Lately, I've been doing go problems pretty regularly, and I've started to think about categorizing the types of moves in the hope that I might be able to narrow my focus quicker. Here are some of my initial observations:

For simplicity's sake, let's just look at Black to Kill type problems. They invariably end with what I call a "fatal blow," and there aren't many different types. Most fall into these three categories.

1. a killing shape is formed inside the group
2. a killing group, or a stone making a false eye links up to the outside
3. a killing group creates a shortage of liberties that prevents white from approaching


My thinking is that it might be helpful at the onset to plan for one of these fatal blows. How to get there is the next question. The moves up to this point must in some way threaten to lead to one of the three blows. Again, there seem to be a limited number of types of threats. A move can:

a. reduce eye space
b. threaten to connect out
c. threaten to prevent a second eye inside
d. combine threats

From this, it would seem that go problems can't be all that hard, but as we all know, it ain't so. Most threats can be answered in several ways, and it is not easy to keep track of every consequence. Nonetheless, I'd like to hear what you think of this type of approach.

Edit: modified point 3 for clarity. Although the shortage of liberties often comes about due to strings being separated, the separation itself is not necessarily the fatal blow.

_________________
Patience, grasshopper.

I think that it is a promising approach. I suggested a similar idea to Nam Chi-hyung a few years ago, and her initial response was positive. I have seen other groupings of go problems based on where the first move is played and on the key tesuji.

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

Visualize whirled peas.

Everything with love. Stay safe.

I've seen a couple of approaches to this in Korean and Chinese books, but I believe the Japanese approach, which I've mentioned before, works best.

Almost all problems can be divided into three categories (easy, medium, hard), depending on whether they involve one techniques, two techniques or three or more techniques.

The techniques overlap to some degree but the final matador's kill is always one of a relatively limited set and these are considered to be cases where you can just look and play the killing move right between the shoulders instantly (e.g. playing on the vital point of a nakade shape; other types include throw-in to make the final eye false).

The point is that you must learn this set absolutely pat, with instant reactions and essentially no reading and no need to take the opponent's response into account. Japanese magazines very often publish sets of these problems. FWIW I can do most instantly but for quite a few I need to pause and think. There is a deal of genuine work to be done to get these all off pat.

But if you can do that, when you move on to the next category (two techniques), by definition the final technique must be one of those you have already learnt, so you only have to learn the second-tier techniques (which for the most part are the same as tier one, but there are some new ones - preparatory moves mostly). Learning this set does, however, benefit also from doing enough problems so that you get a feel for which combinations of techniques crop up most (e.g. hane then placement), and the other important point is that at this level the opponent's responses - ALL of them - have to be taken into account - no wishful thinking allowed!

One you learn these two-technique problems thoroughly, the same observation applies about three-technique problems: by definition the final two techniques must be exactly those you have already learnt. Furthermore, there are virtually no new elements in the third techniques. That means the learning load is essentially getting a feel for which three-way combinations occur most often, and the reading load is extended to the opponent having an extra set of responses to account for. But reading by this time becomes a matter of recalling chunks that you already know from having assiduously done the earlier types.

This is really all rather obvious, but I have never come across an amateur who has been prepared to put in the work of learning all the one-technique types as well as is frankly necessary. The dazzle of the Igo Hatsuyoron and the like is always too enticing.

We are currently working on a novel approach to Igo Hatsuyôron 120:

viewtopic.php?p=182312#p182312

So wait, and see ...

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

One Thousand and One Life-and-Death Problems seems structured this way. It is divided into three sections: 400 one-move problems, 300 three-move problems and 301 five-move problems. It's not always obvious to me why "one-move problems" are labeled as such - the provided solutions for them only give one move but sometimes I need to read a bit further to see why the solution works.

I'd rather see problems categorized into types of technique rather than easy, hard or number of moves.

The more difficult a problem is, the more hidden is the problem's structure.

"Hidden" means that the problem's vital points (and so the techniques to be used) are not obviuos in the very beginning.

This means that some preparatory work is needed before the vital point / the vital technique appears in mind.

So, John is right with his description of up to three successive techniques to solve a problem. You have to master techniques one, and two, before you can use technique three.

Knowing technique three alone does not help, when you are unable to reach the "point of final decision".

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

That's the point. You need (as I do, too) to do probably several hundred of these problems so that they become obvious at a glance, and we really need to have the self discipline to do that before going on to harder problems . It would be interesting to have an estimate of how many such problems we need to do. I'm certain it's at least in the hundreds, but at a guess I'd expect a level of maybe ten thousand (maybe including repetitions) to capture all the techniques for Level 1.



That approach has appealed to me and I spent a lot of time doing it for Gateway to All Marvels, but even so I think it's inferior (if you want to improve quickly, as opposed to enjoy a problem as a connoisseur) to the above-mentioned three-level approach.

In GTAM I identified about 50 techniques, I think. This did not include a few really basic items, and obviously not every technique appears in GTAM anyway, but I do believe nevertheless that the number of identifiable techniques is genuinely on the low side, so that a really concerted and controlled effort to master tsumego problems to a level such as 5-dan is straightforward, though of course not easy because of the time involved,

Categorising go problems can be done in different ways. E.g., by aims, reading methods, structure of reading, (un)availability of techniques.

daal, there are many more classes of techniques than you dream of. You know where you can find a lot. My most modest estimate is that there are hundreds of techniques, however, I really fear that the count goes into the thousands. There are, OC, a couple of frequent classes of techniques, such as fundamentals, eyespace, liberty shortage.

John, when you say that all of the opponent's responses had to be taken into account, you get much closer to the truth of problem solving than with your count of numbers of involved techniques. However, simplifications exist that sometimes greatly prune from "all responses" to "at least one response". Simplifications are not always applicable, but whenever they are applicable one should apply them in order to reduce the reading volume to a healthy size.

I have to disappoint you about your third description of one, two or three techniques to classify a difficulty of problems. After having studied all relevant variations and decision-making of 100 problems, my conclusion is: In ca. 2/3 of all practically occurring problems, techniques are inapplicable or play only a minor role while the major role is reading and pruning methods for simplifying reading.

At least, there is a relation between your / the traditional Asian approach and the reality of raw reading: Prior knowledge helps because, whenever a sequence reaches a known shape with a know result, reading can be pruned and this knowledge can be applied when proceeding with the reading in other variations / sequences.

Although there are stable nakade shapes, a few standard joseki follow-ups and standard LD shapes everybody should know, most sequences do not do us the favour to result in anything else than nakade, and these are useful most in LD problems but hardly in other problems.

Techniques are useful but basic reading theory is at least four times as important for solving problems. Therefore, the most interesting classification of problems is according to the kinds of necessary reading. Unfortunately, one does not know until one has solved a problem and studied carefully the nature of the reading inherent in the solution.

For too many years, I fell into the same trap of thinking that I could improve my reading mainly by studying techniques or combinations of techniques applied to problems. Luckily, now I have discovered that reading theory is by far the most important.

The consequence? Effort. Reading requires effort, regardless of whether simplifications apply. Even with all available simplifications applied, the remaining reading task still requires effort. Whenever there is no short-cut to reading all (except obvious failures) next moves, one is required to read all of them. With iteration that can sometimes mean having to read dozens (in very difficult problems: hundreds) of variations. Without disciplined reading, we cannot solve such problems, which defy useful application of techniques even in retrospect.

Robert

Since I have demonstrated that ~100% of the practical problems in GTAM can be analysed into techniques, and your estimate is one third, we clearly are using different concepts. Your other remarks seem to confirm that to me also.

I don't deny reading has to supplement (dynamic) pattern recognition, but the proportions in real life seem also to be vastly different from your ideas. Pattern recognition dominates. I base this on various things. For example, analogies with the way people learn to read words - we use shape recognition of almost an entire text in whole words and only decipher new or mispelt words - or music scores. I think it also explains the blind spots even pros have. It explains better the astonishing speed at which pros do complex problems at the first pass (reading is then used only as a check). It even explains the choice of Japanese word yomi for reading (as opposed to calculation).

I forgot to mention that we do have a perfect example of the Japanese pro method in Today We Have a Splendid Feast by the Meijin Inseki. As ever, I'm inclined to pay more attention to a Meijin who could produce also Igo Hatsuyoron than to an amateur. This book shows that some reading is required, but it also shows that the proportion is small, with the right guidance.

You mean like this?



I wonder at what stage of language development this becomes easy to read, would a child learning to read be able to read it or does it come later with higher proficiency. Robert, and other non-native English speakers, how easily can you read the above?

There is no doubt that each problem can be analysed into techniques. This is so because each move can be associated with techniques. However, in most cases, such a decomposition is not useful but a superfluous, unnecessary, extra burden.

Techniques are only useful in the sense of efficient if they REDUCE reading amount. Techniques are counter-productive if they INCREASE thinking amount without reducing reading amount.

If - and only if - techniques significantly reduce reading amount, they are useful for reading / problem solving. This is so only in a moderate fraction of practically occurring problems. In another moderate fraction, techniques help not to overlook uncommon (sometimes: tesuji) moves in branch variations; such do not remove the task to atually read the other relevant variations.

Pattern recognition cannot dominate in practical go reading because it does not meet frequencies in reading. Pattern recognition combined with resulting status recognition is useful in terminal positions (leaves) or in positions reverting to already read positions. Both together are a minority of the reading effort, which consists of creating and parsing the game tree in order to REACH the terminal positions, recognisable patterns (known shapes) and reversing positions. This is so despite following unbranched sequences of successive obvious moves.

Analogies from non-go topics are irrelevant for how a go game reading tree evolves.

Yes, pros can read fast and often seemingly at a glance. However, this does not mean that they would not be doing the actual reading. (And it also does not exclude the possibility that some pros might make a too thin reading to achieve the correct result in a great fraction of problems, but not in 100%.)

Reading is independent of meijin versus amateur status. Not social status decides reading, but the skill at reading is decisive for reading well. The skill of explaining reading is decisive for explaining reading well.

The problem with classic (and most modern) problem collections is: their problems are artificial and designed around showing techniques or combinations of techniques. OC, as a consequence, the ability to solve artificial problems focusses on techniques. Practically occurring problems are, on average, different because they are not clean. They do not tend to be focussed just around a few techniques. Instead they tend to require the standard: plain reading.

No problem reading it. Maybe 99% of my normal English reading speed. I don't have a stopwatch handy ATM, maybe later...

Versus this?



Edit: "detritus" is a bowdlerization of S-h-i-t.

Or this?





I learned to read using phonics, and that worked well, despite the vagaries of English spelling. However, teaching kids to read by recognizing whole words caught on in the US, and I suppose, in the UK. Apparently the learning curve is pretty steep at first. In addition, parents were not able to help their kids learn to read by the whole word method. That's why we started seeing ads on TV for aids to teaching phonics. I don't see those ads much these days. Maybe phonics made a comeback in the schools. Or maybe parents who were taught by the whole word method were happy with their kids learning that way.

BTW, the shortened version illustrates the fact that most of the meaning of text is carried by the consonants.

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

Visualize whirled peas.

Everything with love. Stay safe.

I thought we'd already reached the height of amateur arrogance with the contumely for Japanese 9-dans. I see now there was scope for more - and not just any meijin but the Meijin Inseki!

Is this a subtle form of "there's no such thing as bad publicity"?

If we replace "reading" by "brute force calculation of variations", research on chess indicates little difference between master and patzer in that regard. I doubt if that is the case with go pros vs. amateurs, because of brain sculpting as kids. Still, reading is much more than calculation.

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

Visualize whirled peas.

Everything with love. Stay safe.

Dear Robert,

Pattern regognition is a means to easily truncate the variation tree. Thus, it minimises the amount of reading that will be necessary in addition.

Usually, you will easier find your treasure, when you know what you are requested to look for.

"There might be two Nakade shapes in different orientations." gives you a hint to start your considerations with two potential vital points.

"There will arise a line of false eyes, if I play there." might be a useful insight to begin with.

+ + + + + + + + + +

Reading can stop at a moment, when I securely know about the rest.

E.g. there is absolutely no need to fill-in two of four one-point eyes of a group to prove that this group is an independently alive two-eyes-formation.

+ + + + + + + + + +

As a matter of course there will be single examples, where solely trusting "intuition" (i.e. pattern recognition, in my opinion) does not lead you to the correct solution of a problem.

Inoue Dosetsu Inseki apparently liked a lot to integrate such pitfalls for his professioanal collegues in Igo Hatsuyôron 120.

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

Instead of meta-discussion or reference to proprietary file formats, show us Inseki's skill of explaining how to solve a problem well so that we can praise or criticise it properly.

Yes, and for the context see my earlier messages.

Show us RJ's skill of explaining how to solve a problem well so that we can praise or criticise it properly.