Finally I've gathered some time to write my long overdue post about spaced repetition and tsumego.
If you've been following my Week by the numbers blog, you may have seen that for the past 30 days I've been using Anki, an spaced repetition program, to go through the tsumego in Cho-1-Elementary.
Before getting into the details, first some background on Anki, SRS, tsumego et al.
Anki is a SRS program. This raises the question what is spaced repetition? Explained quickly:
- The best way to remember A is to revisit A very often. If you do it daily, quite likely you won't forget, ever.
- But the best moment to revisit is when you are about to forget A, this way take full advantage of your recall power.
- It is known that forget-ness follows an exponential or power decay, and Anki helps you managing this, by tracking what you learn, how good is your recall (for every asked question you are asked for how good did you remember it). So, you don't need to remember you are forgetting something.
That was it, in a pinch. Anki is a multi-platform (really: there are native programs for Android, iOS, Mac, Linux, Windows and even a web-based version: I could even use it in Plan9 I think) SRS program, so it's a very good choice for it (also, can be scripted with Python and has a lot of add-ons to create/manage knowledge cards.) Forgot to mention: knowledge for SRS or Anki is flashcard-based.
Studies have shown that SRS is best for knowledge-based facts ("rain in spain stays mainly on the ...") rather than for performance-based facts. I.e. if you suck at playing Autumn Leaves on guitar, using SRS to check it won't improve it (for these tasks, deliberate practice is the supposed holy grail). For a full account on what is, what isn't and a lot of research about it check this post.
Now, enter tsumego. There are two sides in tsumego (well, maybe even 3):
- Reading the solution and variation branching, brute force.
- "Knowing" the shape and the key point (being shown the solution repeatedly.)
- Related to both there's "knowing there's something in there".
When I was asked by Namii to go through all Cho-1, I dreaded going from problem 1 to problem 900 and repeating ad-nauseam. Also, the source of Cho-1 I'm using (Tasuki's) doesn't have answers for them, so no matter what, I'd need to read the solution every time in fullest. So, the questions asked by Anki aren't see the problem, solve it and see the solution as a check, but see the problem and read the f******g problem out until sure. To wrap up the effectivity, when a problem gives me trouble (i.e. I take a long time to read it, or there's something in the shape that I can't just wrap my head around to) I mark it as "not very good," or even "wrong" so I'm shown it more and more often (here's a small deliberate practice part.) Takeaway: I'm doing a lot of problems per day, randomised automatically.
Before you rush to download Anki, some caveats:
Don't be overzealous about new cards:
The modus operandi of Anki is: get some data, put it into a deck (virtual deck of flashcards,) set a limit for new-cards-per-day (N for this explanation) and for review-cards-per-day (R) and start.
For the first day you'll see N cards, and no R cards. Assuming everything is in default and you can solve/know the answers to all N cards easily (for tsumego is likely, for other types of decks isn't, unless you are the creator of the cards) you'll mark them as "good" and you won't be shown them again for 4 days. On days 2 and 3 you'll again see N cards and no R cards. On the 4th day you'll see R+N cards. Etcetera.
If no-one tells you (and even then,) you'll be overzealos and set a high N. For instance, 50. And even then, for the first few days you'll feel 50 problems is just too few and will add a few more extra (you can raise N by the end of the session to see some more problems in the day.) But beware of the cumulative compound effect. Assuming there are 900 problems.
- 50 | On day 1 you'll see 50 N
- 100 | On day 5 you'll see 50 N and 50 R (from day 1). If all R are perfect again, all go to +7
- 150 | On day 12 you'll see 50 N, 50 R (from day 1, 4+7=11) and 50 R again (from day 8.) Cards shown with +11 get now a ~+20, the +4 get a +7 (so they aren't shown until day 35, when there are no more N cards)
So, you'll be seeing in a best-case scenario, at least 150 problems per day (as soon as you don't answer perfect this will raise.) Which is quite a lot. For my own Anki I used 30 (so as to finish in 30 days) and had some pretty long and tiring review sessions (had a couple days with around 130 problems taking almost 1 hour.) Now that I'm slightly wiser, if I add some new cards to my decks I try to limit the overall N to 10. I know, it's very small, but I'm already solving 60+ cards per day.
Also, after running out of new cards, things get somewhat easier, so for short decks I kind of recommend eating it and using a very large N so as to run through all new cards as fast as possible.
Some more figures: After 32 days, I have run over Cho-1 completely. During this period I have solved problems from it 2751 problems, 86 per day. And has taken me 17 hours with an average of 30 minutes per day. Has my reading improved? I don't feel like my reading is any better for problems that are not in Cho-1, even for some of them my reading is lousy (funny bit: two problems out of the 900 gave me trouble... and the three were in last month's [url=http://www.nordicgoacademy.com/[/url] tsumego assignment). I do feel though that my reading clarity feels better, but this is very subjective and 32 days is probably too few yet to know what its effect was.
Getting data into Anki
The best way to get cards to Anki (specially if you don't care about solutions of problems) is using images. There's a plugin for creating decks from folders full of images. By default it will generate cards with the format filename (front)-image (back.) I edited the decks to have image on front (by exchanging back-to-front) and nothing on back (setting back to none, in the same menu.)
For Cho-1 I generated (from the LaTeX sources, good luck if you try) a 1-problem-per-page PDF and converted the PDF to JPG pages with Imagemagick, then cut them in half also with IM. For other sets (LCH-Tesuji 1, Cho-2) I used my sgfrender program with auto-generated captions (for who plays in LCH-Tesuji). For problems I'm inputting by hand, I create the diagram with GoWrite in B/W mode, and just take a cropped screenshot I put into an Anki card. Fastest way I've found (specially since Mac has a key command for screen-selection screenshot.)
What I'm using Anki for now?
Currently I have 6 go decks, 2 non-go decks (one for general programming language stuff I have a tendency to forget, the other for APL builtins). The go decks are:
- Cho-1-Elementary
- Dictionary of Basic Tesuji
- From reviews
- Proverbs
- LCH-Tesuji-1
- Opening Theory Made Easy
For DBT, I'm inputting the problem, and as answer an explanation of the tesuji used. For now I have only cards for the first simple separating sequences not considered tesuji (because they are very interesting and good to know) and for the first 10 problems and techniques of "separating". (Started creating this deck just 5 days ago)
From reviews holds sequences suggested in reviews by the NGA. So, if in the game I played A and a teacher says B is far better, I remove A from the board, screenshot, phrase a question (like "Black is being attacked, what to do?") and then show the teacher answer variation together with the explanation (probably one of the most useful decks, because it's purely memory, removing bad habits)
Proverbs holds random "go comments", in particular from my coaching sessions with Jeff. Also comments on technique or shape from these sessions.
OTME: I started with Principle 8, and converted the chapter into a problem-question format. Seems also pretty useful (Started creating this deck 5 days ago)
I guess that's it. A very long post... Probably a little chaotic. If you have any question about this, please ask
