Spaced Repetition Software for Tsumego

[Boidhre]

I think none exists specific to go (I'd really welcome being corrected here). Has anyone had any success using Anki or Memosyne as part of their tsumego practice? You could hack something together using a simple set of flashcards with problem numbers on them and answering honestly how difficult a problem was for you and/or if you got it wrong. This kind of technique is very big in language learning for drilling vocabulary and other memorisation tasks and I don't see why it wouldn't work for go. I'm actually surprised that go grinder or uli-go haven't already implemented this feature already, I'd very happily pay for such a feature.

Edit: There's a sgf viewer for Anki but it's bugged and hasn't been updated in ages which is a real pity.

[Mike Novack]

I think none exists specific to go (I'd really welcome being corrected here).

I could be wrong but I believe that it does exist. It's not exactly easy to be sure about the behavior of something throwing problems at you (with color reversal and rotations) to be sure to what extent your statistics getting a particular one right or wrong has affected the frequency of reappearance.

Since you say you are willing to pay ...... well you shouldn't do that until you are sure of the behavior. I believe that the MFOG trial version will let you access the easy set of problems. You could try to keep statistics on reappearance (and the rate of appearance of new problems) dependendent on whether you answer correctly or not. You could intentionlly choose to always answer one wrong and see if that caused it reappear more frequently.

[Boidhre]

I think none exists specific to go (I'd really welcome being corrected here).

I could be wrong but I believe that it does exist. It's not exactly easy to be sure about the behavior of something throwing problems at you (with color reversal and rotations) to be sure to what extent your statistics getting a particular one right or wrong has affected the frequency of reappearance.

Since you say you are willing to pay ...... well you shouldn't do that until you are sure of the behavior. I believe that the MFOG trial version will let you access the easy set of problems. You could try to keep statistics on reappearance (and the rate of appearance of new problems) dependendent on whether you answer correctly or not. You could intentionlly choose to always answer one wrong and see if that caused it reappear more frequently.

Thanks for the idea but in order to be a SRS system there has to be some grading of how hard it was to get right by the user, not just whether your answer was correct or not. That allows you to get tested rarely on material you find trivial and very regularly on material that still challenges you even though you're still getting the answer right. Assigning the same frequency to two problems that I get right first time doesn't make any sense, the first might have been solved at a glance, the latter requiring reading. I need to see the latter again very soon and regularly to try and get it to the point where I can solve it at a glance, the former I can leave for far longer before looking at again. MFOG definitely doesn't cater for this and neither does any other go software I've come across so far.

I can do what I already do and use an Anki deck of cards from 1-100 and have each number assigned to a problem in smartgo and have copies of the deck assigned to different batches of 100, but that's cumbersome solution.

[hyperpape]

There have been a few discussions and dfan was apparently trying something: viewtopic.php?f=12&t=1740&p=33206&hilit=spaced+repetition#p33206

[Boidhre]

There have been a few discussions and dfan was apparently trying something: http://lifein19x19.com/forum/viewtopic. ... ion#p33206

Yeah, I saw that when I was searching for mentions of SRS in go, but didn't want to drag up a thread from a year and a half ago about it. I should have referenced it in my OP though.

[RBerenguel]

xiangz has it in his todo for EasyGo (see http://lifein19x19.com/forum/viewtopic.php?f=18&t=5356&p=93105&hilit=Easygo#p90143), low priority though.

[Boidhre]

xiangz has it in his todo for EasyGo (see http://lifein19x19.com/forum/viewtopic.php?f=18&t=5356&p=93105&hilit=Easygo#p90143), low priority though.

Excellent, thank you very much!

[RBerenguel]

We should all thank xiangz, EasyGo is awesome!

[Boidhre]

We should all thank xiangz, EasyGo is awesome!

Indeed.

[Actorios]

I think none exists specific to go (I'd really welcome being corrected here). Has anyone had any success using Anki or Memosyne as part of their tsumego practice? You could hack something together using a simple set of flashcards with problem numbers on them and answering honestly how difficult a problem was for you and/or if you got it wrong. This kind of technique is very big in language learning for drilling vocabulary and other memorisation tasks and I don't see why it wouldn't work for go. I'm actually surprised that go grinder or uli-go haven't already implemented this feature already, I'd very happily pay for such a feature.

Edit: There's a sgf viewer for Anki but it's bugged and hasn't been updated in ages which is a real pity.

I've been looking for the exact same feature.
I did use Anki on iPad for this purpose (coding problems as pictures).
It was a lot of work for 1 month practice upon 500 problems. At the end, I felt it was doing me good and I do believe spaced repetition shape up your instinct but it was a shame not to be able to test variations or play move by move.

Now, as RBerenguel stated it, Easygo proved to be a much easier solution even if I'm missing the spaced repetition. It's still better to have the opportunity to play out the problems even if you use simple sorting to go through your library than having spaced repetition with pictures.

I'm still hoping this features get released either through Easygo or, maybe, through Smartgo (I wouldn't be surprised if a "Smartgo problems" application was released one day or if Smartgo Books was improved to tackle specifically go problems books).

[Boidhre]

I think none exists specific to go (I'd really welcome being corrected here). Has anyone had any success using Anki or Memosyne as part of their tsumego practice? You could hack something together using a simple set of flashcards with problem numbers on them and answering honestly how difficult a problem was for you and/or if you got it wrong. This kind of technique is very big in language learning for drilling vocabulary and other memorisation tasks and I don't see why it wouldn't work for go. I'm actually surprised that go grinder or uli-go haven't already implemented this feature already, I'd very happily pay for such a feature.

Edit: There's a sgf viewer for Anki but it's bugged and hasn't been updated in ages which is a real pity.

I've been looking for the exact same feature.
I did use Anki on iPad for this purpose (coding problems as pictures).
It was a lot of work for 1 month practice upon 500 problems. At the end, I felt it was doing me good and I do believe spaced repetition shape up your instinct but it was a shame not to be able to test variations or play move by move.

Now, as RBerenguel stated it, Easygo proved to be a much easier solution even if I'm missing the spaced repetition. It's still better to have the opportunity to play out the problems even if you use simple sorting to go through your library than having spaced repetition with pictures.

I'm still hoping this features get released either through Easygo or, maybe, through Smartgo (I wouldn't be surprised if a "Smartgo problems" application was released one day or if Smartgo Books was improved to tackle specifically go problems books).

What I currently do is this:

Anki on my PC has a simple deck of 1-100 corresponding to problem numbers in GoGrinder. What I do then is have Anki tell me what problems I need to do in GoGrinder this session. Cumbersome but it works well enough and as you say, you get to play out the sequence. I break up problems into batches of 100 for simplicity and copy the same deck out multiple times but there's no reason one couldn't make a larger deck if one had the patience. Ideally yes you want the one program doing both jobs but that doesn't exist at the moment. You could do use this as well with a book with numbered problems for revision purposes.

[Mike Novack]

I'm going to disagree a bit but that is actually more of a philosophical matter about what do we know and how do we know it -- questions about "operational" reality.

Forget for just a moment that the computer is a machine and not another human. It (or he or she) has presented you with a problem and recieves in return the information whether you got that problem right or wrong. The internal processes of your mind are not disclosed to it (or him or her) and in fact, not even the information whether you are a human or a machine.

Besides, this immense difference you imagine between what you saw at a glance (or because of some fundamental principle)vs what you had to read? How long does that difference hold up? How long does it affect the frequency with which you are fed particular problems? If the problem is truly trivial to you, you will always get it right (when presented with it) and so it drops more rapidly to low frequency of reappearance than a problem which you got right in spite of being harder and so will presumably will sometimes get wrong.

You suppose that is not the case? That you will not "learn" the answer to the one which required reading? That it continues to be a deep reading problem for you? Well then answer me this. How did you spot that this problem required deep reading and not just the application of some fundamental principle? Even if you do not remember the full reading haven't you at least learned "this is a reading problem"?

Remember, the is an important difference between the way "problems" are usually presented and real situations on the board -- because it was presented as a problem (you are being told there is something here). Imagine instead being presented with "problems" mixed in with "non-problems" (no solution exists). I find problems much more difficult when the exact objective isn't spelled out, not "black to kill" but "black to kill, or getko, or seki, or gain a few points in sente" (balck to gain something). Another step up in difficulty were that mixed in with examples where there wasn't anything particularly gainful. Problems "if there is something here, say what it is (slve it), or report "nothing available".

The point of that is there is more learning involved with problems than the exact solution. Along the way we also learn to remember/recognize problems in terms of their achievable objective (go for a kill; get a ko; get seki; gain a few points in sente).

[Boidhre]

Remember, the is an important difference between the way "problems" are usually presented and real situations on the board -- because it was presented as a problem (you are being told there is something here). Imagine instead being presented with "problems" mixed in with "non-problems" (no solution exists). I find problems much more difficult when the exact objective isn't spelled out, not "black to kill" but "black to kill, or getko, or seki, or gain a few points in sente" (balck to gain something). Another step up in difficulty were that mixed in with examples where there wasn't anything particularly gainful. Problems "if there is something here, say what it is (slve it), or report "nothing available".

Interesting, but isn't the point of doing (easier) tsumego to improve you ability to spot killing patterns quickly so when looking at a board a pattern might jump out at you and say "there's a kill here" or "there's bad aji here" or whatever? You speed up your reading by taking shortcuts like this, allowing you to reject certain eye shapes (for instance) as unconditionally alive, some as dead and some as killable and requiring urgent action. Ditto for certain shapes within eye spaces for living and killing. I'm a complete beginner though, so I may be completely wrong in this however.

I'm unsure how useful SRS would be for hard problems that you do for reading practice but since I'm nowhere near a level where this would be useful for me to be doing (as another on this forum put it, beginners don't really have the base from game experience to be building reading up anyway) this is not a question I can answer.

[Bill Spight]

Here is an algorithm that I used to use for overlearning. It depends upon then number of failures (F) and the number of successes (S) for a problem. A new problem starts off with F = 1. A problem is declared overlearned when S > F/2.

So if you solve the problem the first time out, S = 1, and 1 > 0.5, so it is overlearned and you are done with it. If you fail the first time, then you have to succeed twice in a row. Then S = 2 and F = 2, and 2 > 1. Etc., etc.

One obvious problem is that you could declare a problem overlearned when you still get it wrong more often than not. What if you have a success or two and then fail? In that case I added 2 to the number of successes needed. One thought is that if you already have a success, count a failure as double. So if you fail three times and then succeed twice, so that F = 4 and S = 2, and then you fail again, add 2 to F so that F = 6 and S = 2, and you need two straight successes for overlearning.

[RBerenguel]

As interesting as SRS is (in my fun-todo is a bash & text based SRS small system for fun,) currently I'm overlearning with the crude method. Randomize the list of problems and do as many as you can without failing. When you fail, reshuffle and repeat. When I can get the list straight... Well, it can take a while still (record 28, there are 123 problems)