Personal library of mistakes

[acosmicjoke]

Hi! This is my first post on this forum. I had a very similar experiment last summer so I thought I should share.

What I did was write a shell script using sgf2dg that converts go problems in an sgf file into png images, then creates anki flashcards out of them. The workflow was to load up a game record, manually find the mistakes with some help from ai, then create multiple games in a single sgf file, each containing a single problem created from the situation when the mistake occurred, and the correct solution. Next, run the script on the sgf to create an anki deck with all the problems which I can import into anki.

I don't think creating the actual problems is something that should/can be automatized. For one, because I don't necessarily only check the big winrate drops, but also the parts of the game where I was unsure if what I did was a good idea. Also, the "optimal" ai variations, where it keeps tenukiing all the time, are not very digestible for me. I'm usually more interested in the good local moves, and than I have to ask more questions from the ai to understand why those are good moves. Just slapping the standard ai variations at the big winrate drops on the flashcards wouldn't be very educational.

My experience was that srs can certainly be helpful, but you should work with very different settings that what you use for learning languages, otherwise you can easily end up wasting time overreviewing mistakes that you are already sick of instead of playing more games. The goal isn't really to learn solutions to particular problems by heart, but to recognize certain underlying patterns. The standard spaced repetition algorithm is supposed to ask you the question again just barely before you forget the answer, but I think for go problems, It's best to be asked the question again after you did forget the solution to the actual problem but solving it again is still easier than originally because you remember some patterns to look out for and viewpoints to consider. I ended up with setting the minimum interval between getting the same problem to 1 week (the default is 1 day), and the new interval in case you fail to answer correctly 50% of the original interval (the default is to set it back to the minimal interval). I also use an addon called "MIA retirement" that automatically suspends cards in my deck which reach a certain interval length, it's set to 3 months currently. This because I don't really want to keep reviewing the same mistakes until the end of time. Even with all these adjustments, the anki time felt too long, and in most cases, I still just outright remembered the solutions of the problems on the cards. When this happened I considered the criteria for pressing "good" to still read things out and verify that what I remembered is correct.

Unfortunately, my go time was cut back around mid-autumn, so I stopped doing the spaced repetition. It did make me improve, especially in the direction of play department, but I think it was a bit of an overkill. Nowadays I have more time again, but I'm content with just reviewing my reviews about 1 week after originally doing them.

[Jæja]

I'd like to do something seamingly simple: add a variation to an existing game that improves on the actual game and add a comment to it, e.g. "Black is strong around here, so the double hane is powerful." I'd then like to export a series of diagram pairs: actually played and improved variation.

I'm thinking about a poor man's solution where I comment a position in a special way that can then be parsed by a script, e.g. "Black is strong around here, so the double hane is powerful. <43>". In this example, this will use the moves until move 43 of the main variation (the actual game) for one diagram and the complete variation (the improvement) for the other diagram, with the comment used as the caption for the latter.

Does something like this exist, or do you think I'll have to write it myself? Or perhaps there's even a better way?