4 ms·

The consensus within the Anki community seems to be that yes, the SuperMemo scheduling algorithm (SM-17) is a bit better than Anki's (SM-2 variant), but not by

The consensus within the Anki community seems to be that yes, the SuperMemo scheduling algorithm (SM-17) is a bit better than Anki's (SM-2 variant), but not by enough to matter, so you shouldn't worry about it. But when you look into tests people have done to see how much better it is exactly, it's around 30% more efficient. I still use Anki and am grateful for its existence, but this is a huge difference in efficiency. I wish there was a better open source scheduling algorithm available.

SamBam 5y ago
How come Anki cant update it itself? Is SM-17 just a different set of exponential-decay functions from SM-2? Is that even copyrightable? Also, if Anki is open-source, I'm surprised that there is no unofficial port that uses SM-17. But maybe I'm just ignorant of how much work that would involve.
- SirensOfTitan 5y ago
  Woz goes into detail about the SM-17 algorithm here: https://supermemo.guru/wiki/Algorithm_SM-17 https://supermemo.guru/wiki/Algorithm_SM-17 ... he's built these through decades of thinking of and tinkering and working on this singular problem. SM-2 is from 1987 for SM for DOS 1.0, and is fairly simplistic comparatively.
- Asraelite 5y ago
  SM-17 is much more complicated and also closed source. You would have to spend a while reverse-engineering it.
tmaly 5y ago
Is there a version of Anki that uses SuperMemo or is there another software that implements it?
andai 5y ago
I'm curious about those tests, do you have a link? I'm also curious if there's been any attempts at open-source versions of more recent supermemo algorithms.
- Asraelite 5y ago
  Unfortunately there's no single good source for this; 30% is just an approximate aggregate figure I've come up with after seeing various people's posts on different sites. Some things I could find just now: Anki vs. SM-8: https://anki.tenderapp.com/discussions/effective-learning/1258-sm-2-vs-sm-15/page/2?full_site=1 https://anki.tenderapp.com/discussions/effective-learning/12... SM-2 vs. SM-17: https://supermemo.guru/wiki/Universal_metric_for_cross-comparison_of_spaced_repetition_algorithms https://supermemo.guru/wiki/Universal_metric_for_cross-compa... Anki vs. custom machine learning scheduler: https://old.reddit.com/r/Anki/comments/d0i8uy/improved_algorithm_for_anki_scheduling_w/ https://old.reddit.com/r/Anki/comments/d0i8uy/improved_algor... Anki vs SM-17: https://unrelatedwaffle.medium.com/battle-of-the-spaced-repetition-heavyweights-anki-vs-supermemo-part-ii-10ef2cd0379d https://unrelatedwaffle.medium.com/battle-of-the-spaced-repe... SuperMemo seems to fall behind Anki in the short term (<6 months), but eventually makes up for it. Some of the links reflect that. As for open-source implementations, there this: https://github.com/slaypni/SM-15 https://github.com/slaypni/SM-15

SamBam 5y ago

How come Anki cant update it itself? Is SM-17 just a different set of exponential-decay functions from SM-2? Is that even copyrightable? Also, if Anki is open-source, I'm surprised that there is no unofficial port that uses SM-17. But maybe I'm just ignorant of how much work that would involve.

SirensOfTitan 5y ago
Woz goes into detail about the SM-17 algorithm here: https://supermemo.guru/wiki/Algorithm_SM-17 https://supermemo.guru/wiki/Algorithm_SM-17 ... he's built these through decades of thinking of and tinkering and working on this singular problem. SM-2 is from 1987 for SM for DOS 1.0, and is fairly simplistic comparatively.
Asraelite 5y ago
SM-17 is much more complicated and also closed source. You would have to spend a while reverse-engineering it.

tmaly 5y ago

Is there a version of Anki that uses SuperMemo or is there another software that implements it?

andai 5y ago

I'm curious about those tests, do you have a link? I'm also curious if there's been any attempts at open-source versions of more recent supermemo algorithms.

Asraelite 5y ago
Unfortunately there's no single good source for this; 30% is just an approximate aggregate figure I've come up with after seeing various people's posts on different sites. Some things I could find just now: Anki vs. SM-8: https://anki.tenderapp.com/discussions/effective-learning/1258-sm-2-vs-sm-15/page/2?full_site=1 https://anki.tenderapp.com/discussions/effective-learning/12... SM-2 vs. SM-17: https://supermemo.guru/wiki/Universal_metric_for_cross-comparison_of_spaced_repetition_algorithms https://supermemo.guru/wiki/Universal_metric_for_cross-compa... Anki vs. custom machine learning scheduler: https://old.reddit.com/r/Anki/comments/d0i8uy/improved_algorithm_for_anki_scheduling_w/ https://old.reddit.com/r/Anki/comments/d0i8uy/improved_algor... Anki vs SM-17: https://unrelatedwaffle.medium.com/battle-of-the-spaced-repetition-heavyweights-anki-vs-supermemo-part-ii-10ef2cd0379d https://unrelatedwaffle.medium.com/battle-of-the-spaced-repe... SuperMemo seems to fall behind Anki in the short term (<6 months), but eventually makes up for it. Some of the links reflect that. As for open-source implementations, there this: https://github.com/slaypni/SM-15 https://github.com/slaypni/SM-15