't count as a pass and is therefore not cancelled in the event of a resumption.

I actually prefer "One player chooses (half-integer) Komi, the other one chooses sides." the most

I don't think there will be any 100% "fair" value for komi as long as we make it tie-breaking, no matter it is breaking by fraction value or by extra rules.

Let's assume two Gods play Go in a minimax manner (because Gods have infinite computational power so they can brute-force the whole game tree), and the games are played with the super ko rule (because Gods have perfect memory and don't feel any difference in simple ko and eternal life). In such case, each game they play will always end with the same score difference no matter what kind of scoring method they choose. Now we set that score difference as the komi value. This will be the "fair komi" for Gods or perfect players, which allow Gods to declare their games a draw.

Any other definition of "fair komi" will be subjective to the players' intelligence, and increase/decrease when people find new ways to play stronger as black or as white.

luigi wrote:

HermanHiddema wrote:Theoretically, the perfect komi is an integer, and any fractional value is not 100% fair.

That's true, but it's perfectly possible that a fractional value works best in practice. The score difference in a game played perfectly needn't correspond to the komi value that brings the most statistical balance in games played imperfectly

Yes, but seemingly it still happens to be the same in practice, since AFAIK all current data shows that 6.5 is a bit too small, and 7.5 is a bit too high. Which is not surprising considering how close current top levels are (supposedly) to perfect play. From a fairness perspective ddk komi is more practical problem.

billyswong wrote:I don't think there will be any 100% "fair" value for komi as long as we make it tie-breaking, no matter it is breaking by fraction value or by extra rules.

No question about that.

I think a better way of phrasing my original question would be: if a certain integer komi value yields equal win rates for Black and White plus a number of ties, will adding the button as a tiebreaker on top of that still yield equal win rates for Black and White?

Reddit's AMA reveal that DeepMind's Julian Schrittwieser experiment and think komi 7.5 (probably with Chinese rule) favor White with 55% winrate. (Confirm what everyone's afraid of)

But the most interesting thing actually come from DeepZen's Hideki Kato that said Zen think komi 6.5 (Japanese rule) slightly favor white. (I talk to him in KGS room)

Combine with contrasting statistics from 8D+ game of GoBase data in 2008 that black win a little bit more (50.58%) in Japanese rule with komi 6.5, we can roughly conclude that Japanese rule is more fair than Chinese rule.

pookpooi wrote:Reddit's AMA reveal that DeepMind's Julian Schrittwieser experiment and think komi 7.5 (probably with Chinese rule) favor White with 55% winrate. (Confirm what everyone's afraid of)

But the most interesting thing actually come from DeepZen's Hideki Kato that said Zen think komi 6.5 (Japanese rule) slightly favor white. (I talk to him in KGS room)

Combine with contrasting statistics from 8D+ game of GoBase data in 2008 that black win a little bit more (50.58%) in Japanese rule with komi 6.5, we can roughly conclude that Japanese rule is more fair than Chinese rule.

And 6.5 komi under Japanese rules is almost equivalent to 7 komi plus button under Chinese rules. The message is clear...

luigi wrote:And 6.5 komi under Japanese rules is almost equivalent to 7 komi plus button under Chinese rules. The message is clear...

Agree, but I'm still standing my statement.
In Chinese rule both statistics and AI agreed that white has advantage with 7.5 komi.
In Japanese rule DeepZen also agree that 6.5 komi favor white but statistics says otherwise (black slightly win more), thus making it more fair.
Now the reason behind what's make black win more in Japanese rule is to be discussed. Both rule should give similar result, isn't it?

pookpooi wrote:

luigi wrote:And 6.5 komi under Japanese rules is almost equivalent to 7 komi plus button under Chinese rules. The message is clear...

Agree, but I'm still standing my statement.
In Chinese rule both statistics and AI agreed that white has advantage with 7.5 komi.
In Japanese rule DeepZen also agree that 6.5 komi favor white but statistics says otherwise (black slightly win more), thus making it more fair.
Now the reason behind what's make black win more in Japanese rule is to be discussed. Both rule should give similar result, isn't it?

I don't understand. Black wins more in Japanese rule compared to what?

luigi wrote:I don't understand. Black wins more in Japanese rule compared to what?

Compare to Chinese rule.
Black only win 49.51% in Chinese rule

source https://senseis.xmp.net/?Komi%2FStatistics

pookpooi wrote:

luigi wrote:I don't understand. Black wins more in Japanese rule compared to what?

Compare to Chinese rule.
Black only win 49.51% in Chinese rule

source https://senseis.xmp.net/?Komi%2FStatistics

I see. Well, that's because 7.5 komi under Chinese rules is more than 6.5 komi under Japanese rules.

As I said, 6.5 komi under Japanese rules is nearly equivalent to 7 komi plus button under Chinese rules, which is less than 7.5 komi without button under Chinese rules.

It's also not surprising that White's advantage with 7.5 komi under Chinese rules is more apparent in AlphaGo games than it is in human games, as AlphaGo is closer to perfect play.

pookpooi wrote:

luigi wrote:And 6.5 komi under Japanese rules is almost equivalent to 7 komi plus button under Chinese rules. The message is clear...

Agree, but I'm still standing my statement.
In Chinese rule both statistics and AI agreed that white has advantage with 7.5 komi.
In Japanese rule DeepZen also agree that 6.5 komi favor white but statistics says otherwise (black slightly win more), thus making it more fair.
Now the reason behind what's make black win more in Japanese rule is to be discussed. Both rule should give similar result, isn't it?

Hmm.... DeepZen is not strong enough for me to trust its own statistics as heavy as human stats. A person can be stronger in black or stronger in white. Same for AI.