Then you have to define card similarity, and that’s a whole new can of worms. And no, you can’t use text/audio/images.
I mean “similarly” as in same Difficulty, Easy, Stability, whatever…
- For example, gather all cards with a difficulty range of 70-80%, 60-70%, 50-60% and then make parameters out of their reviews
versus
- Gather all cards pertaining to Presets 1, 2, 3 and then make parameters out of their reviews.
But Difficulty, Stability and Retrievability themselves depend on parameters. So you get a circular dependency.
Just go over your 237 presets, man. And combine them until you only have 5-10.
Hen-or-egg-first conundrum. I get it. Sorting cards by their presets vs sorting cards by their Difficulties (which depends on params) to calculate params. I was trying to brainstorm to find ways in which cards can be sorted in a non-arbitrary (like presets) way to calculate params, because for the time being, the number of cards with a particular preset sizes are arbitrarily defined by the user, where Difficulty is not. But I get the argument.
Perhaps another sorting method (which is not presets) which does not depend on FSRS params in the first place is the solution 
I did 
2 weeks ago. Now I only have a preset for each subject.
You’re effectively asking for some kind of clustering algorithm to be applied to cards. This is unlikely to be feasible given the relatively small review history each individual card will have and the low signal-noise ratio in that review history.
It would be nice if there was some compromise though, e.g. Anki periodically automatically optimize every deck/subdeck and deck/subdeck split by card type, then use some heuristics based on card count and log loss to determine which parameters to use for which card.
2 Likes
Something like that. My point is that presets are an inefficient way to group and sort cards for calculating parameters based on them because they are arbitrarily set. I know that FSRS needs data to work with, there is no question about that.
But any non-arbitrary way, be it an algorithm like you said or something else, to group cards and build params on them would be better than a haphazard arbitrary way.
1 Like
Presets aren’t arbitrary, though. They are made by the user (who, I assume, is doing something more rational than throwing darts at a dartboard with deck names on it while he’s blindfolded), and they are based on the content of the cards.
1 Like
If I could theoretically give decks 1-10 all a single preset, I could. If I could split the decks into groups of 2, making 5 presets, I could. If you could make a preset for a group of 5, making 2 presets, I could. Multiply that if you have 100s of decks and then you get endless combinations.
Yes, in the sense that presets are assigned by the user to decks that are about the same context (e.g. subject, topic, etc.). In that sense, I could say they are not arbitrary. But the problem is that they still are with regards to:
- how the user “defines” which decks belong to which topic. Different users have different interpretations. If I give user A and user B the same topics and tell them to group them in a certain way, they will most likely come out with different outcomes.
- what gets to decide how many presets apply to how many decks? Different users would most probably come to different outcomes.
That is arbitrary.
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.