I have just got this idea by reading that KataGo prefers not going into 3-3 point under a 4-4 stone with the group tax rules because of the 2 point penalty.

Could a very logical and simple territory-style ruleset be derived from the concept of "territory is the amount of excess moves you can make". How to make that into a consistent ruling?

Maybe after the game is considered finished and the play is hypothetical it should be something like this:

1. The side to move doesn't matter anymore, you can access the position from both sides when the side to move has one more move that's a dame.
2. When one side can make an unanswered move, that's a point (in other words, pay 1 stone per pass)




This difference game results in:

a. black first captures one stone (+1), white captures three stones (-3) and then black cannot continue to play without giving up more points (-1)
b. white first captures (-2) and black cannot continue to play (-1)

in other words it is worth 3 points for white in both cases
of course, if you LOSE more than one point by moving first you just pay the 1 point to pass instead

if neither side wants to move, the position is scored as worth nothing for both players (seki)

How would this ruleset work in other weird cases? My point is to have the SHARPNESS of territory counting (wins by 6 and 7 possible), the ability to leave dame on the board in online play, but also be able to judge the result of all finished boards, and allow for hypothetical resolution of life and death

In 2, you are probably talking about "pass stone", right? (I am not sure if you are already familiar with this concept or not.)

Even though you used the word "territory", I guess what you suggest is basically an area scoring with group tax. Yes, group tax resolves a few things, but most of them (if not all) are already resolved by adopting the area scoring. In addition, all the superko and whole board repetition related problems are still up in the air.

_________________
Jaeup Kim
Professor in Physics, Ulsan National Institute of Science and Technology, Korea
Author of the Book "Understanding the Rules of Baduk", available at https://home.unist.ac.kr/professor/jukim/bbs/board.php?bo_table=notice&wr_id=5

Professor Berlekamp did that in the 1980s with No Pass Go with Prisoner Return. See https://senseis.xmp.net/?PrisonerReturn

Here are some links to what I have written about that here.

See this post, forum/viewtopic.php?p=179724#p179724 . See also forum/viewtopic.php?p=179786#p179786 and forum/viewtopic.php?p=195133#p195133 .

Also, recently, this post

https://lifein19x19.com/viewtopic.php?p=261629#p261629

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Bill, I will adress you as the friend of ancient Chinese weiqi rules.

_________________
Zhang-hu 章浒
Committed to the restoration Chinese traditional Weiqi
Research on ancient Weiqi rules & Classic (Dunhuang Classic and the Thirteen Chapters Classic)
From Suzhou, Jiangsu Province, China

I want the final dame order not to matter. In other words, I don't want area scoring because the score difference can only change by two. At the same time, I want a consistent scoring of impasse situations that don't involve "you get no points" judgements

From a practical standpoint it seems 6.5 komi in territory scoring rules is the most fair komi for professionals

Then, what you want is "an area scoring with group tax plus a button Go (or whichever way to achieve the same scoring)". It is still quite different from the real territory scoring, but most (95%?) games will be scored as if one is using the territory scoring, so.. how to name it is up to one's definition.

Actually, I am interested to see if 6.5 komi is still proper when "group tax" is applied. I guess Black has a slightly better chance of reducing the number of separated groups, so Black's advantage may increase a little bit. (Maybe not enough to increase the komi to 7.5, but who knows?)

_________________
Jaeup Kim
Professor in Physics, Ulsan National Institute of Science and Technology, Korea
Author of the Book "Understanding the Rules of Baduk", available at https://home.unist.ac.kr/professor/jukim/bbs/board.php?bo_table=notice&wr_id=5

I am not sure what iopq means by impasse. Three Points without Capturing comes to mind. The most usual impasse is seki.

No Pass Go with Prisoner Return, which is a form of real territory scoring fits the bill. It is territory scoring because, once the dame are filled, you can stop play at any point and count the territory plus prisoners and dead stones to determine the winner. And it has a group tax.

Here is how to play it with a 6.5 komi. At the start of the game Black hands White 7 stones for komi. Black has a special move. At her turn she may rap the table or say "Button" or "Go" or something to indicate that play, even "Pass", unless White has returned a prisoner.

To illustrate how that works, let's suppose that we have a close game when the last dame is filled. For simplicity, assume that neither player has captured any stone during the play.

Case 1, Black fills the last dame and, after accounting for the group tax, has 6 more points of territory plus dead stones on the board than White. The players could agree to stop play at this point and count the score. White would win by 1 point. Suppose that Black does not agree to stop play. Then White returns a prisoner. That costs one point, so the net score is 0. Under no pass go a score of 0 is a loss for the player to move. Since that is Black, Black loses. If Black makes a play, the score is back to a 1 point win for White. However you cut it, White wins.

Case 2. White fills the last dame and, after accounting for the group tax, Black has 7 more points on the board than White. The net score is 0 with Black to play. But Black raps on the table or otherwise indicates her special move. Now the score is 0 with White to play, and White loses.

OC, if there is no reason to continue play after the last dame, you could just count the territory and add 0.5 point to Black's score, or say that in that case Black wins jigo.



Good point!

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

The only thing is that online players don't like filling dame. Of course the only practical way to test a ruleset (for fairness in practice, for example) is to add it to some online server.

Is there any way to chill the scoring where dame are not worth anything even if you forgot to fill them?

On the question of komi you can just run KataGo with the territory ruleset and turn on group tax. I might just do that when I have time for 9x9 and investigate actual lines of play

There is no reason why "territory scoring" and "0 points in seki" have to go together. They are choices that the Japanese rules happen to bundle up together, and no popularly-used ruleset currently has one without the other... but they truly are independent choices that have no bearing on each other. This is why they are separate options in KataGo - try "scoringRule=TERRITORY" along with "taxRule=NONE" - these are perfectly valid rules.

Area scoring with a button of course is an option too, if you want to get rid of all the theoretical tricky bits that come up with endgame disputes in territory scoring and replace them with just "play it out", and have a completely clean scoring rule, and still retain finer granularity. But yes, you'll have to fill the dame.

The rule may be clear in itself but it involves pass-fights as correct strategy. When you add "White passes last", it is not territory scoring but is area scoring.



It is not the only thing. Other things include escapers.

As an online player, I do like filling dame because
- it avoids rules disputes
- clarifies the position
- involves interesting strategy (especially under area scoring)

When the J89 rules came out, given the new definition of seki, it seemed to me that to avoid losing territory because a group's dame had not been filled, to go ahead and fill the dame. I discovered that, since to count the score you fill the dame, anyway, filling the dame cost very little time, if any. Also, always filling the dame means that when you do you are not alerting your opponent that they have to make a protective play.



Sure. Chill the game by, for instance, ⅓ point, and you don't even have to play the final ⅓ point ko fight. There is usually a natural drop of ⅓ point then, anyway. And, OC, you could just play chilled go, but that may cause some difficulty with counting the game if there are fractional plays left. But everybody knows how to count ⅓ point kos.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

It is not clear how to score san moku torazu, since both players will have a valid claim to the other side's stones.

My idea is to prove via difference game that the first person to move loses something. I used the example of a snapback which is not controversial, but it's still kind of an impasse because the white stone is not yet alive and can be captured.

If there's no judgement about it, stones in a snapback would be considered in a seki, forcing white to capture. In the case of a dispute I do want to just be able to play it out. Intuitively, you could just give your opponent a stone and ask them to make an extra move to prove they can kill.

But what I see on GoQuest 9x9 games is people losing when they forget to fill one dame. A lot of the players are used to Japanese rules so they don't realize not filling a dame can cost them points.

Given the snapback in the #1 post of this topic, we can write the game tree this way:

{1|-5||-4}

IOW, White to play captures two Black stones for a local score of -4. Black to play captures one White stone and threatens to connect, for a score of 1, but White can reply with the snapback, for a score of -5. In theory you can prove that is worth -5 (5 points for White), and you could do so with a difference game, where Black has 5 White prisoners that he can return instead of making a play on the board. You could also use Lasker-Maas rules, where each play is made with a prisoner, so if White captures the two Black stones the result is -4 for the local score on the board - 1 point for reducing the White prisoner count by one.



As Gérard Taille has pointed out recently, with no ko threats, Torazu Sanmoku where Black can capture 4 White stones has this reduced game tree:

{2||3|3}

You can show that this is theoretically worth 3 points for Black, or you could use a difference game, as described for the snapback, or you could use Lasker-Maas rules, or you could use No Pass Go with Prisoner Return. The last one also gives you the group tax for free.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

This is not entirely correct. Pure territory scoring identifies dead stones somehow. One way for this is to define territory first (transformable to pass-alive vs resistance), then dead stones are those in territory. In this case, seki exception is in effect even if the rules does not mention seki at all. (Btw I would also prefer territory rules without seki exception - but the above seems to be the most solid approach to life.)

Well, that seems to be the Japanese approach, except that independently living stones need not be pass-alive. But the whole point is to distinguish independently living stones from mutually living stones (by the traditional definition) or living stones with dame (the new definition). That definition need not mention seki, but the distinction from seki is its raison d'etre.



We still have the question, in regular territory go, of why each dead stone counts as the same as one point of territory. Well, that's how things are, and we can speculate why.

In straight No Pass Go dead stones count as territory, but not in a simple way. Here is a small example.

Click Here To Show Diagram Code

[go]$$Wc No Pass Go
$$ -----------------
$$ | . . . . X O O . |
$$ | X X X X X X X X |
$$ | O O O O O O O O |
$$ | . O X . O X X . |
$$ -----------------[/go]

Obviously, the score is equal on this board, with 9 points for each side.

But that's in regular go. In No Pass Go the score is 3 to 3. The top left corner is worth 3 points for Black (Don't ask) and the top right corner is worth 2 points (Ditto). Subtract the group tax of 2 points and you get 3 points for Black. The single eye in the bottom left corner is worth 1 point for White, and, OC, the bottom right corner is worth 2 points. Finally the middle eye on the bottom side is worth 2 points. Subtract the group tax and you get 3 points.

Plainly, territory in straight No Pass Go is quite different from territory in regular go. But both forms of territory include dead stones. It's just that in regular go there is a simple relationship between dead stones and empty territory. In straight No Pass Go they interact.

Suppose that Black to play captures the two stones in the top right, leaving this board.

Click Here To Show Diagram Code

[go]$$Wc No Pass Go
$$ -----------------
$$ | . . . . X . . X |
$$ | X X X X X X X X |
$$ | O O O O O O O O |
$$ | . O X . O X X . |
$$ -----------------[/go]

Now the top right is worth only 1½ points. The prisoners do not count, OC. Suppose that White now plays inside that eye.

Click Here To Show Diagram Code

[go]$$Wc No Pass Go
$$ -----------------
$$ | . . . . X . O X |
$$ | X X X X X X X X |
$$ | O O O O O O O O |
$$ | . O X . O X X . |
$$ -----------------[/go]

Now we are back to 3 to 3.

I won't burden this discussion any further with territory values for No Pass Go, but note that dead stones are an intrinsic part of the definition of territory. In regular go we can consider dead stones and empty points separately, in straight No Pass Go we can't.

In regular go, we don't have to define territory without dead stones. We could say that dead stones are two points of territory, while empty points and prisoners are each one point of territory.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

I am surprised to see that no one mentions this. If button Go is introduced properly to give slight benefit for the first passer, one player can pass when only dames are left and the other player can simply agree to score, just like the way life goes on in many internet Go servers. To make life even easier, the server may let the first player to suggest to score instead of pass. If the second player agrees, then the server treats as if the first player gains the benefit of the first passer automatically.

There is no harm to deny such a process and to keeping playing, but there is nothing to gain by doing so. (Isn't this a well known fact?)

The result will be interesting, but I am sure the reinforcement learning for the KataGo network is performed with "no group tax", and the result will be different from that of the network trained with group tax. But anyway, it is worth investigating.

_________________
Jaeup Kim
Professor in Physics, Ulsan National Institute of Science and Technology, Korea
Author of the Book "Understanding the Rules of Baduk", available at https://home.unist.ac.kr/professor/jukim/bbs/board.php?bo_table=notice&wr_id=5

Using ancient rules of Song Dynasty as below is OK.

1. "dame is not allowed except to balance moves"
2. "even number moves then game over"
3. "territory not including basic liberties/eyes for groups alive forever(group tax)"
4. komi = 0

if you like, komi will be changed.

_________________
Zhang-hu 章浒
Committed to the restoration Chinese traditional Weiqi
Research on ancient Weiqi rules & Classic (Dunhuang Classic and the Thirteen Chapters Classic)
From Suzhou, Jiangsu Province, China

Why would I go through all the trouble of implementing group tax rules in KataGo's engine, and then at the last moment, decide not train the neural net on those rules**?

The reinforcement learning DOES have all of the rules. KataGo's neural net is trained to handle the whole range of the basic rules that the engine is capable of supporting, which are described precisely here: https://lightvector.github.io/KataGo/rules.html. That means that KataGo should have reached far-superhuman levels in group tax Go, exactly at the same time as it learned "normal" Go rules, and should be a master at the specific differences you should need to adjust your strategy and joseki and openings by for group tax rules. And yes, that means that KataGo should also be experienced with things like Button Go already too, because Button Go is also part of the implemented rules.


**Well, actually the latest "g170" run technically isn't trained on non-square boards, like 9x13, or 10x16, or whatever. So that is a thing that I went through the trouble to implement, but didn't train on. Oops. Because neural nets are amazing though, they still seem immensely strong on those boards, probably superhumanly despite no training - because of all the experience learned from normal boards. Also humans probably don't have much experience either on those sizes.

Great. I knew KataGo has quite a few scoring modes, but I admit that I haven't followed the way KataGo is implemented. I am surprised to hear its ultra-flexiblility. (I really need to learn it someday.) I hope other AI makers follow such a strategy. I was tired of them saying "My AI is super-strong, but it can only play under area scoring with 7.5 komi". (Well, that was about a year ago, and maybe the trends have changed now.)

_________________
Jaeup Kim
Professor in Physics, Ulsan National Institute of Science and Technology, Korea
Author of the Book "Understanding the Rules of Baduk", available at https://home.unist.ac.kr/professor/jukim/bbs/board.php?bo_table=notice&wr_id=5

What value button is that? One point? If there's only one dame left it's to your advantage to fill the dame. I think I need a bit more explanation.

At 2 points now you should take the button over some real move like taking a ko