Life In 19x19

Bantari wrote: Hmm... so you have a method to evaluate something, but no method to tell you if you can or cannot apply the first method to begin with?

Interesting...

A pseudometamethod?

Bantari wrote:so you have a method to evaluate something, but no method to tell you if you can or cannot apply the first method to begin with?

1) It depends on what one calls "a method". For go theory, I am defensive with using the word only for step-by-step advice with very clear specification of each step and of the order in which to proceed from step to step. The joseki evaluation method is clear enough to deserve the attribute "method". The context of having a meaningful sequence, or a position which must have been created by a meaningful sequence, I have not worked out sufficiently precisely yet to deserve the attribute "method". What you see here is that I have not solved the entire go theory yet but I have only reasonably solved part of all go theory. You may scold me for not having solved the whole game yet but only done better than everybody else by reasonably solving the numerical joseki evaluation part;)

2) Although I cannot supply a method yet for judging whether a sequence is meaningful, I have provided lots of aspects and related principles relevant for such a judgement. A meaningful sequence must consist of meaningful moves. A meaningful move is a move achieving good purposes for its player and avoiding clearly inferior "purposes". Aspects for such purposes can be listed: territory, influence, development directions, weaknesses etc etc etc (study strategic concepts, strategic choices, positional judgement, positional context etc). For each such aspect, the player prefers his advantage and the opponent's disadvantage. So there are related principles, such as "a move shall increase the territory count in the player's favour and avoid decreasing it", for every aspect. Furthermore, the combination of all such aspects should be advantageous for the player and disadvantageous for the opponent. Finally, the combination of all such aspects should be as good for the player as possible. Especially concerning this combination, go theory has not evolved far enough for the general case of an arbitrary whole board position. Distinguishing meaningful from bad moves remains in part ambiguous. However, in practice, it often is clear. E.g., playing a neutral stone hardly achieving anything is clearly worse than playing a territory-making, influence-making or own-weakness-eliminating move.

However, when we are not confronted with the general case but have a joseki-like situation, the judgement whether a sequence is meaningful and its moves are close to being as meaningful as possible is much clearer because the most important aspects (stone difference, territory, influence, other major strategic advantages, other major strategic disadvantages) can be judged very well in almost all cases. Besides, these few (classes of) aspects suffice for the local judgement. A meaningful-assessment of these few aspects suffices to feed my joseki evaluation method.

So although the presuppositions of what constitutes a meaningful sequence cannot be described as a method in general case, the lists of relevant aspects and principles relevant for a joseki-like situation are good enough to suggest how future research can work out the combination of these things to develop another method. (Even better: I have provided definitions or sketches for definitions for the low level terms, which would be applied for that other method.)

3) My joseki evaluation method presumes close-to-equality of both players' moves so that it is meaningful to relate one excess move to a) 7 points komi, b) 1 extra influence stone or c) playing of a joseki follow-up. This means that the method requires meaningful, striving-to-be-optimal moves by the players so that is it meaningful to relate every move to one of the optimal ideals (a), (b) or (c). Contrarily, too bad moves must not be related because their value does not equal, and cannot be viewed to approximate, a move of such ideal value. I.e., although there is no method for the general case of any good or bad moves, one can be the more confident the more the moves approach the ideal of having such meaning. One can judge the results (of the aforementioned aspects) also after a sequence of good and bad moves, but one cannot necessarily be sure any longer from my joseki evaluation alone whether the sequence is joseki-like because it measures the results but does not detect whether the result might have been created by also bad moves, of which some might cancel each other.

I have an idea: apply my joseki evaluation method after every move. If it says "balanced" at each moment, then each move can be viewed as "good". This presumes the move sequence is given. (Of course, it does not work when somebody wants to discuss an illegal sequence by rearranging a stone where there is another stone.)

4) Instead of trying to let sound my joseki evaluation method irrelevant because of not also solving the entire go theory immediately, you should appreciate how frequently applicable and meaningful it is in joseki-like situations because, in practice, it is very often possible to distinguish meaningful from meaningless moves. Difficulties arise only when one wants to achieve more and distinguish margerine from butter so as to choose the best joseki move variation in the global positional context.

My joseki evaluation method presumes close-to-equality of both players' moves

Robert, as I understand this and a similar earlier remark, you are saying that you assume that some individuals (pros) or a collective of individuals have already "solved" go theory and come up with sequences that are widely agreed to be equal. What you have done is to provide a numerical method, which you regard as superior because it can be applied by non-pros. Pros have used another but opaque method (and were first). Is that right?

A problem I have with that sort of approach is that it's as much like mathematical rigour as a beauty pageant. A group of women (qua opening sequences) who have already been told by friends, relatives, prom nights, etc that they are beautiful (qua evenly assessed joskeis) are put forward for a so-called rigorous evaluation in which a panel uses numbers to grade each contestant. 1-5 for a curvaceous figure, 1-5 for "personality", etc). By and large they end up with a winner that causes little controversy, so they can claim to have "solved" the evaluation of beauty.

But the problem is that the panel is part of the same cultural background as the friends and relatives, so they are not really measuring anything - they are just applying the same (pseudo)cultural norms. In other words, such methods are circular.

I can see the value of your method for a computer. But for a human I have doubts. In fact I could claim you are demeaning josekis, objectifying them as numerical objects. Ban joseki books! Josekis of the world rise up, and burn your bikinis!

Did anyone comment on the second point. W12 seems to be important due to the presence of N15, old coffee for most commenters it seems, but maybe beneficial to others reading this thread.

Click Here To Show Diagram Code

[go]$$W Not played - White shut in
$$ ------------------
$$ | . . . . . . . . .
$$ | . . 1 . . . . . .
$$ | . X O . O . . X .
$$ | . . X O . . . . .
$$ | . . X 3 2 5 B . .
$$ | . . . 4 O 6 . . .
$$ | . . X . . . . . .
$$ | . . . . . . . . .
$$ | . . . . . . . . .[/go]

John,

I am not saying that some would have solved the entire go theory. In reply to Knotwilg, Bantari and DrStraw, I have said that my method should not be depreciated on the grounds that I (nor anybody else) has risen everything of related go theory (and ultimately the entire go theory) to the same standard as my method, and that therefore not everything else should already be called a "method".

I do not regard my method to be superior because of being applicable by non-pros. It is superior because

- it can be applied by non-pros and pros (and, with some extra effort, even programs),
- this is so independently of a consensus of professionals' opinions on what is or is not joseki,
- this is so independently on how frequently applied some sequence has been in (especially professional) games,
- therefore the method is applicable during one's games and one need not wait years until possibly some professionals express opinions, reach a consensus or apply particular sequences sufficiently frequently to allow for empirical data,
- it is applicable to every joseki-like local position (applicability rate ca. 99.75% for josekis that do not necessarily immediately evolve as middle game fights and therefore mainly depend on the positional context)
- AFAIK, it is the only joseki evaluation method systematically relating the stone difference to territory and influence,
- AFAIK, it is the only joseki evaluation method expressing amounts of influence at all and therefore allowing systematically relating influence to stone difference and territory,
- AFAIK, it is the only general joseki evaluation method almost always allowing to distinguish josekis from non-josekis,
- AFAIK, it is the only joseki evaluation method explaining different joseki value types (such as 'ideal equality' or 'territory versus influence) by a common, consistent evaluation scheme.

Professionals have not used only one other "method", but they have been, and are, using a variety of "methods", such as tewari or 'forming a consensus by trial and error'. All such methods have some use, but fail to be generally applicable and achieve the aforementioned advantages.

I do not share your scepticism about numerical approaches. Numbers of stones and their difference are numbers. Amounts of territories and their difference are numbers. Numbers of outside influence stones and their difference are numbers. Komi is a number. The number one of an influence stone elsewhere on the board is a number. The territory per influence ratio is number. My statistics that every ordinary joseki has a ratio between 1.5 and 3.5 relies on numbers that can be checked by recalculating them. I do not care about whether beauty can be expressed in numbers, but I care when numbers can be consistently used in a method. The method and its theoretical foundation let such numbers have good meaning.

The method is not circular. Of course not. You should know better. Study the entirety of my research results and you will notice that I start with the rules, go to the lowest level terms and work my way up to the higher level terms and concepts. There may still be some gaps, but the bottom-up structure is there. My joseki evaluation method applies first move value, stone difference, territory count and influence stone difference, which I have already defined (sometimes formally, sometimes with less rigour). There is no vicious circle.

Your doubts about human applicability are refuted the most easily: I apply, and profit from, my joseki evaluation method in my games or discussion about games. It is straightforward to determine stone difference, territory count, influence stone difference, calculate the ratio, interpret it etc; everybody can do so if only he wants to learn and apply.

Of course, I am not demeaning josekis. Quite contrarily:

- the method appreciates josekis as something so outstanding that it can even be recognised numerically,
- the method is deliberately approximate (exact values are not necessarily required) and flexible (applicable to different joseki types, ackowledging dominating strategic concepts, having a variety of value transformation options) to allow for a variety in josekis existing or to be discovered later,
- the method can be embedded in almost arbitrary global positional contexts.

Of course, my method does not oppose joseki books. Quite contrarily, it confirms, summarises and combines (most of) their assessments. However, my method greatly enriches the scope of what traditional joseki books tell their readers. There are times when I criticise traditional theory as wrong. Traditional joseki theory is not wrong - it is "only" lacking generality of evaluation and consistent discussion of strategic decisions. My method goes beyond traditional understanding to raise joseki understanding to a new level of evaluation.

EDITED

Bantari wrote: Hmm... so you have a method to evaluate something, but no method to tell you if you can or cannot apply the first method to begin with?

Interesting...

It’s a little mysterious to me why Robert’s calculation is met with such snarkiness, but thinking back, that's how Bantari communicates almost always. Clearly Robert is not claiming to have solved go, so the method cannot answer every question --- rather it's a simple heuristic, which assigns a number to a particular type of local situation. (So in this sense it’s probably similar to the influence calculation Bill alluded to --- I mean similar in spirit, they might be measuring different things.)

I personally find Robert’s proposed measure quite natural, but of course it’s coarse grained (the proposed "reasonable" range of 1.5 to 3.5 is pretty wide). So it’s unlikely to resolve questions about details, as Knotwilg points out:

Knotwilg wrote: In order to evaluate the joseki, imagine white's P17 were at O16 instead. Would you feel white is better? Would robert's method yield different quantics?

There are various different kinds josekis; not all of them simply share territory versus influence. Therefore, josekis without one-sided other (dis)advantages can have a value range 1.5 to 3.5. Values outside this range must be compensated by a clear other advantage for the player having the disadvantageous value. E.g., a value around 4 to 4.5 can occur if a player makes much territory in the corner but the opponent prevents his easy access to the adjacent sides; there are a couple of such josekis.

Without a clear local explanation for a player's value advantage, a locally one-sided result can still be acceptable in the case of particularly fitting global circumstances for the opponent.

Robert: Thank you for the detailed reply, but you still haven't convinced me. In themselves the points you make are reasonable. Part of the problem is that you consistently change the other person's context to stress your own.

For example, yes, your system can be said to build up in blocks from a base. But it's still circular in that it depends on pro assessments via a feedback loop. If by some fluke pro thinking changed, you would have to change your system. There is nothing that I see in your system that would enable you to say to a pro, you are wrong because... I say "by some fluke", but if we consider the changes in go in the 20th century, we can say that there has been a gradual change in thinking in favour of central thickness, as players have got used to using it better. This has already changed the significance of many joseki evaluations. Current changes in go thinking seem likely to change evaluations further. I expect pros will rely on their own methods, not yours, to make them.

Secondly, yes, your system can be applied quickly on a single position even by a human, but this is practical only in leisurely home study. In a game context, if an opponent plays a new move, you have to look at several, maybe many, lines and evaluate each one independently. I don't believe your system can cope with that. Furthermore, pros do cope with this situation often and seem more often than not to come up with a good reply, so clearly they do have a system of their own that works.

The problem with applying numbers can be illustrated by looking at the stock market. Many numerical systems are touted - the Japanese started the ball rolling a very long time ago with their candlestick theory - but isn't it strange how all these tipster companies ask you to let them use your money instead of their own.

Essentially, complex things like go and stocks remain impervious to numbers and are best studied in a hermeneutic way. (Or best of all, get inside information - in go terms, pro advice). And in that context, the old chestnut from medicine applies, in which new students are told by the professor that half of all the "science" they will learn in the next few years will turn out to be wrong. He then adds: "The only problem is that we don't yet know which half."

For the creation of my method, I have used mainly dictionary / game database josekis as input to detect the existing joseki value types and calibrate the bounds ca. 1.5 and ca. 3.5. Now, the method does not need fresh, circular input. However, it is possible to view the bounds and the first move value as parameters, which can be re-evaluated from time to time. I think this is all that needs observation. If you want to include this process of re-evaluation in future, you can perceive some circularity but I perceive it as possibly repeated input with possibly adjusted parameters.

In a game context with various candidate joseki-like sequences, I can apply my method easily. It is possible that you can't do it yet; I recommend regular practice.

I dislike the stock market analogy because it is chaotical. Go is a perfect information game. This has a very different behaviour: the same positions always have the same best next moves (if one is able to understand which they are). In a stock market, the same initial situation can result in different continued market developments.

Sennahoj wrote:that's how Bantari communicates almost always

You think you know me?

As a matter of fact, my posts alternate freely between the scathing wit underlining thoughtless remarks from others and wise informative style presenting well though-out arguments on my part, virtually dripping with wisdom - if you only open your eyes wide enough to see it. Heh...

As for Robert, we do have a long history, and the hatchet not always can stay buried, no matter how I try. My failure, doubtless, so I concede this point to you.

But ultimately, I really have a very high respect for him and his work, regardless on how I need him now and then, and I think he knows it.

These days, the people I don't respect I don't talk to - and when you search through my posts you might notice a few prominent names I stopped responding to a long time ago, no matter what weird stuff they write.

Bottom line: I consdier RJ one of the most valuable poster on this forum, and MVP if you will, one of maybe 3 or 4 people I consider to be such MVPs. And I have said so both publicly and personally to him in the past, not only on this forum.

Hope this clears the situation for you.

John Fairbairn wrote: For example, yes, your system can be said to build up in blocks from a base. But it's still circular in that it depends on pro assessments via a feedback loop. If by some fluke pro thinking changed, you would have to change your system.

I think you're giving Robert both too little and too much credit.

The calculation he gives implies that for "joseki" positions the ratio of territory to "influence stones" is 1.5-3.5. Setting aside the ratio for a moment, he is looking at the relative efficiency of the player's stones. I've certainly seen pro commentaries indicating that efficiency is of paramount concern when pros evaluate a new sequence. He's also noting empirically that an "influence stone" is worth 2-3 points of cash territory. Again, it sounds plausible.

The method looks fine as far as it goes, but the problem is identifying those influence stones. That's largely a matter of opinion, at which point you have to ask why bother with the mathematics if it only substitutes opinions at one level with opinions at another?

I think the method seems most reasonable for strongish kyu players, who will have demonstrably inefficient stones that can be identified during a post-game analysis. It's just another tool in the kit, although whether it's a Swiss army knife or a specialty tool remains to be seen.

He's also noting empirically that an "influence stone" is worth 2-3 points of cash territory. Again, it sounds plausible.

The method looks fine as far as it goes, but the problem is identifying those influence stones. That's largely a matter of opinion,

It's more than plausible. It's been backed up by at least a couple of pros, although a little indirectly. Writers like Abe Yoshiteru have usually been concerned with measuring thickness alone in the opening or early middle game, and not just with josekis, and use methods they often describe as wielding a sword. In their calculations an extension from the wall is normally assumed (this is the sword sweep), and this typically heightens the value of each stone in the wall to closer to 4 (also increasing as the size of the wall increases, which seems reasonable). But so long as an extension stone is eventually possible, its absence can be factored in and the value of the wall reduced accordingly to more like 3 per stone. They also make remarks about which wall stones are valid: bumps, gaps and second-line stones are typically discounted. This method works also round corners. There are also refinements such as very big walls needing not just their "magic sword" extension but also a "magic dirk" extension (the analogy used is the samurai's two swords).

In other words, putting a numerical value on influence is not new. But I have never been able to establish that pros use it themselves. I half-suspect it's a device they've come up with only to teach amateurs. But then again, surely they must assess influence, and it may seem there's not much else to go on apart from a numerical estimate. The only alternative I've deduced is that they simply assess the efficiency of each play (a kind of tewari-lite), and FWIW I've seen far more evidence of that than for using numbers. However, if they do decide a move is inefficient, they quite often say it loses 1 point or 2 points, or whatever, so in that sense numbers come into it.

pwaldron, I have not repeated the full method here but only given hints that it includes more cases. A joseki belongs to one of the value types 'ideal equaly' (stone difference, territory count T and influence stone difference I are all 0; this triple indicates a joseki; dividing T by I is unnecessary and would not work), 'small values', 'ordinary sharing of territory and influence' (with ratio values from 1.5 to 3.5), 'territory disadvantage combined with other advantage' (two cases, of which one is a ratio <1.5, the other is of course I = 0), 'territory advantage combined with other disadvantage' (ratio >3.5).

Identified are those influence stones with significant outside influence. The outside is given clearly by perception. Stones with significant outside influence are, with a bit of experience of thinking about these details, reasonably easily identified.
- In the simplest case, they are the outside live stones.
- If influence is generated by thickness so that this generation is given by a shape of which influence-generating stones are not situated on the outside but have clear impact on the outside, then all such essential, non-superfluous stones are counted (consider, e.g., a thick small nadare shape)
- Ignore those influence stones whose influence is dominated by other same-colour influence stones. E.g., first and second line stones of a high wall.
- Ignore those influence stones being "covered" by strong opposing stones.
I will work out and publish this aspect of positional judgement more carefully later.

It can sometimes happen that one is not sure whether one wants to perceive a stone as "half" an influence stones. Round to either adjacent integer. The method tolerates such approximation. If the result is near to a bound of the standard ratio range, it does not matter because the bounds are circa-values. Of course, nevertheless one may look for compensating other strategic advantages.