Hi. I have some problems with FSRS. In particular, the calculated difficulty levels often seem to make NO SENSE based on my review history. For example, I will show you a card I haven’t gotten wrong in about a year. Why is it calculating difficulty at 95% when this is obviously an easy card for me? When I was originally learning it, I may have gotten it wrong a few times. But that is why I set the optimizer to only take into account reviews after 7/1/23. I feel as though this cannot be intended behavior.
I think difficulty not only depends on reviews but only time passed. Correct me if I am wrong @Expertium
Difficulty doesn’t depend on the timing of the review.
So is this really a bug in FSRS, or how would you explain it
We have tried a dozen of methods to include timing of review in the difficulty calculation. But none of them could improve the accuracy.
Sorry, I meant to reply this earlier. It sounds like it may be Due counts on calendar graph can change after midnight · Issue #1025 · ankitects/anki · GitHub
If it’s not specific to the beta, it doesn’t belong here.
Thanks, it was a rounding issue. Fixed.
The report is appreciated. There was no need to post it a second time, however.
It’s something almost no FSRS users will use, and I felt hiding it away is better than keeping all FSRS options together. The alternative would be a dropdown in the FSRS section, but it’s also a bit confusing to have multiple advanced sections.
Presumably you failed it multiple times in the early stages? I think the rate at which difficulty decreases on passing depends on your FSRS parameters, but the FSRS people could confirm that for you.
2 Likes
Don’t forget about this, too
Does this happen all the time? I added a fix for this while working on the cloze handling change but it can probably fail sometimes. I’ve not managed to reproduce it yet though, so a sample card would be appreciated.
I couldn’t reproduce this too. Can you post the field text or send a sample card?
I haven’t been able to reproduce this one yet. Presumably the recommended value is still shown at the end?
Yes, it appears to happen 100% of the time (I’m on Windows 10, qt6, no add-ons). It also does not seem to be influenced by which ‘cloze’ note type I use: I am also able to reproduce the behavior by using the default Cloze note type.
Seems to occur both if I add the cloze deletion by using the keyboard shortcut and the toolbar button, if it matters.
Anki 23.12.1 (testing on this version because I already uninstalled all previous 24.04 releases, but I don’t recall this occurring):
Anki 24.04rc2:
Here is a sample card: https://gofile.io/d/MDAele
2 Likes
I’ll see if I can find a consistent way of reproducing it.
Here you go.
| Klappe | anatomische Projektion | Auskultationsstelle |
|---|---|---|
| {{c21::Aortenklappe}} | {{c22::linker Sternalrand in Höhe des 3. ICR}} | {{c24::2. ICR rechts parasternal}} |
| {{c23::Pulmonalklappe}} | {{c25::Sternalansatz der 3. linken Rippe}} | {{c26::2. ICR links parasternal}} |
| {{c27::Trikuspidalklappe}} | {{c28::rechter Sternalrand in Höhe der 5. Rippe}} | {{c29::4. oder 5. ICR rechts parasternal}} |
| {{c30::Mitralklappe}} | {{c31::Sternalansatz der 4./5. linken Rippe}} | {{c32::4. oder 5. ICR links parasternal bis MCL}} |
| {{c33::Erb-Punkt}} | {{c34::–}} | {{c35::3. ICR links parasternal}} |
1 Like
Presumably you failed it multiple times in the early stages? I think the rate at which difficulty decreases on passing depends on your FSRS parameters, but the FSRS could confirm that for you.
Yeah, I probably did fail it a couple times while still learning it, but shouldn’t FSRS know all those times I clicked good it got easier? You can see that this card was never seen since I enabled FSRS, but with a card that has been, pressing 3 or 4 usually decreases the difficulty.
Difficulty doesn’t depend on the timing of the review.
What does it depend on? Also, if I’m setting it to only take into account reviews after 7/1/23, shouldn’t it only have seen 3 as the worst/initial review and have set the difficulty to 50% like it does for any new card I have that I press 3 on?
It only depends on the grade aka which button you press.
As for ignoring reviews before a certain date and how it affects difficulty, idk, ask @L.M.Sherlock
It depends on the FSRS parameters. In FSRS, w[7] controls the mean reversion of difficulty to avoid ease hell. If this parameter is very small, pressing 3 usually decrease the difficulty a little.
And FSRS only considers the first review for each card in each day. If you press again in the first time you see the card today, the difficulty will not change even if you press easy in the following reviews of the same day.
The problem description sounds very similar indeed, but that issue in GitHub was closed as completed a year ago, whereas this is occurring (recurring?) in the latest 24.04rc2 (b84fce4). Can it be un-closed, or reported anew?
PS,
I’ve edited my original message to give a simple one-paragraph description. The anomaly happens whenever the current time of day falls precisely within the one-hour time interval immediately after Anki’s “next day starts at” changeover hour.
I’ve reopened it: Due counts on calendar graph can change after midnight · Issue #1025 · ankitects/anki · GitHub
THAAAANK YOUUUUUUUUUUUUUUUUUUUU ANKI DEVELOPERS
CURRENTLY USING 24.04 BETA version
MY issue of “STUCKING AT PROCESSING” got resolved now in this update
after VERSION 2.1.65 , now it get resolved
now happily i am using ANKI without any huslte.
many times i raised that issue , but finally this 24.04 update fix all the issue
THANKYOU SO MUCH , you save me .
Have the same issue