-
|
Why haven't you used popular spaced repetition algorithms for your scheduler? (SM2 or FSRS) |
Beta Was this translation helpful? Give feedback.
Replies: 2 comments
-
|
The first phase of the algorithm is close enough to those algorithms in the sense that it's doing exponential decay to simulate a forgetting curve. However, the problem is sufficiently different once you add hierarchical information that I didn't think it would be ideal to port an existing algorithm. |
Beta Was this translation helpful? Give feedback.
-
|
Ok. So I had some time to look into FSRS and replaced the existing exponential decay scorer with a power law scorer that has a simplified stability and difficulty measure. It's far from being as full-fledged as FSRS, but as the comment explains, there are differences in trane that make trying to optimize individual scores not as important as it is for products that just have a deck of cards with no structure. New version: https://github.com/trane-project/trane/blob/master/src/exercise_scorer.rs |
Beta Was this translation helpful? Give feedback.
Ok. So I had some time to look into FSRS and replaced the existing exponential decay scorer with a power law scorer that has a simplified stability and difficulty measure. It's far from being as full-fledged as FSRS, but as the comment explains, there are differences in trane that make trying to optimize individual scores not as important as it is for products that just have a deck of cards with no structure.
New version: https://github.com/trane-project/trane/blob/master/src/exercise_scorer.rs