So here's what I did, briefly, and I'll add more if there is interest:

First of all, I did read books such as Moonwalking with Einstein and others and I also hung out on mnemotechnics.org for a while. I learned the theory and practice of such techniques and spent a long time figuring out what would make sense for Go and what I could get out of it for time invested. I concluded that the main benefit that would be of interest to me is remembering games, both professional games and my own. And I'll be completely frank about the scenario I was envisioning. There are certain situations in one's life where one has to sit around and it's fine to sit quietly and say nothing, but is not considered socially acceptable pop out a go book or electronic device and start obviously clicking through games. These times are different for different people---it could be at church or at a boring meeting, for example. So I wanted the ability to think productively about Go during these times.

As has been mentioned, Go moves are not random. So at first I tried to create a system to remember sequences that have some Go meaning, such as haengma. But I gave up on that because upon examination, the moves that were hard to remember in pro games were tenukis. The timing of a kikashi, for example. So I reluctantly concluded I needed to be able to encode moves more precisely, via something like coordinates or very close to that.

The system I came up with was to divide the Go board into 4 quadrants and then assign each point a two-digit sequence. I then created a multi-component system described here. This would allow me to encode 4 Go moves in a single locus, so about 60 loci for a whole game.

I took the hit of losing the information about which quadrant the move was in, but that's actually resolvable if one reduces the desired moves per locus to 3. I'll explain the math later. In any case, remembering which quadrant a move is in once given the hint of the exact relative position to the nearest corner is not much of an issue for me anyway.

So, first of all, learning such a system is very hard. You have to come up with encodings that are meaningful for to you to remember. Then, once learned, the system has to be practiced or else you forget it. I wouldn't recommend a 3-digit system to anyone who isn't either homeless or planning to become that way. Of course, my system is 40% of the way there. More later as my kid is up. I did not go homeless.

So to follow up more...

In addition to the overhead of keeping one's system rehearsed, there as also the question of loci if you are using that system. You need a lot of loci and the journeys through these also need to be practiced regularly. Memory competitors will often spend several hours a day training nearing competition. I could not invest such time, so I had to take it a little bit at time.

One thing I learned is the value of imagination in remembering things. The other thing I gained is a better understanding of the variations in my performance depending on whether I was tired, angry, etc. This doesn't come across as well in Go as in memory sports, because in Go whether you win a game or not depends on your opponent as much as you. When using memory techniques, there is no lying to yourself about your progress. It was very enlightening to learn the effects of fatigue and other distractions.

These techniques are quite powerful, but just remember that what you learn has to be reviewed. Memory athletes in fact rely on forgetting. It's a blessing. If you didn't forget, you'd have to keep building new journeys (or palaces) for every practice session! As it is, most people can reuse their loci after a couple of weeks of disuse. But keeping them around isn't too hard because one can use something spaced learning patterns for that. But it does take time and effort.

I think the problems are that people give up before they get any benefits or they try something too ambitious and get discouraged.

Below is the example of the 2-digit encoding I use. I think this is good for go players because the 3-4 point, etc. appears in a canonical direction. If you can imagine a game with the first 4 moves all on 3-4 point, for example, those would be natural placements. So for example, now that one has a digit encoding, one can use any of a number of systems, such as Dominic System or Major System, to encode go moves.



If one has a 3-digit system, which I do not recommend for beginners, then one can do a numbering of the quadrants, and prepend these digits to the numbers above:



So one of the reasons I gave up on the three digit system, besides it being insane, is that well, it's overkill for Go by a big percentage if I wanted to use the same system for digits, which I do. For a 3-digit system, you have 1000 images when you only need 361: 36.1% efficient. In my system, 100 images but I only need 91, so 91% efficient. Well, I actually I have 400 image components, but that's another story and is not related to the waste a 3-digit system would have.

So I think I've disclosed everything that is needed. So I'm out of the closet as an amateur mnemonist on this forum. What a weight off my shoulders! Now I just have to come out as a go player on that other forum, and my karma will be clean. Well, cleaner

Thanks for the detailed posts, snorri, fascinating stuff. Some questions:

Are you able to successfully store and retrieve a large set of joseki's? How many, if I may ask? How do you deal with variations?
How do you decide which one to use? Are you still relying on vague ideas of the colour and distance of neighbouring stones or influence vs. territory? Or something more systematic, along the ideas of what Robert discusses?

One thing about the "meaning of the moves" issue-a well-known Korean 7d amateur spent some time at our Go club and spent an hour or so teaching a 3k before the club meetings. Many times I overheard them and she was teaching him josekis or other moves saying "just play here" and he was continually saying "I don't understand the meaning/purpose of this move". It became clear to me over hearing this many many times over the course of a few weeks that she regarded the meaning or purpose of a move as surplus information not necessary to play Go really well, and she would definitely lean in favour of the "just remember the joseki moves by whatever means" side of this argument IMO.

I haven't tried using this for joseki, just for pro games which are linear and for which the method of loci works better. But I have thought about the variation problem. One way to solve it would be via linking, which is a different technique than the method of loci. The problem with linking is that it's fragile---there is a significant risk of losing an entire branch. This would be a problem if the variations were truly meaningless, but in go I'd expect to be able to use my general knowledge of shape to recover, so maybe it's doable with practice.

As for volume, let's set some expectations. The grandmaster norm for digits is memorizing 1000 in an hour. In 1993 the world record was 900. Now the world record is 2660. So let's say 500 moves in one of these training runs. Probably keep around 10 journeys of this size for training, but if one ditches training and just uses for review of joseki encodings, 5000 moves. Kogo's had about 62,000 moves the last time I checked. I think 5000 moves would be in the ballpark for a grandmaster who knows nothing about go. I am not a grandmaster and I don't know how long it will take to meet that norm. I was probably 30% of the way there at my peak, but my journeys are kind of rusty.

So the conclusion is that I wouldn't do it this way. If I wanted to learn a lot of joseki, I would do it mostly the traditional way and only use mnemotechnics in tough spots.



Well since I don't memorize joseki using mnemotechnics and would only use them partially if I did, I'm sorry to disappoint, but I don't think it would make me think differently about the game. The moves I make are influenced by the same ideas from pros and game experience as anyone else's. Would I train up on some crazy joseki variations? Maybe... But the difficulty there is not just about the number of variations in the books, but in how to handle complicated fights in general. Reading ability and shape knowledge tend to dominate there. In real amateur games those fights will get out of the book pretty quickly. But you know, I might try this. The thread is kind of inspiring...

When I memorize pro games, I associate the move counter with the move. I hadn't realized I was doing this until I read this thread. So I can look at a stone on the board and say, "this was move 60 (or whatever)". If I lose track of of the move number, I have a hard time remembering where the next move is played. I also have trouble starting the game, but once past the first 8-10 moves, I can usually remember the rest.

I'm terribly sorry for misreading you, or at least, not concentrating, and just replacing your "pro games" with my assumptions of "joseki". Anyway, your descriptions are indeed very interesting and rather inspiring.

Some more questions:
- you never did elaborate on the maths of how 4 moves became 3
- am i to understand that 5000 moves is a realistic limit for this encoding system + your memory? Using 250 moves per games suggests that 20 games would be the limit of your memory? Is that correct? While not bad, it seems a bit less than what one would hope to achieve. I would have thought a strong player could remember most of the moves of many pro games without any special techniques, although as you say, the problems mostly being around tenuki/kikashi, and probably end game. Did you consider some sort of hybrid of coordinates along with local joseki sequences?

I took a weekly class for a while with a strong pro from Korea. Homework each week was memorizing a pro game we'd review in class. In one review, I point to a move up on the demo board ask a question and said "If move 130 were one row down ... " or something like that and the pro and the stronger players in the class gave me quite a quizzical look. Turned out I was memorizing the moves attached to the move numbers, speaking the numbers like a metronome when playing over the game and also associating key shifts from one part of the board to another with some of the numbers. I knew they moved from lower left to right center on move 74 ... helped me not forget any less memorable moves along the way. All the strong players used the numbers to find moves on the game diagram, but then just memorized the moves. My approach bemused them rather.

You'll notice that in my first diagram there are duplicates of the 2-digit encodings. If I memorize moves and discard quadrant information, I can use 2-digits per move. Since I can store 8 digits per locus, that's 4 moves per locus.

But if I really care about keeping the quadrant information, then I could sacrifice two of the 8 digits in the following way:

relative move 1 (2 digits) + relative move 2 (2 digits) + relative move 3 (2 digits) + (6 bits of information encoding quadrants for moves 1-3) (2 digits)

This is because the quadrant information for 3 moves is 6 bits which can fit in the range 0-63 and therefore can be represented as a two-digit decimal number.

I don't like that much, though, because I like my persons to always be black. (Wait, that didn't come out right. Anyway, the parity shift of an odd number of components would bug me.

Another option is to use a smaller, PAO system, and two moves per locus with complete information. You only have 16 possible actions. I don't like that as much because the actions would repeat a lot. Consider how many times in a game a white move is in the same quadrant as the previous black move! I studied this. It's a lot. Excessive repetition is a source of errors when using memory techniques, and most memory athletes will go to some length to avoid them.



It's an estimate, but I'm not saying one couldn't do more with more dedicated practice or by starting younger, etc. or just being crazier. Take a look at this interview with Dr. Yip Swee Chooi who has memorized a huge Chinese-English dictionary.

5000 moves doesn't sound like much, but ask a pro to memorize twenty 250-move games with completely random moves and see how that goes. I treat this as a case of overlearning. There is a coloring to the moves and sequences that gives them a personality. Why do pros encourage you to play over games on a real board? It adds the sense of sound and touch. It's good for the memory to just pile on additional input that may not be pertinent to the thing that is being memorized. That's just the way the brain works. It is not like a hard disk.

Efficiencies for encoding josekis are possible, but most of the game is not joseki, so you really don't save that much. I evaluated a hybrid haengma-coordinate approach, but it suffers from a kind of Amdahl's law. The non-optimized moves become the bottleneck.

Opinions vary on this, but it's well known warning. 33 for me is Michael Caine, who can do whatever wants without getting me into therapy.

5000 is a massive amount. It does just seem a bit of a shame to spend them equally on "normal" moves and the peculiar moves. But I take your point about a hybrid being hard to get right.

Yeah, the "meaning of the moves" (or worse, the "meaning of each move") always sounds a little bizarre to me. Like something out of that old Kung Fu series. "Snatch the pebble from my hand" stuff, it seems. Because you're supposed to study this stuff that no one really seems to know.

Some moves have a clear meaning, or, if they are very good, multiple meanings. Others have so many follow-ups the meaning is not clear until the results of an enormous number of subsequent variations are analyzed. Joseki books, for example, don't tell you the meaning of the moves. Rather, they give evaluations of results. If some new variation is discovered that can be forced, a joseki may be abandoned in a branch pretty far back from the refutation. This happens all the time. Do we then suggest that professionals didn't understand the meaning of the previous move before? If you answer no, I'd reply that the meaning has limited use. If you say yes, then maybe the meaning is inaccessible to amateurs.

It's almost misleading, too. I can take any 20k game and make up a "meaning" for most moves. (We'll leave out: take sente by pointless self-atari.) That doesn't make them strong or easy to remember.

Can't get much more vivid than Michael Caine. It was detriments like 'im that built this bloody empire!

Hats on!

I don't understand he point of all these complicated encoding algorithms. The purpose of memorizing pro games is too get a feeling for shape. If all you remember is " bob triangle fish" clearly that isn't happening.

Now, I`m really trying not to be negative but could it be that all this memory stuff is barking up the wrong tree?

Isn`t the point of studying the meaning of a move not to remember the move, but to be able to play that move or others like it when the situation calls for it? For example, if Onimoto Sousuke, 11 dan, says "this move b is a situational joseki. Normally you lose five points compared with the alternative at a, but you might choose b to function as a ladder break or to form a large-scale formation with your 5-4 point in the opposite corner", then it`s not so much the move itself you`re learning from as the reasoning behind it.

Again, there could be applications for mnemonics, for example, I use "Di, Dick and Ed" to remember "Descent - Ikken tobi; Diagonal - Keima; Empty Triangle - Diagonal" as a way of remembering ways to play a supported 3-3 invasion against 4-4 and keima enclosure, but generally I`ve found it`s just easier to remember the reasons for choosing each move.

I think the problem with just memorising stuff is that every position on the board affects everything else to a greater or lesser extent, and so even a perfectly memorised sequence could be a poor choice or could be profitably altered in some way according to the needs of the situation. If you`ve studied the thinking behind the moves, then you`re going to be able to act in a much more flexible way than someone who`s simply retrieving items from a memory palace.

Put it another way: "use complicated mnemonics!" or "think carefully about why you play x in move y situation!". Which piece of advice would you bet on Lee Chang Ho giving you, if one of the two?

_________________
Learn the "tea-stealing" tesuji! Cho Chikun demonstrates here:

Have you studied a good or a bad set of possible meanings? With bad meanings, it is easy to conclude that they would be bizarre. Use a good set and structure of meanings!



Many meanings are basic and many people know them somehow. A problem is to be inaccurate, have important gaps and lack structure in the knowledge of meanings. Studying material on meanings gives you a reasonably complete related knowledge, regardless of whether other players fail to do related study.



In such cases, dynamic meanings apply: meanings referring to strategic or tactical choices. Decisions such as "Giving the opponent the strategic choice to achieve either X or Y.". If the player dislikes at least one of the reasonably possible opposing choices (X or Y), then the player must NOT choose a move with such a meaning of giving the opponent the related choice.



This is, of course, wrong: my joseki books tell you the meanings and strategic choices, even if they are dynamic.



Usually, they don't, or, if they do, the "evaluations" are very weak. "Black is better." Occasionally, comments are bit more informative: "Black has 12 points in the corner and White has a wall facing the upper side." It is a shame! All joseki books should give evaluations of stone difference, territory, influence, strategic choices etc. for each joseki and explain possible meanings carefully, quite like my books do.



It is better to understand limited knowledge before the value of new moves is fully understood. If variations are complicated, then knowledge must admit that it is partial or makes an assumption such as "no unexpected refutations have been discovered thus far".



Meanings must be relevant for achieving a result helping the player making the move.

Possibly, which is why I'm inquiring. However, some of the examples discussed so far seem legitimate:

- a checklist
- remembering pro games to permit their study in conditions not conducive to using a board
- remembering joseki sequences, although this one is a little unclear, as the relationship between the sequence and the result is murky at this point

I don't think anyone is presenting this as an either/or thing. At least I hope I'm not coming off that way. Memory has gotten kind of bad rap in Western culture over the past few centuries. Because of this, I think the pendulum has swung a a bit too far in the extreme direction of insisting that understanding precedes remembering all the time, or that remembering things just for the sake of remembering them is somehow bad. Taken to the extreme, that would of course be a silly proposition. No one would say, "don't remember the words in a foreign language. Instead, just try to understand the meanings of the phonemes and their flow." It is not 100% wrong---sensing the cadence of a new language does help in acquisition---but eventually one has to learn something.

There are a number of ways of remembering things better. One is to understand them. That's often the best if it is possible. I'm not arguing that point. Another is brute force repetition. Another option---the one being discussed here---is memory techniques. There is no one correct solution for each situation.

A few more:

- the rules for capturing races, to save time during game play (I'm not saying one shouldn't try to read, but when counting is faster or more accurate, why not?)
- the faces and names of people you meet at go events, to make new friends
- some words in an Asian language, to read go books
- some poetry, to calm your mind or share
- the sizes of endgame moves, to win close games