Then how are draws possible?
Fair Komi is that Komi which maximizes draws. If the button(s) net a .5 to one of the players, Fair Komi will end in a .5. This is not guaranteed to be Perfect Komi, which is the issue none of these are addressing
Yes we have gotten off topic. And yes I agree that for beginners none of this matters and they should learn area rules (as I said in my first post above).
That said, we’ve covered some interesting ground in terms of the reasons for these rules differences. I’ve learned some new things. And I think the diversion from the original topic contains a lot of useful info for people interested in rules details.
2 Likes
Same way as under area now: strange seki that splits neutral points in a fractional way.
You could also think of this as having a fixed komi of 7 with a 0.5 point move. That move nets to 0 on average. Half the games it goes to one player, half to the other.
Since territory is perfect under these conditions, then these rules should be as well. And it should still be fair since fair komi is 7, it’s just that we’ve made ties much less frequent. That komi should still maximize them.
Again, this is theory, I don’t know if an AI analysis exists that backs it up.
I haven’t been able to find much clarity about how to exactly define this concept either. I think this complexity makes such a rule clearly inferior to simply using a button.
From the Taiwan rule Sensei’s Library page (which does not offer much clarity and instead states that there are some ambiguities), there’s a link to some interesting discussion here: GoBase.org - Ikeda Toshio, On the Rules of Go
Those are rare circumstances, so a different Komi is going to increase the draw rate, and thus you don’t have Fair Komi there
Just like 6.0 is not Fair Komi for Area Scoring
Well, perhaps I’m not considering the impact of the button properly. But it seems to me that if we had area rules and a komi of 7 + at the end, we flipped a coin and gave half a point to the winner of the toss, then all the appropriate game theoretical properties would be unaltered.
So it would be fair in the sense that we care about, that since it is a perfect information game, it should be theoretically drawn. And perfect in the sense that both sides have equal chances given equivalent mistakes from the other.
Am I missing something critical?
Today I learned that the original button rule called for 3 passes. The game ends with 2 passes in a row after the first pass.
Subsequent “first pass” rules have dropped that and ended up with strange problems.
His original version is substantially clearer about what is going on, the importance of sente, etc.
yes, you’re interpreting “Fair Komi” as being komi which is “fair” in the dictionary sense. That is not what I mean by Fair Komi: I mean that Komi value which maximizes draws. A coin flip to give 0.5 points, forces Fair Komi to also end in .5, because in order for draws to exist, the score difference must be an integer
I’m treating it game theoretically, which because we added a random element, now means maximizing them in expectation along with makingthe outcome unbiased to either player. Something that 7 still does assuming it was fair and perfect before we added the coin toss.
Maximizing draws for the sake of it, isn’t meaningful.
So, based on this crude analogy, I would think that a parity correction would maintain the relevant properties in a practical sense.
Then you’re not getting Fair = Perfect Komi, you’re claiming Fair Komi doesn’t matter. That’s a completely different claim
You’re not responding to me, you’re defining away the disagreement and then claiming your position is compatible with mine
For that matter, is anyone aware of a complete catalog of “hybrid” rules variations? Even if there isn’t a single place that parses out all the minute differences between these proposed compromises, having a complete list would be useful. (Though, I wish there was an automated way to check this type of thing against the various rules beast collections for a large number of minor rules changes).
Here is a partial attempt:
Aside from the already linked pages on Sensies and Ikeda’s book, I found this discussion on 19x19 where Bill Spight points out that there are major problems with parity corrections attempting to stick to 2 passes without the “double button” mechanism. In particular with the specific one that @jannn was concerned about. Spight recommends a variation on double button:
It is easy to modify AGA scoring to incorporate the equivalent of a button. Under current rules White must pass last, handing over a pass stone. Under the equivalent of button go White must pass last only if Black made the first pass, otherwise Black must pass last. OC, with the pass stones, territory counting is used.
(It seems from the discussion that komi would need to be 6.5 since ideal komi is 7 but this double button rule adds in an extra half-point in general.)
There are also Jaisek’s commentaries on the Ikeda I rules. Of particular interest is this pdf modifying the Ikeda territory I rules to be more in line with how territory scoring handles end game kos. This is what I referenced earlier as seeming to be very close to what Katago actually does, and I wonder if any of the Katago developers were aware of it when they created their territory mode (which is even more faithful, but requires 3 phases and a ko-pass move).
I wonder if there is a way to construct an area-with-button version of this similar to how Ikeda area III and territory I correspond. For example, would it match if you adopted his proposed ko rules instead of super ko and modified Ikeda area III to say that after the first pass, the game ended after 3 passes in a row (keeping the rule that if white passes first, 1/2 point is subtracted from black and added to white)?
You are right. I am clearly misunderstanding you. I think I made a bad assumption based on what “fair” and “perfect” mean in other contexts. I did not mean to offend, only to ask if you had seen an analysis, using AI or otherwise, that looked at these properties for the various hybrid rules.
I am open to understanding your meaning of “fair” = “perfect” better if you are open to explaining it.
Here’s the definition that I gave right near the start
Beware the different common use (see komi on senseis).
This follows from what I wrote in earlier posts (you may want to skim through them again as I feel you missed quite a few things). Key points:
1: area_score = territory_score + 1_if_even(territory_score)
(from B’s view and w/o odd seki or onesided passing)
2: winning probabily under area scoring ~= the probability of reaching the closest territory score that rounds to a winning area score
3: depending on current territory parity, one color can always make a free 1pt mistake, which will be canceled by the also flipped last move parity (if_even() above) - assuming perfect play onwards. in 9x9 opening it is W who has this leeway (you can verify this on the links)
4: for a human W however, losing a territory point this early means he will need to reach that closest territory score from -1pt territory deficit, which is still possible but harder. Making his allowed -1 mistake now means he is not allowed any more mistake in later moves (as well as transfering this right for 1 future mistake to B). W now has a narrower path to victory while B has a wider one, for a similar win% loss as playing against 1 territory komi diff.
If you mean the example on senseis, in that case W makes the first pass as well as the last move and the last pass (assuming two pass stops), so it seems this rule would still award him an incorrect point (since area and territory agree there, there is no room or need for any adjustment). Also the played stones (and the number of passes) are equal which I think is the most robust indicator, closest to territory / human logic.
1 Like
After reading that and searching a bit on OGF, it seems I (or whoever I saw it from first) reversed the definitions of Fair and Perfect Komi. I’m not too concerned for 2 reasons, though: I nearly always define it when I bring it up, and I noticed a distinct lack of consistency in how the terms there used on OGF, so I’d definitely still have to define it whichever way I used it
@hexahedron did a good job making this make sense for me with the idea of rewarding ending in sente. That makes the rest of your example make sense now.
Yes the last example here.
Using area scoring, with the button as a dedicated move, I agree with you. Same with most of the variations on the idea that I’ve found.
But, I think I’m getting confused about the two-button version using pass stones and equivalence scoring with the rule “the first player to pass hands over a pass stone, as usual. If the last player to pass is the same as the first player to pass, she does not hand over a pass stone for the last pass” (and 2 passes ends). (Or, equivalently, they hand over the pass stone, but are awarded 1 additional point for both passing first and last.)
Under that rule, for your example, since white passes last in the second sequence and also passed first, they don’t hand over a second pass stone. The result is that under both sequences, players have each given 1 pass stone. Under the second sequence, they have each taken the ko once. And the captured white ko threat stone offsets for the extra stone black had to play in response taking away from his territory.
So shouldn’t the score now be the same under both sequences under this rule?
I think I’m missing something obvious that creates a problem because, it seems like under normal AGA rules, where white still has to give the extra pass stone when they pass last, they end up worse off under the second sequence, even though the area on the board ends up being the same. Does something about this sequence break equivalence scoring? Or is it just breaking my head?
I expect I am doing something silly here, but I am not seeing it. If you could point it out, I’d appreciate it.
Regardless, is there a straightforward implementation of the Taiwan rule that doesn’t have practical difficulties when it comes to teire? Difficulty defining “the last competitive move” seems to be the motivation for button rules to begin with.
Also, when you said “assuming two pass stops”, was that for clarity? Or is there some version with more passes that doesn’t have this problem?
In the second sequence I count two passes for each (and no B discount).
See senseis for possible implementations. I think this may be worked out without the “competitive” part, preferably in connection to handicap compensation points, but I suppose the issue is that in Taiwan this rule got edged out by Ing rules, so could not really mature in the usual slow way.
Well, Ing’s rule also didn’t pan out and never got very popular (only tournaments sponsored or associated with the Ing’s foundation still use them). We mostly just like everyone else, adhere to the area scoring with a simplified 185 black win threshold (don’t even bother with the komi) in amateur tournaments. And the pros mostly adapt the territory scoring just like the Japanese rules.
I could only find some vague descriptions, not a proposed rules text.
Is the idea: “If at the end of the game, black has played more stones than white, he pays white and additional 1 point of komi”? In practice, do you track that by keeping pass stones separate and doing some math to then derive the correct adjustment?
Has anyone done a major examination of this to confirm that it doesn’t introduce other weird corner cases?
It seems like this was the intent behind the World Mind Sports rules, even though that’s not how they worked in practice.
Am I correct in thinking that the issue is unbalanced passing throwing off the naive parity rule? Or is there more to it?
Edit: by “math”, I specifically mean: use AGA rules but with 6.5 komi. Track pass stones separately (still used for fill-in). At the end, black pays extra komi to white equal to white’s excess number of passes.
If that sensei’s page is right, then this should achieve the equal stone result you want.
1 Like
Nice, 9 wins so still effectively 7.5 I guess. How does this simplified view see ko and long cycles?
I doubt the Taiwan rule was ever well documented or studied, even in its historical/“competitive” variant. Since it predates the button/pass approach, it’s advantages vs early pass problems were probably not even recognised in its time.
Wrt combining with pass stones an encore is another option (no pass stone before first stop avoids parity issues instead of triggering-then-correcting them).