I changed the way I grade my cards. Should I defer optimization to give FSRS more data first?

Anki FSRS
1

Backstory

I have always used 3 buttons only (all except for “easy”). Now, I want to push cards further into the future if they are ridiculously easy. In the past, I had graded those cards as “good”. But now, I started to grade them as “easy”.

My parameters had been optimized with my three button usage only (so they have exactly 0 review data for “easy” cards, because I never used “easy” at that point).

Question

Now I’m wondering: should I wait with optimizing my preset if I change the way I grade myself? My deck options screen does show me a reminder to optimize but I’m wondering whether it’s better to wait until FSRS has more data on it.

I also ask this question because it might be an important consideration for Anki, if / when automatic optimization should be implemented.

There is also this (source: fsrs4anki/docs/tutorial.md at main · open-spaced-repetition/fsrs4anki · GitHub):

the optimizer requires several reviews to fine-tune the parameters.

But I don’t think it’s clear whether my 3-button usage data is enough, or if I need 4-button usage data.

Version

Anki 25.09.2 (3890e12c) (ao)
Python 3.13.5 Qt 6.9.1 PyQt 6.9.1
Platform: Linux-6.12.41+deb13-amd64-x86_64-with-glibc2.41

4 Likes
2

Can’t give you a technical answer but I always wait for some time before optimisation.

2 Likes
3

Judging by the fact that noone in Anki > FSRS seems to know the answer I’m guessing that this is nothing that has been documented yet.

@Expertium, @L.M.Sherlock: I only know of you two that are involved in FSRS directly. Maybe one of you can answer this?

That sounds like you are changing the way you grade yourself quite often. Probably not what you meant, though. I also am waiting regarding optimization currently, mostly because I am having the question as outlined above. Otherwise, I would probably have optimized right away.

4

I don’t know for sure, but optimizing early shouldn’t be harmful because

  • there isn’t much data for the parameters to significantly change anyway
  • the parameters trained on old + new data are more likely to be better fitted to the newer reviews than the parameters trained only on the old data.
1 Like
5

I think your instinct to defer re-optimizing is a good one. Better to have some data there than none.

What are your current parameters, though? I would suspect that either –

  • w[2] and w[3] (the 3rd and 4th parameters) are nearly to the same value – “I don’t know anything about Easy cards, so I’ll treat them like Good cards”
  • or w[2] is normal and w[3] is 100.0000 – “I don’t know anything about Easy cards, so I’ll treat them like infinity.”

Neither of those would be quite what you want when you’re shifting to a new habit. :sweat_smile:

1 Like
6

Good point. Interesting enough, none of those are the case for me. My params:

0.6043, 3.2790, 12.1755, 37.2240, 6.0814, 0.7519, 3.0183, 0.0010, 1.9701, 0.2486, 0.9183, 1.5683, 0.0228, 0.3606, 1.7208, 0.1724, 1.8729, 0.5931, 0.4230, 0.0273, 0.1000

And here my params in the visualizer: Anki FSRS Visualizer.

7

Nothing alarming (but you might want to see what happens with a few Easy grades in a row). It wouldn’t be the worst thing in the world for you to optimize now, but waiting is probably more prudent.

[Adding]
@Anon_0000 I re-read my message and I think it was unclear. When I suggested you “see what happens with a few Easy grades in a row” – I meant in the Visualizer, not testing that on actual cards. :+1:t4:

1 Like