-
-
Notifications
You must be signed in to change notification settings - Fork 123
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[Question] Would FSRS work correctly for cards with copied scheduling data? #136
Comments
It will cause problems 😫. In my view, these logs are invalid. |
How serious could it be? |
Here's what I think: I'm curious about what fraction of your cards had their scheduling transfered, I suspect it's a really small number and that would have little effect on your optimized weights. |
FSRS helper needs the first rating (corresponding to the learning type of review) to initiate the rescheduling of the card. |
Ok. But, what would the helper do in the case of these cards? Also, my other two questions remain unanswered.
|
It would cause unknown bugs, because I haven't tested this case. 5% data doesn't affect the parameters too much. And maybe I need to use the add-on and check the effect. |
It's not the responsibility of FSRS. In my opinion, the history of reviews shouldn't be mutable. |
I agree that it's not the responsibility of FSRS. However, for the benefit of the users, I think a small check could be added to look for such cases. If such a case is found, the helper can use only the later parts of the review history to set the difficulty and other factors. These factors would then adjust as more reviews are done.
So, this would mean that there is no need to tweak the optimizer.
I would wait for the results. |
I prefer to skip these cards, because the helper can't create the first rating out of nothing. |
Could you share the deck file with scheduling information with me? It will speed the check. |
Here is the file: Selected Notes.zip Note 1: I have changed the file extension to .zip because GitHub wouldn't allow me to upload .apkg file. Please change the extension to .apkg before using it. Note 2: I have only shared some of those cards which were affected, not my whole collection. Also, I have deliberately replaced the contents of the fields in the note because I don't want to share the contents online. |
Thanks! I will analyze it tomorrow. |
I guess your retention before is higher than 90%. Could you paste the analysis generated by the optimizer here? Like this: |
I noted that the retention is extremely high when the first rating of the |
By the way, I recommend using learning steps shorter than one day. Let FSRS control the long term schedule. |
What should I do then? Should I increase the target retention to something like 96%? |
I will set learning steps shorter than one day when I start using FSRS. These logs are from my previous reviews. |
It depends on your goals. Higher retention or less reviews. You could check the stats given by Anki to see the retention in the past. |
Is this what you are talking about?
I would want similar (if not greater) retention to that achieved by standard Anki algorithm. So, if I use a retention like 95%, would FSRS reduce the number of reviews as compared to standard Anki algorithm? Or would I be fine with using the standard algorithm and not bother about FSRS? |
You could simulate the comparison via the fsrs4anki simulator: https://github.com/open-spaced-repetition/fsrs4anki/blob/main/fsrs4anki_simulator.ipynb |
1 similar comment
You could simulate the comparison via the fsrs4anki simulator: https://github.com/open-spaced-repetition/fsrs4anki/blob/main/fsrs4anki_simulator.ipynb |
So, I simulated the comparison. Here are my results for 94%, 95% and 96% retention respectively. Based on this analysis, I think 95% retention reduces my review time (slightly) as compared to standard Anki algorithm but gives the maximum number of remembered cards (5906). So, I think 95% retention would be best for me. What do you think? |
Sorry for being too inquisitive. I am just trying to ensure that FSRS gives me the most optimal results possible. |
I prefer 94% because its time per remembered card is the least (i.e., efficient). |
This seems good. |
The retention per day doesn't equal the retention of all cards. For example, the retention of mature cards is often lower than young cards in Anki, because Anki gives a longer interval to those mature cards. The lower retention of mature cards could not be included in the daily retention until they were due. So the retention per day could be higher than the average retention of all cards. |
Thanks. I think you should include an explanation of the values generated by the simulator in the simulator's description. This would avoid many such queries from the future users. |
I have a question about whether FSRS would work for me correctly. But, before I ask my question, here is some background:
While working through my cards, I often discover old cards that I think should not be a single card but must be split into multiple cards. Then, I split those cards and based on my assessment of my memory of those answers, I copy the scheduling data from the old card to the new cards using this add-on. However, this add-on doesn't copy the complete review history. Here are the card infos of two such cards:
All the reviews seen in the review history are those that were done after the splitting.
Now, since the Optimizer and the Helper use the complete review history of the card, would they work correctly with these cards?
I don't know how much of my collection consists of such cards, but I guess it should be close to 10%.
The text was updated successfully, but these errors were encountered: