Tsumego: Easy or hard?

[dfan]

In real life it's not necessary to read out a situation completely. You only need to find the best next move.

This feels to me like such a weird thing to say that I wonder if I am misunderstanding you.

Locally, if I know I am going to play in this area, all I have to do is find the best move and play it.

But to decide what area to play in, I need to know the status of every area. It matters a great deal whether the best local move kills a group, or makes it fight a ko for life, or is just a sente endgame sequence, or is a lot of points in gote.

[John Fairbairn]

I don't find the problem weird. Or, if you do insist on calling it that, once you've been round the block a few times you have to accept that weird positions come up in games all the time. Either way, studying them is practical. It may not be the most efficient use of your time if you have gaps in even more basic knowledge, but these weird positions should still be on your list to work though eventually.

But there is a more fundamental question to address. Are you studying tsumego for entertainment or to become a pro? Or as an in-betweener? As the latter, you are prepared to study hard to see how far you can get, but you don't really expect to reach pro level. Or are you a rank beginner?

Each category merits its own type of tsumego collection. Beginners get fed a diet of very easy problems, always practical, and the only "entertainment" provided is a pat on the head or some other warm, fuzzy feedback. The equivalent of cheese for rats in mazes.

By far the biggest category is for in-betweeners. The usual serving is a large dollop of practical problems but with a generous helping of sauce made up of a cocktail of "weird" classical problems such as under-the-stones gems. The psychology seems to be that studying the practical problems is hard work, so we need to have some relaxation. Seeing the crown jewels happens also to have the bonus of their beauty providing inspiration and motivation. This category is not just the largest but is also by far the most popular. Various such collections are set at various grades of course, so there is something for everyone.

The category of tsumego collections for entertainment only is rather small. Quite a lot of what is available does offer "weird" problems and they tend to be rather hard. It is actually the fact that such problems are outliers that makes them interesting to many people. Some composers try to entertain in quite different ways, of course, but entertainment is still at the heart of the collection. For example, Hashimoto Utaro produced a famous collection in which he composed a problem for each path on Basho's Journey to the Far North. There wasn't much connection between Basho, his poems or go (though he did stop off at one place named after go stones), but it's easy to imagine Hashimoto's readers recalling Basho's haiku at each point, taking a sip of sake as they ponder whether depth-first is best, sip, or is width-first best, sip, or is there a clue in the name of the place visited, hic.

As far as I can recall, that type of book is almost non-existent in English, and the only example I can think of, sip, is Gateway To All Marvels where I explain in depth the names of all the problems. But it's huge, and if you try to sip your way through that too quickly, you may end up needing rehab.

The final category of books for becoming a pro the "pure" ten million reps in the gym way is small but does exist, though mostly at lower levels. Some Korean series in particular fall into this category. But they don't really take you very far - at least not as far as all your hard work leads you to expect. What happens beyond these books in real pro life is that you have to change your whole modus operandi. Typically you become a member of a sort of clique. At pro level, the most common - maybe the only way - way of studying tsumego is to compose problems and to share them with fellow pros. You take your latest offering to the next study group session and challenge your pro mates (in the old days teachers would challenge their pupils). Nowadays you don't even have to wait for the next session - you just pop your creation into your smartphone app and "send to all". Recall, for example, that Cho U used to send his compositions to Kobayashi Izumi when courting her. This has even been made into a major Hollywood film called Romancing the Stone.

The point about composing your own problems is that is how you really learn to "frame" them. Take any two (or three) themes at random and see whether you can combine them into an organic whole. That way you don't think in terms of reading moves out. You just do the problem like a jigsaw puzzle - fitting the pieces, i.e. themes, together. Of course you can throw in lots of blue sky or make the joins deceptively similar, but still you are "chunking" more than reading.

Not every pro composes problems, though those who don't tend to stay near the bottom of the pecking order. And of course being good at jigsaws doesn't automatically make you good at Lego and is probably no help at all in learning to play the banjo. So some pros prefer to study josekis, fusekis, tesujis or whatever. Each to his own, and amateurs make similar choices.

But if tsumego is your shtick, wherever you lie on the spectrum you really do first have to decide which category you want to belong to, and then adapt your study to that category.

Exactly how you study within each category is an open, and possibly entirely subjective, question. But if you fancy yourself as part of the "wannabe pro" or "as good as I can get" sets, it seems to me that there is enough information out there already (from games other than go and from activities other than games) to make some generalisations. I would say that the most important traps to avoid are to stare (or glare) at a problem or to pretend to yourself that you can read out variations to infinity. These are deceptive traps because quite often with such effort you can successfully do a problem. But what you are practising is not really solving (i.e. understanding or analysing) a tsumego position. You are practising concentration, self discipline or masochism. All, except perhaps the latter, very useful even in real life, but it's still like analysing a rock sample by glaring at it and smashing it with a hammer instead of using reagents and a chromatograph. The elements (or chunks) that make it up need to be identified. In go, it seems to me, both from simple observations and reading about what existing pros do, composing thematic problems is the way be sure you are identifying the graph peaks properly. That's the way to achieve warp speed in analysis. It may even prove to be the way to achieve DeepTsume.

[gowan]

Another movie (and novel) with an allusion to Go, is Cutting for Stone

[iopq]

In real life it's not necessary to read out a situation completely. You only need to find the best next move.

This feels to me like such a weird thing to say that I wonder if I am misunderstanding you.

Locally, if I know I am going to play in this area, all I have to do is find the best move and play it.

But to decide what area to play in, I need to know the status of every area. It matters a great deal whether the best local move kills a group, or makes it fight a ko for life, or is just a sente endgame sequence, or is a lot of points in gote.

Yes, sometimes that's useful, but in a tsumego context I need to know if I'm eating my time or not. Often I'll spend way too long searching for a miracle solution. In a game I have a time limit so I'll just check if I can play a kō or not. How often do I need a perfect answer? Once every thousand games?

I'm actually very interested in the "beginner to online player" amount of learning. There's a gap between "knows the rules" and "can legitimately have fun playing against other beginners online and get a ranking"

So basically a collection of all the basic shapes, tesuji, endgame, life and death, opening on a small board.

But how do people learn go? Is it enough to memorize as many of these as possible? Let's say you simply use a spaced repetition approach. You solve for like a second, then memorize the solution. So on and so forth, for several thousand of these.

Check by giving a timed test. Schedule reviews on failed problems. Does this make a 10k player from a beginner?

[Bill Spight]

I'm actually very interested in the "beginner to online player" amount of learning. There's a gap between "knows the rules" and "can legitimately have fun playing against other beginners online and get a ranking"

So basically a collection of all the basic shapes, tesuji, endgame, life and death, opening on a small board.

But how do people learn go? Is it enough to memorize as many of these as possible?

No, because you also have to learn the similar positions where the plays you have memorized do not work. Otherwise you will make mistakes in such positions. Relying solely upon memory and recognition is brittle. That's why you have the advice not to learn (i.e., memorize) joseki. When your opponent makes a mistake, what do you do? What do you do when the joseki you have memorized is not right for the occasion?

You can learn such positions by being thorough and taking your time when you study. You can also learn a lot by playing with positions, as Nigel Davies says.

[iopq]

I'm actually very interested in the "beginner to online player" amount of learning. There's a gap between "knows the rules" and "can legitimately have fun playing against other beginners online and get a ranking"

So basically a collection of all the basic shapes, tesuji, endgame, life and death, opening on a small board.

But how do people learn go? Is it enough to memorize as many of these as possible?

No, because you also have to learn the similar positions where the plays you have memorized do not work. Otherwise you will make mistakes in such positions. Relying solely upon memory and recognition is brittle. That's why you have the advice not to learn (i.e., memorize) joseki. When your opponent makes a mistake, what do you do? What do you do when the joseki you have memorized is not right for the occasion?

You can learn such positions by being thorough and taking your time when you study. You can also learn a lot by playing with positions, as Nigel Davies says.

But who said the training data doesn't have those positions already? Or who said they actually come up enough? For example, a player who misses under the stones tesuji can be 1d, a player who can see them but misses nets probably has difficulty breaking into SDK.

But let's take another view, doesn't a beginner miss a lot of important tesuji that experienced players can see anyway? What's the difference if it's because of not memorizing them or because of not seeing them enough?

I see zero downside of memorizing the exact moves of the tombstone tesuji. If you know them by heart, you can see if the resulting shape will win the semeai or not. This is not the same as memorizing joseki. All of the moves are forced. If the opponent goes off the move order you just capture their stones that are in atari.

[Bill Spight]

I'm actually very interested in the "beginner to online player" amount of learning. There's a gap between "knows the rules" and "can legitimately have fun playing against other beginners online and get a ranking"

So basically a collection of all the basic shapes, tesuji, endgame, life and death, opening on a small board.

But how do people learn go? Is it enough to memorize as many of these as possible?

No, because you also have to learn the similar positions where the plays you have memorized do not work. Otherwise you will make mistakes in such positions. Relying solely upon memory and recognition is brittle. That's why you have the advice not to learn (i.e., memorize) joseki. When your opponent makes a mistake, what do you do? What do you do when the joseki you have memorized is not right for the occasion?

You can learn such positions by being thorough and taking your time when you study. You can also learn a lot by playing with positions, as Nigel Davies says.

But who said the training data doesn't have those positions already?

Current collections do not include them, as they are of little interest. E. g., Black to play and White lives anyway.

Or who said they actually come up enough?

I imagine that non-tesuji positions are more common than tesuji positions.

I see zero downside of memorizing the exact moves of the tombstone tesuji.

The problems with memorization come from what you do not memorize and from overspecialization. You have to be able to handle novel situations, and you should not neglect any go skills.

[John Fairbairn]

I see zero downside of memorizing the exact moves of the tombstone tesuji.

OK, you've memorised the moves of the tombstone tesuji. Do this easy tombstone problem. Black to play.



Prince Hamlet dressed in black holding the white skull. I suspect poor Yorick didn't think much of being remembered.

[EdLee]

major Hollywood film called Romancing the Stone.

[SoDesuNe]

But let's take another view, doesn't a beginner miss a lot of important tesuji that experienced players can see anyway? What's the difference if it's because of not memorizing them or because of not seeing them enough?

I guess the difference is the same with learning stuff in general.

For the beginner the problem is often not missing a tesuji per se, but missing the problem's problem. Ie tomb-stone tesuji is involved in a liberty race, if you can't figure out the liberty race or maybe you don't even see a liberty race happening, the knowledge of the sequence of the tomb-stone tesuji is of not much use.

That's why it is better to learn how to tackle problems, learning how to figure stuff out instead of learning solutions to problems. In go that means (on a basic level) being able to count liberties, visualise stones in your head (read) and knowing defects in shapes (candidate moves).

A really advanced idea, concerning liberty races, is presented in the last game of AlphaGo vs Yi Se-tol. AlphaGo "botched" that liberty race because he sacrificed stones he could have saved. But it wasn't at all about winning that liberty race but judging that gaining sente moves on the outside is better for the overall game. My (old) LZ still agreess with this decision (over 300k playouts).

Memorisation will never give you this idea.

[iopq]

Current collections do not include them, as they are of little interest. E. g., Black to play and White lives anyway.

Those positions exist as "is this group alive? If not, play to kill" type problems

When you say alive, you switch colors and defend an attempt to kill

[Bill Spight]

Current collections do not include them, as they are of little interest. E. g., Black to play and White lives anyway.

Those positions exist as "is this group alive? If not, play to kill" type problems

When you say alive, you switch colors and defend an attempt to kill

Sure, status problems exist, but I am talking about specific positions where a tesuji or line of play would work with a very slight change. The point being that it is easy to confuse the two. Memorization of moves is not enough. You have to be able to distinguish positions from each other.

[iopq]

Current collections do not include them, as they are of little interest. E. g., Black to play and White lives anyway.

Those positions exist as "is this group alive? If not, play to kill" type problems

When you say alive, you switch colors and defend an attempt to kill

Sure, status problems exist, but I am talking about specific positions where a tesuji or line of play would work with a very slight change. The point being that it is easy to confuse the two. Memorization of moves is not enough. You have to be able to distinguish positions from each other.

You can make both of those different problems.

You can't cover everything, but you can't cover everything anyway. Even pros try tesuji that don't work.

[Bill Spight]

You can make both of those different problems.

That's why, when people make flash cards, or make a set of problem positions for computerized review, I advise that they also make cards with slightly different positions where the play does not work. That way they won't end up recognizing the problem position based on only some of its features. Doing so can lead to mistakes in actual games. I have seen that happen.