Expertium
Expertium
@L-M-Sherlock do you plan to remove it now? If not, close this
Can you describe it step-by-step, please?
Also, regarding estimating the lowest value. For AUC and logloss, it can be done in the following way: 1) For each user, will will need alpha and beta to model...
Previously, I estimated alpha and beta an very crude way: 1) Take probabilities given by FSRS 2) Plot a calibration graph, like this:  Well, _plotting_ isn't necessary, but I'm...
You can plot one here: https://colab.research.google.com/github/open-spaced-repetition/fsrs4anki/blob/v5.3.1/fsrs4anki_optimizer.ipynb It's plotted after optimization I wonder if we can get alpha and beta from the bins themselves, without using FSRS predictions. It would probably...
When I say "Oracle", I mean "an algorithm that knows the true probability of recall for a given card at a given time". I DO NOT mean "an algorithm that...
> So you want something like {bin1: (0.90, 16), bin2: (0.95, 23), ...}? Yep.
> That is, do this calibration on the train set and see if it improves performance on the test set. If so, we may come up with an FSRS v4.1...
 Alright, this looks promising. Kinda. Code: ``` import numpy as np from scipy.optimize import curve_fit from scipy.special import gamma def betafunc(x, a, b, scale): return scale * gamma(a+b) *...
Alright, I'll do it myself and then report the results.