What should komi be (perfect komi, for Japanese rules & Chinese rules)?

General Go Discussion
1

I think 6.5 is the best komi. What do you think?

2

7.0 for area
6.0 for territory

3 Likes
3

If the perfect komi is 7.0 for area & 6.0 for territory scince komi is 7.5 for area & 6.5 for terriotory white is better by 0.5 in go?

4

Yes correct white wins a small majority of games. They still do even with integer komi, but it’s closer than how we do it now.

2 Likes
5
1 Like
6

Um, no, I never open that thread, but I did open this one…

I guess by “perfect komi” you mean white should win 50% of the time? Then it has to be a whole number. If both players are playing perfectly, then every game will have the same result. So for 50%, it has to be all draws.

1 Like
7

Same, but wish I didn’t :laughing:

8

“white should win 50% of the time” does not necessarily imply whole-number komi. There may be some human bias (as @BHydden mentioned) where even with integer komi, one side may be preferred.

The criterion that requires whole-number komi is “result is a tie, given perfect play”

Example:

We can say 5x5 perfect komi should be 25, because a skilled player can always take all as black. However, given imperfect players, we will end up with some white wins, so the 50% criterion is not met.

1 Like
9

yeah, I think they just misspoke. as we all know, perfect (best play is a draw) komi is 6.0 territory and 7.0 area for board sizes 9x9+. fair (closest to 50/50 winrate) komi is 6.5 territory and 7.0 area

perfect = fair komi under area but not territory is one of the big advantages of area scoring

1 Like
10

Don’t forget the possibility of no-result! For when you don’t have a full superko rule (and in professional Go you often don’t).

Though unlikely, in theory the following is a possible state of affairs in such a case:

  • There exists a strategy for black where black can guarantee either least 7 points on the board OR achieve a no-result via triple ko or other long cycle.
  • There exists a strategy for white where white can guarantee either black gets at most 6 points on the board OR achieve a no-result via triple ko or other long cycle.

Then, depending on how you value “no-result” relative to a draw, any komi in the range from 6 to 7 could be considered “perfect”. Symmetry and the desire to make the margin from the nearest imperfect komi equal on both sides would then point to 6.5, a non-integer value, as the “most perfect” komi in such a case. :slight_smile:

2 Likes
11

It’s probably not the case. Still, it would be a funny cosmic joke on Go-playing humanity by the universe if somehow it did turn out that optimal play in Go under one of the commonly used rulesets always was a forced triple ko like that.

3 Likes
12

Wait then what’s the %of Black winnig with Japanese rules and 6.5 komi?

13

but some thing is wierd white has only 0.3 point better (from AI) than black how can 0.3 point be possible

14

I found this

Handicap (Komi) Black Win Rate Black Advantage
5.5 points 58.4% Black is ahead by 0.7 points
6.0 points 53.0% Black is ahead by 0.2 points
6.5 points 47.6% Black is behind by 0.3 points
7.0 points 42.2% Black is behind by 0.8 points
15

(Japanese rules)

16

so white is better in go

17

white is better by 4.8% in go then black

18

People are not AIs, the AI winrate is very different than using actual human pro games for statistics. And for human pro games, the statistic is very close to 50/50 at 6.5 komi (in fact, last year, the statistic slightly favor black at 50.3%)

http://mamumamu0413.web.fc2.com/statistics/handicap/data.html

1 Like
19

Is that for all recorded history or post AI revolution?

1 Like
20

It has year-to-year breakdown and different komi, just check the website, it has very detailed records. And post AI era has the most even winrate between black and white

1 Like