Open Expertium opened 1 month ago
I'm waiting for the new dataset.
What do you mean?
I mean I will re-benchmark everything with the new dataset.
Did Dae say that he will make a new dataset soon?
Oh, I see. I think 20k is better though, I'd prefer not losing accuracy.
My disk would complain. And the benchmarking time has been too long.
Will the new dataset have deck names or preset names or both?
It includes deck_id and preset_id.
Currently, all models only use interval lengths and grades. I tried incorporating time of the day and the number of reviews done on that day into FSRS, but it didn't improve results. However, it's possible that a neural network would be able to utilize that information. Please make GRU-P-F (or whatever you want to call it, I'm not good with names), which will have 4 input features.