srs-benchmark

srs-benchmark copied to clipboard

Expertium

Weighted average by reviews: | Algorithm | Log Loss | RMSE (bins) | Parameters | | --- | --- | --- | --- | | FSRS-4.5 | 0.3252±0.1514 | 0.0533±0.0334...

L-M-Sherlock

Some weird first forgetting curves

11

     

L-M-Sherlock

Weighted average by reviews: | Algorithm | Log Loss | RMSE (bins) | Parameters | | --- | --- | --- | --- | | FSRS-4.5 | 0.3252±0.1514 | 0.0533±0.0334...

L-M-Sherlock

- https://github.com/open-spaced-repetition/fsrs-optimizer/pull/81

L-M-Sherlock

Ebisu?

7

Thank you for this very interesting analysis! If you all feel inclined to include it, I'd be curious to see how [Ebisu](https://fasiha.github.io/ebisu) compares.

andymatuschak

L-M-Sherlock

Inclusion of any of the boosting models

7

Would it be possible to include any of the boosting models like catboost, lightgbm or xgboost, since they are very good with tabular timeseries data such as this and there...

Gilfaro

#102 Sample usage of Catboost on top of FSRS-5. It uses "delta_t", "power_forgetting_curve", "stability", "difficulty" from FSRS-5 as input to the boosting algorithm.

Gilfaro

Add features to the dataset: time (in hours) since 'day_starts_at' and workload of today before current review

16

This issue is a reminder to add the following features to the dataset and make a version of FSRS that can use them: 1) Time (in hours) since 'day_starts_at' when...

Expertium