Fix/keep data if count >= 6

[L-M-Sherlock]

No description provided.

[user1823]

@L-M-Sherlock, the new code doesn't work as it was originally designed.

The original design was:

• If count < 6, remove all the cards in that delta_t (even if it is more than 5% of the total). This is because when count is less than 6, then the value of recall is not reliable. For example, if I get 4/5 correct, the recall is 80% and if I get 3/5 correct, the recall is 60%. This is a very significant difference that is caused by only a single additional lapse.
• Even if count ≥ 6, we can remove up to 5% of the total reviews (or up to 20 reviews, whatever is greater). The reason is that the case of 6 reviews or 7 reviews is not very much different than that of 5 reviews. So, we shouldn't stop filtering after we have at least 6 reviews in a given delta_t. The condition of 5% will prevent filtering of too many reviews.

[L-M-Sherlock]

This PR made the implementation consistent with the FSRS-rs. And it could keep more data from the filter.

[user1823]

I would say that it is better to make FSRS-rs consistent with the previous behavior of fsrs-optimizer. I have explained the reasons in the previous comment.

And it could keep more data from the filter.

Yes, but pre-train is sensitive to outliers and needs a robust outlier filter. In fact, the need for a stricter outlier filter in pretrain and a less strict outlier filter in train was the motivation for open-spaced-repetition/fsrs-rs#121

[L-M-Sherlock]

I’m satisfied with the current implementation. So it will not be changed unless someone reports that the optimizer provides bad parameters.