Thoughts Regarding Komi Tie-Breaker

#68

I’m not assuming perfect komi, what I am assuming is that perfect komi is an integer value (which I believe is a logical necessity.)

What this means is that we know for a fact that perfect komi is not a fractional value and so any fractional komi will favor one side or the other. If we want both sides to have a balanced chance of winning, then we must avoid a fractional komi. (Flip a coin in the event of ties may be more arbitrary, but it is absolutely more fair than a fractional komi is.)

#69

It is a necessity for perfect play, but in the imperfect play of us humans, this is not guaranteed.

I still say there should be some research done before we assume perfect komi for human players is a certain way.

#71

I doesn’t matter if the players are human, robot, or dogs. The nature of the rules dictate that correct komi, the points that must be given to white such that, everything else being equal will result in a draw, is an integer value.

The only other option is to say that if everything is exactly equal between the two players, the result will be that black (or white) wins, and neither of these outcomes are appropriate in tournament play.

#72

That’s not correct. Value of komi is simply a constant derived empirically, which gives white and black similar chances of winning a game.

Perfect komi is a different thing, a theoretical concept, which is likely different from the practical value.

1 Like
#73

The rules of the game dictate that such a value will be an integer, not a fraction.

#74

And which rule is that exactly?

If statistically komi of 6.5 would give a perfect ratio of 50% wins for black and white, then using a komi of 6.0 would give black some undeserved draws, whlle komi of 7.0 would do the same to white.

1 Like
#75

You keep saying this, and I’m not sure I know what you mean anymore, since the rules say that komi is currently 6.5 (Japanese) or 7.5 (Chinese).

#76

I am saying that given the rules as they currently stand, it is impossible for a komi of 6.5 or 5.5, or 7.5 or any other fractional komi to produce “a perfect ratio of 50% wins for black and white.”

This is why for example that alpha go teach puts black’s chance of winning at less than 50% before the first move is made.

#77

Again, which rule exactly are you talking about? You never explained that.

And let’s leave AlphaGo aside, because a komi that’s practical for AlphaGo may not be useful for human games.

#78

I’m still not sure what principles you are calling on for this. I have suggested an experiment to test this, and your absolution seems to not be grounded in anything.

#79

Which rule do you think I’m excluding when am talking about “the rules as they currently stand”?

#80

I don’t understand. If you check OGS stats for a game with komi of 6.5, you’ll likely find that the wins by black and white are very similar in numbers. How do you know that its not close to a perfect 50/50 ratio? Suggesting that it’s because of ‘rules of Go’ doesn’t make any sense to me.

#81

It’s funny how this discussion is rather closely following a 2.5-hour argument that occurred in the chat room about a year ago, where the leading contenders were @thought and Pan Piper. Thought also argued that the natural number for komi must be an integer, since go points other than the tie-breaker are integers. Of course, when trying to derive the number from a large data set, you will get a fraction. Someone said that analysis of a large number of pro games had yielded a number very close to 7 under Japanese rules. Since the 0.5 functions, in practical terms, as another full point in tie breaks, it has the effect in such circumstances of bringing the number up to 7. But as I understand it, the 0.5 does not derive from analysis, it is simply tagged on as a tie-breaker.

There is actually a small, OGS, data set that someone compiled, that shows odd variation in win rates by rank, which suggest that komi is a sliding scale depending on rank (see Win rate by players ranks in OGS). Like someone in the discussion above, I would like to see a large analysis of.wins by rank and color, which might establish a more realistic komi at different ranks (I talked about this here: New way of deciding Komi).

1 Like
#82

So, time for some facts I think.

Without komi, if both players played flawlessly under Chinese rules and the game does not contain seki, the ending position must distribute 361 points between the two players. Since both players are perfect, you would want the komi to be such that if they swapped colours, the result would be the same (i.e.: under perfect circumstances, the game would be purely testing the skill of the players, which should be equal). This means that Chinese komi should be an odd integer, or else the 361+komi points are unevenly distributed.

We aren’t perfect players, but we do have a large record of games that have been played. Statistical evidence shows that komi should be around 7 for these games. What is not being accounted for, is that players might play safer if they are ahead, and therefore win with a smaller margin than if the komi was different. Another thing that is not taken into account, is whether the statistical difference in score accurately reflects the “perfect komi” value or if it is skewed in favour of one of the colours (perhaps for us ‘simple humans / computers’ it is easier to play black than it is for the perfect player). Actually the human komi doesn’t even have to be an integer.

So the result is that we can estimate 7 to be the correct komi (hence 6.5 in the Japanese rules and 7.5 in the Chinese, and most of the other komi also are such that they are close to 7), but we don’t have a way to be certain that it is the right amount to be “perfectly” fair. It is statistically fair, though.

The question about whether you allow tie-breakers does not really have much to do with what the perfect komi is, and just depends on the rules that the players agree on before playing. If you absolutely want to avoid this, you could use something like the pie rule (where the first player plays the first black stone, and the second player chooses which colour he plays with) or things like betting on the komi. Both of those are fair ways to play without any of the players feeling he has a disadvantage.

2 Likes
#83

Ugh, there are many confusing ways to put it, and I think I’ve spent enough time in this thread. But let me try to summarise, before checking out from this thread completely :wink:

Imagine deity playing a Game of Go. He plays with itself, and since it knows the result, it just does what most deities do in such situation - ungodly fun while trying to create a perfect game. There is no komi this time, black aims to max it’s score at the end, as white it tries to minimise it. The game is thrilling, and after myriad of whole board kos, dragons barely escaping with their lives, and subtle tesujis, which Michael Redmond would love to explain to us by putting his favourite variations in a tiny 1GB .sgf file, the game ends. Black +8. That’s the ‘perfect komi’. It’s an integer, no discussion about it.

After watching the game on youtube, IGS officially confirms that 8 points is the komi that everybody should use. Want or want not, all the federations change their rules, @anoek hardcodes a strict komi of 8 into OGS code and the world moves on. But something isn’t right… everybody notices that white is better off, and since noone wants to play black, whole Go community dies…

Some players do not give up. @GreenAsJade develops illegal branch of OGS which is deployed deep in the dark net that allows komis breaking the 8 point rule. Numbers of Go vagabonds with @mekriff as their secret leader hang around with a single goal, to find a fair komi and make Go great again. Despite being undercover, the community strives, with solid support from other dark net industries, they organise thousands of dark Go tournaments. They gather millions of data points from players from all sorts of backgrounds, styles and ranks. Eventually the dark secret is discovered. Fair komi does not exist…

That comes as a blow. Who could have thought that was is good for dan players simply doesn’t work for DDKs. Distraught by the failure, the community leaders disperse in the depths of the dark net, digging doge coins and trading them for recreational drugs, they try to erase the experiments failure from memory. They do fail to do one thing, shut down the server. After years from the failure, people still enjoy dark OGS playing out all kinds of games, not all of them fairly. Some of those games do end in draws.

6 Likes
#84

I too will leave with one comment, a comment that will probably sound like I agree with @Jokes_Aside.

I used to play a lot of 5x5 Go on Cosumi and had heard that “proper komi” or “perfect komi” is 25 points (Chinese, or 24 point Japanese.) This means that for black to do anything other than take every white stone played means that white out-played black. However, for the life of me I couldn’t figure out how black could possibly capture every white stone. No matter what I did, white managed to make a 2 eyed, uncapturable, group.

Now, I’m only around 13kyu so I was quite willing to accept that Cosumi was the better player, but with such a small board, I felt I should be able to find the key to winning. I found that key one day when I saw a gif of the so-called perfect 5x5 Go game.

It turns out that in that perfect game, the black player sacrifices 2 stones. Now, at my level of play the idea of sacrificing stones to achieve a better score wasn’t in my vocabulary. Once I saw that game, I was able to find the required moves in subsequent 5x5 games. Now I seem to be able to play black “perfectly” on that size board.

This anecdote is my way of saying that @Jokes_Aside may very well be right. It may be the case that, in order to play perfectly, Black must make moves that most people would never consider. Maybe even moves that pros have never played. It may be that playing black well is just harder than playing white well.

5 Likes
#85

Thank you for confirming my memory that the statistically derived komi is about 7…

Score estimator, final game score disparity (the phantom point)
#86

I got it from here. It also shows how trivial the discussion is, since the the skew for both 6.5 and 7.5 is about half a percent deviated from 50/50. And that’s for 8d players, my guess is that for less experienced players the result gets even more diluted.

#87

I found your (@Jokes_Aside) narrative entertaining, but I have no idea why it was addressed to me.

Actually, I did, a year ago:

Looking back, I see that it was you who echoed this point in post 45 of this thread, which means, as distasteful as it may be, we actually agree on one major point.

Lest I be misunderstood by anyone, I am happy with the status quo, and do not advocate any change. At the same time, I take a jaundiced view of “tie-breakers” in all contests, for reasons well explained in the sport philosophy literature.

1 Like
#88

Here’s one: https://online-go.com/game/13342434
2

I didn’t know this was possible.

1 Like
Which rule set do you think is better, Chinese or Japanese?
Bug in scoring