How to use the next-generation spaced repetition algorithm FSRS on Anki?

好的,谢谢老哥解惑 :kissing_closed_eyes:

I’ve tried using this algorithm twice but failed each time. Last time my 1st good review was rated 4.5 month :slightly_smiling_face:

Could you share the parameters that you used? And the true retention in your past review.

Retention 90%
[4.7107, 7.036, 135.0907, 135.0907, 4.8477, 0.1862, 0.7813, 0.2704, 1.4758, 0.5463, 0.948, 2.3465, 0.0316, 0.4232, 1.7249, 0.227, 2.7415]

135 is something I wasn’t expecting
(Note: I used the hugging face website and couldn’t use the helper add on to reschedule all my cards, I don’t have any laptop)

I guess you real retention is very high, maybe larger than 95%.

Then the frequency shouldn’t be less?
I don’t know what’s wrong with my cards, but I really want to use fsrs
What should I do ?

When did you generate these parameters? The last important update to the optimizer was on August 14.

If re-generating the parameters today solves the problem for you, then fine. Otherwise, you can create an issue on GitHub: Sign in to GitHub · GitHub

Creating an issue there would help in faster resolution of your problem.

@L.M.Sherlock Im looking at the commits for Anki GitHub. Are you integrating FSRS into Anki naturally in the next release?

Yup. Dae and I are working on the integration.


It would be great for some instructions on how manage on this feature as native. But also want to thanks for this new amazing resource

Is it possible to filter out certain reviews that the users specifies based on criteria such as rating and review platform?

I am asking because Ankiweb doesn’t have a bury, suspend or skip feature. I worry that this will mess up FSRS in cases like this example:

Case: There are times of the day where a user can only study via Account Login - AnkiWeb . They have listening cards with only audio on the front side. At their study location, they cannot listen to the audio.

Problem: They are forced to give a rating without having conducted an actual review.

I don’t know. You should ask @dae.

It is not possible to filter by platform. You’ll need to put them in a separate deck and avoid that deck, or use one of the mobile clients instead of AnkiWeb.

If a learning step being longer than one day is bad because FSRS can’t see when the review happens, are steps like 23h that will be due on the next day also bad for it? What about short steps, but which I reviewed when they were long overdue?
Does having long steps in the revlog over the years affect rescheduling significantly now? What about the “Next day starts at” setting? I changed it many, many times.

Yeah, it’s bad if you only use the custom scheduling. But the helper add-on or the built-in FSRS in 23.10 could deal with it correctly.

1 Like