It is currently Mon Mar 18, 2024 8:20 pm

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 42 posts ]  Go to page Previous  1, 2, 3  Next
Author Message
Offline
 Post subject: Re: J2003 problem
Post #21 Posted: Mon Sep 20, 2021 10:39 pm 
Judan

Posts: 6082
Liked others: 0
Was liked: 786
jann wrote:
a local enable rule is almost as big a theoretical flaw as a global ko pass.


In which sense do you speak of a "theoretical flaw"? What, IYO, is the theoretical flaw of a local enable rule? What, IYO, is the theoretical flaw of a global ko pass?

Top
 Profile  
 
Offline
 Post subject: Re: J2003 problem
Post #22 Posted: Mon Sep 20, 2021 11:23 pm 
Lives in sente
User avatar

Posts: 1308
Liked others: 14
Was liked: 153
Rank: German 1 Kyu
"Local enable" is just as little a theoretical flaw as "global enable" or "rebirth enable" is.
"Global ko-pass" is just as little a theoretical flaw as "ko-pass for every single ko" or "no ko-pass" is.

In each case (and also each combination of these) a consistent, error-prone ruleset can be built around.

You may not like the game that comes out of it.
However, this is SOLELY a result of your unfulfilled expectiations.
If you can't live with that, just play another game. There is more than enough choice.

_________________
The really most difficult Go problem ever: https://igohatsuyoron120.de/index.htm
Igo Hatsuyōron #120 (really solved by KataGo)

Top
 Profile  
 
Offline
 Post subject: Re: J2003 problem
Post #23 Posted: Tue Sep 21, 2021 12:30 am 
Lives in gote

Posts: 445
Liked others: 0
Was liked: 37
RobertJasiek wrote:
What, IYO, is the theoretical flaw of a global ko pass?
jann wrote:
The theoretical defect of such rule is it may allow passing for two remote ko fights in one move, for non-Japanese results

RobertJasiek wrote:
What, IYO, is the theoretical flaw of a local enable rule?
jann wrote:
You definitely want to detect ANY kind of problem with a capture, never allow throwing away large parts without consequence

RobertJasiek wrote:
In which sense do you speak of a "theoretical flaw"?

In both of the above cases the problem is the same: the rule invention does not correspond to a meaningful go concept.

  • A local ko pass rule (as used in Japanese rules) has a theoretical meaning of complete ko isolation.
  • A global enable rule (again as used in J89) has a theoretical meaning of detecting any kind of negative consequence/compensation anywhere that may be attached to the capture, which could mean it is not really a simple dead stone capture from sure territory.

A global ko pass rule and a local enable rule can easily lead to incorrect results (see the examples in this thread for both) because they are random inventions without such theoretical ground.

Top
 Profile  
 
Offline
 Post subject: Re: J2003 problem
Post #24 Posted: Tue Sep 21, 2021 12:33 am 
Lives in sente
User avatar

Posts: 1308
Liked others: 14
Was liked: 153
Rank: German 1 Kyu
jann wrote:
A global ko pass rule and a local enable rule can easily lead to incorrect results (see the examples in this thread for both) because they are random inventions without such theoretical ground.

What is YOUR "incorrect" based on IN the respective rule set(s)?

Are YOU sure YOU are IN the right game?
Why don't YOU play a game that follows YOUR interpretation of "meaningful"?

_________________
The really most difficult Go problem ever: https://igohatsuyoron120.de/index.htm
Igo Hatsuyōron #120 (really solved by KataGo)

Top
 Profile  
 
Offline
 Post subject: Re: J2003 problem
Post #25 Posted: Tue Sep 21, 2021 12:42 am 
Lives in sente
User avatar

Posts: 1308
Liked others: 14
Was liked: 153
Rank: German 1 Kyu
jann wrote:
A global ko pass rule and a local enable rule can easily lead to incorrect results (see the examples in this thread for both) because they are random inventions without such theoretical ground.

Same task as for Gérard in another thread:

Click Here To Show Diagram Code
[go]$$B
$$ +---------------------+
$$ | X X . O X X O X X . |
$$ | X O O O X . O X O O |
$$ | . O X X X X O X X O |
$$ | O O X . O O O X . . |
$$ -----------------------
$$[/go]

Reverse tsume-go.
Please be so kind to show us the last seven moves before this position was reached.

_________________
The really most difficult Go problem ever: https://igohatsuyoron120.de/index.htm
Igo Hatsuyōron #120 (really solved by KataGo)

Top
 Profile  
 
Offline
 Post subject: Re: J2003 problem
Post #26 Posted: Tue Sep 21, 2021 4:48 am 
Judan

Posts: 6082
Liked others: 0
Was liked: 786
First are the rules, then come the strategic concepts.

Do you want to make that

First are the strategic concepts, then come the rules?

If one invents a new game, rules design can fit to produce some desired strategic concepts. Go is an exception to game design because rules and strategic concepts were developed in parallel. Even worse, Japanese professional players developed their particular understanding of some desired dependency between rules and strategic concepts but furthermore changed it over time.

Life was perceived as a local concept unless covering the whole board but play for clarifying local life is global. Ko is both local as to currently prohibited intersections and global as to availability of plays.

Hence, it is difficult to identify a rules concept as a flaw on the grounds of general concepts.

Top
 Profile  
 
Offline
 Post subject: Re: J2003 problem
Post #27 Posted: Tue Sep 21, 2021 12:21 pm 
Lives in gote

Posts: 445
Liked others: 0
Was liked: 37
RobertJasiek wrote:
First are the rules, then come the strategic concepts.
Do you want to make that
First are the strategic concepts, then come the rules?

Even when creating a new game from scratch you need rules that are not too arbitrary and carry some concept, to get people interested and convince them there is intellectual value in your game. With Go, the game and its rules already exist for a long time. So not only your newly written rules need to be convincing and consistent, they need to match the real game as well.

Quote:
Click Here To Show Diagram Code
[go]$$B
$$ +--------------------
$$ | X X . O X O X O O O X X X |
$$ | X O O O X O X O . O X . O |
$$ | . O X X X O X . O O X O O |
$$ | O O X . X O X X X O X X O |
$$ | O X X X . O . X O O X . . |
$$ ----------------------
$$[/go]

In the wider version of this example, the bent4 corner itself seem to be alive in your rules. Capturing enables a new permanent W stone on dame next to his 10 stones, barely within your enable region. But your rules are not consistent. It matters if the right side is single ko or double ko (since your ko pass works vs double kos and vs non-ko-based threats, only not vs single kos), and it also depends on the middle (since your enable rule doesn't see the right edge, it matters if B or W gets the last play on dame after the seki collapses).

In comparison, Japanese rules say something like "bent4 is dead since ko fight is not allowed in confirmation, it doesn't matter if the unremovable threat is near or far, neither if it is single ko, double ko or something else". Which can be questioned of course (Japanese isn't my rules preference), but at least it's logical and consistent.

Top
 Profile  
 
Offline
 Post subject: Re: J2003 problem
Post #28 Posted: Tue Sep 21, 2021 1:05 pm 
Lives in sente

Posts: 1226
Liked others: 21
Was liked: 57
Rank: 1d
jann wrote:
RobertJasiek wrote:
First are the rules, then come the strategic concepts.
Do you want to make that
First are the strategic concepts, then come the rules?

Even when creating a new game from scratch you need rules that are not too arbitrary and carry some concept, to get people interested and convince them there is intellectual value in your game. With Go, the game and its rules already exist for a long time. So not only your newly written rules need to be convincing and consistent, they need to match the real game as well.

Quote:
Click Here To Show Diagram Code
[go]$$B
$$ +--------------------
$$ | X X . O X O X O O O X X X |
$$ | X O O O X O X O . O X . O |
$$ | . O X X X O X . O O X O O |
$$ | O O X . X O X X X O X X O |
$$ | O X X X . O . X O O X . . |
$$ ----------------------
$$[/go]

In the wider version of this example, the bent4 corner itself seem to be alive in your rules. Capturing enables a new permanent W stone on dame next to his 10 stones, barely within your enable region. But your rules are not consistent. It matters if the right side is single ko or double ko (since your ko pass works vs double kos and vs non-ko-based threats, only not vs single kos), and it also depends on the middle (since your enable rule doesn't see the right edge, it matters if B or W gets the last play on dame after the seki collapses).

In comparison, Japanese rules say something like "bent4 is dead since ko fight is not allowed in confirmation, it doesn't matter if the unremovable threat is near or far, neither if it is single ko, double ko or something else". Which can be questioned of course (Japanese isn't my rules preference), but at least it's logical and consistent.

BTW what is the result in J2003 for the above position. It seems all groups are alive except the four white stones at the right => black wins by 11 points. Is it true?

Top
 Profile  
 
Offline
 Post subject: Re: J2003 problem
Post #29 Posted: Tue Sep 21, 2021 1:23 pm 
Lives in gote

Posts: 445
Liked others: 0
Was liked: 37
You got me thinking for a moment - nice point on the 4 stones! :)

Top
 Profile  
 
Offline
 Post subject: Re: J2003 problem
Post #30 Posted: Tue Sep 21, 2021 6:07 pm 
Lives in sente
User avatar

Posts: 1308
Liked others: 14
Was liked: 153
Rank: German 1 Kyu
Simply allow hypothetical play inside local-2 only, and everything will be fine.

_________________
The really most difficult Go problem ever: https://igohatsuyoron120.de/index.htm
Igo Hatsuyōron #120 (really solved by KataGo)

Top
 Profile  
 
Offline
 Post subject: Re: J2003 problem
Post #31 Posted: Tue Sep 21, 2021 11:47 pm 
Judan

Posts: 6082
Liked others: 0
Was liked: 786
"J2003 not consistent": What you mean is that different remote environments can result in different local LD status. I agree that J2003 should be improved for the sake of better modelling intention of Japanese professional players.

Concerning specific examples, it is necessary to show that they would remain final positions to be relevant.

Japanese 1989 rules do not say something like bent4 is dead since ko fight is not allowed in confirmation but are independent of the specific shape (bent-4), allow some ko fight but rewrite the confirmation rules to restrict such a ko fight to hopefully make it ineffective.

J1989 intention logical and consistent? Uh, what a joke. First of all, the intention would have to be stated unequivocally.

Allow hypothetical play inside local-2 only? Maybe this solves things but needs exhaustive testing. It might be fine if all examples behave well but local-only play violates the spirit of global play in go.

Top
 Profile  
 
Offline
 Post subject: Re: J2003 problem
Post #32 Posted: Mon Nov 29, 2021 7:44 pm 
Dies in gote

Posts: 29
Liked others: 4
Was liked: 0
Rank: 3k
Are we 100% sure that Japanese pros would rule both the left and right parts in these examples as truly independent for ko analysis? Do we have a source, or is it speculation? My feeling is that indeed they would, but I would not be completely sure without an authoritative source. The position feels like a giant seki/semeai thing among many groups at the same time, where none of the groups has two eyes yet, thus it has some feel as something similar to "Life-and-Death Example 4", where many attacks can be made at the same time but you do not ignore the "other part" in the analysis but rather play the whole thing together. That one does not have a ko though, so it is less problematic.

Without a rule where a pass lifts "more than one" ko shape, how do you solve local moonshine life? (Which leads to infinite repetition without being able to capture under the more direct interpretation of the written 1989 pass for ko rule). There are other things like "1 two-stage-ko + 2 basic-kos seki", molasses ko, round robin kos and similar things that can all easily fail under different interpretations of the "pass for single ko" version.

To me it looks like "the Japanese way" actually is "global play where you can pass for a specific local-ko-region", so that you lift "all the kos in a specific region", regardless of the exact shape. That behaves almost always the same as J2003, unless you have a really complicated tangled mess of related ko's that look independent but are not really separated by completely independent live stones (like in this example).

This example by Jann for example is resolved under such a rule by understanding the left and right kos as different "local-ko-regions", and thus have independent passing for each. But passing for "all the kos in the locality" at the same time solves the complications created by the other weird kos, which do seem to require passing once to free "the whole shape" (or other complex ko rules).

Of course, the difficulty of such "intuitive" rule is to define what exactly a "local ko region" is :)

Top
 Profile  
 
Offline
 Post subject: Re: J2003 problem
Post #33 Posted: Mon Nov 29, 2021 10:08 pm 
Lives in gote

Posts: 445
Liked others: 0
Was liked: 37
santo wrote:
Are we 100% sure that Japanese pros would rule both the left and right parts in these examples as truly independent for ko analysis? Do we have a source, or is it speculation?

What we know is old Japanese rules had bent4 independently dead by explicit rule, and J89 has almost the same because of the pass-for-ko rule. I also think passing for more than one ko simultaneously makes little sense conceptually or theoretically.

Quote:
Without a rule where a pass lifts "more than one" ko shape, how do you solve local moonshine life?

If you mean the closed double ko loop, I think it has two possible explanations. It wouldn't be surprising if J89 pass-for-ko was intended like this: recapturing in a ko without passing for it is forbidden, but later recaptures in the same ko don't require further passes again.

Top
 Profile  
 
Offline
 Post subject: Re: J2003 problem
Post #34 Posted: Tue Nov 30, 2021 4:46 am 
Dies in gote

Posts: 29
Liked others: 4
Was liked: 0
Rank: 3k
But that interpretation of passing for the ko "only once" does not handle both molasses ko (https://senseis.xmp.net/?MolassesKo) and the following "long double ko seki" at the same time (as far as I understand, both cases should be seki under Japanese rules):

Click Here To Show Diagram Code
[go]$$B
$$ +----------------
$$ | O O O O O O O X .
$$ | O X O . O . O X X
$$ | X . X O X O X O X
$$ | . X X X X X X O X
$$ | X X O O O O O O X
$$ | O O O . . . . . .
$$ | . . . . . . . . .
$$[/go]


If you understand passing-for-ko to mean each single ko-shape of marking specifically the two-intersections of the ko-stones, then you need to pass twice to free the two-stage-ko part of the seki, and this positions becomes seki. If you understand each "shape" like the two-stage-ko to be freed completely with a single ko pass, white dies instead.

For the molasses ko, it is the other way around: if you consider that you have to pass twice to completely free the two-stage ko, then one side is considered dead, while it is (anti)seki if a single pass frees the whole two-stage-ko shape. There is not time in these examples to "recapture the ko again later once it has become a normal-play-ko", so that rule does not change it.

These examples were discussed when creating the Katago rules, as the Katago rules use a version of the Japanese pass for ko rule for each specific pair of ko intersections and thus will consider one specific player to die in the molasses ko if left as is, creating a very small difference (there is a single recorded case of molasses ko in actual play, that happened in an amateur game).

The only way I see to handle both positions is that a pass actually acts like a "normal" play somewhere else and thus lifts all the related kos in the region. J2003 have a global ko pass doing precisely that, but then it seems that it can get some interference from the elusive "enable" rule, which is very hard to pinpoint exactly as to when a stone counts as enabled and when it does not. My feeling is that Japanese pros have an implicit notion of locality, and then consider each locality as "a single ko", even if it has 4 kos inside it like this example, it is "one ko" (one long double ko, say) on the board, and you pass for "it" specifically, to separate it from a bent4 in an opposite corner and such, but the whole shape is what you pass for.

Top
 Profile  
 
Offline
 Post subject: Re: J2003 problem
Post #35 Posted: Tue Nov 30, 2021 5:48 am 
Lives in gote

Posts: 445
Liked others: 0
Was liked: 37
santo wrote:
If you understand each "shape" like the two-stage-ko to be freed completely with a single ko pass

No, I didn't mean like that. A ko recapture needs a previous pass for it, this is clearly stated in the rule, and is true for all kos individually, which the opponent took earlier. So no "shape" magic and single pass for all stages.

A known unclear detail is whether a destroyed and later recreated ko mouth at the same intersection is still the same ko or a new ko. I am unaware of anything in J89 that would clarify this. But this needs an answer regardless of pass-once or pass-each-time interpretation.

About the molasses ko, iirc Davies was quoted saying in Japan it would be no result. I also think this is more reasonable (never safe to pass in response to the opponents pass).

Top
 Profile  
 
Offline
 Post subject: Re: J2003 problem
Post #36 Posted: Tue Nov 30, 2021 6:42 am 
Dies in gote

Posts: 29
Liked others: 4
Was liked: 0
Rank: 3k
You mean that it is not safe to pass after the opponent pass in the molasses ko because if you do, then the game goes to scoring and life-and-death considerations rules your group death and the opponent's group alive? A resumption would not occur, as Japanese rules say that the opponent of the player asking for resumption starts, thus no player wants to resume (one player losses if they let the other start, while the other can restart just to prolong the game for no reason, as in any normal double ko, so it would also not lead to no-result).

That is very interesting, indeed. If the actual Japanese result of molasses ko is no result, then this is one less difference between Katago interpretation of the rules and the actual Japanese rules. I must say Katago rules did a really good job to extend the "hybrid" rules idea to have three different phases and use the Japanese pass-for-ko-rule (together with a special kind of "superko" rule for the special second phase, that prevents abusing the double ko seki). That contribution was key to have a "dumb-computer-player-proof" ruleset with Japanese rules consequences.

Top
 Profile  
 
Offline
 Post subject: Re: J2003 problem
Post #37 Posted: Tue Nov 30, 2021 12:56 pm 
Lives in sente
User avatar

Posts: 714
Liked others: 109
Was liked: 138
Rank: Shokyu
Universal go server handle: CDavis7M
I think people are misinterpreting the Japanese Rules. The 1989 Japanese Rules Committee said they were going to find a rationale for past rulings and that's what they did. Consistent rulings were kept and inconsistent rulings were dismissed. The life and death rule requiring a pass is simply the rationale for how scoring works. It is a procedure that can be followed (as in the examples) but it is not a procedure that would be followed. The Japanese Rules Committee never said they were creating a procedure and the rules do not require a procedure. And the procedure, when it is shown in the examples, follows a particular style consistent with proving life and death (e.g., trying to kill rather then recreating the same position, etc.). From what I've read, the referee in a Japanese professional game would make a decision from the definitions in the rules.

Also, the Japanese Rules state the a ko should be specified. They also state that dame are filled to confirm life and death. The supposed issue with these kos is not actually an issue under the Japanese Rules. The "long double ko" is seki and molasses ko is dead. Though yes, if the players would not agree to end the game because of a molasses ko, then there would be no result.

The "destroyed and later recreated ko mount at the same intersection" is still the same ko. The rules define ko by the "shape." If stones are captured and then placed again in the same same then it is the same ko shape. We can't pretend that the players cannot recall the shape being there.

Going all the way back, the topic "J89's pass-for-ko: Misinterpreted in the Western Go World?" is a misinterpretation itself.

Top
 Profile  
 
Offline
 Post subject: Re: J2003 problem
Post #38 Posted: Tue Nov 30, 2021 4:05 pm 
Lives in gote

Posts: 445
Liked others: 0
Was liked: 37
santo wrote:
You mean that it is not safe to pass after the opponent pass in the molasses ko because if you do, then the game goes to scoring and life-and-death considerations rules your group death and the opponent's group alive?

No, I mean not safe to pass because that would be suicidal if normal play would continue (would depend on the stop and getting two moves in a row in case of a resumption).

This is a doubtful situation and it also involves the "both lose" clause, a rare case where J89 does have a logical problem. It is not clear when and how that clause should be invoked (if ever), since usually one side is happy either by scoring the stopped position as is or resuming (on the opponent's request) with him going first.

But imo the existence of this clause shows Japanese rules do not think it is ok to deliberately play a suicidal pass into a situation with an effective killing move left on the board. The stop on two passes is only a technical thing, the players' strategy should be the same whether the stop is on two passes, three passes - or on verbal agreement like in the old times.

Top
 Profile  
 
Offline
 Post subject: Re: J2003 problem
Post #39 Posted: Tue Nov 30, 2021 5:16 pm 
Dies in gote

Posts: 29
Liked others: 4
Was liked: 0
Rank: 3k
My understanding is that it is safe to pass under the Japanese resumption rule: "for whoever asks to resume, then the other player starts". Your opponent pass, you pass - you don't ask for resumption. So that way the game either goes to scoring directly, or it resumes and it is your turn - thus, your pass was safe, the opponent can at most make you play and go on, when you can play once more in the molasses ko and then be the one that passes first, etc.

But, also in addition to that, when using the 1989 "pass for single ko" version so that you must once for each stage of the two-stage-ko, then all of the stones of the player that passes last are deemed dead in life and death analysis, because no matter who starts the analysis, the player that has to retake that one will not get there in time and the opponent always kills everything.

Anyway, the conclusion ends up being the same, that passing is not safe, only that such "safety" depends on the final result according to the rules: if the rules had said seki, it seems to me like passing is safe and you can go for seki, which would then be the result. But the 1989 single-ko-passing-version rules death, not seki, and thus players will not pass, and thus it is no-result during the main game.

Top
 Profile  
 
Offline
 Post subject: Re: J2003 problem
Post #40 Posted: Tue Nov 30, 2021 5:19 pm 
Dies in gote

Posts: 29
Liked others: 4
Was liked: 0
Rank: 3k
Ah, the "both players lose" clause is another beast altogether. I guess that might change things depending on exactly how you interpret it... I am simply ignoring that rule in my analysis.

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 42 posts ]  Go to page Previous  1, 2, 3  Next

All times are UTC - 8 hours [ DST ]


Who is online

Users browsing this forum: No registered users and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group