Didn’t get it. How is the difference small as shown in your comparison back in 2023 2.5y to 15y is about 7x for one ‘again’, and some comparisons in the topic show a difference that can lead to 60% more reviews for 4 ‘again’ (first time seeing the card).
Does changing the difficulty (multiplier) parameter 8 have a bad impact in the long-term, as it has worked well for me until now?
I used FSRS, but got a lot of reviews per day than usual until I knew this FSRS affected learning cards. And I believe this is a general problem for anyone using a good premade deck, you can go back to examples I mentioned in the topic.
This add-on doesn’t fix the problem, it prevents you from learning the cards.
the point is take your time to learn the card, fail as you like without worry of FSRS.
The community who decides if this change is necessary or not, whatever this FSRS problem, Anki problem
If the cards appear too often, change the “Desired retention”. Your examples don’t make sense since they don’t use your FSRS parameters.
In this case, 1111 is not the same as answering 4 times again in one day.
If I’m not mistaken, 1111 means that you have answered the card “again” for 4 days in a row.
You don’t have to worry. If the number of failures at the learning stage does not affect the further forgetting of memory, then FSRS will take this into account.
e.g.
1.Review the card (but for any reason, you forget what you learned before Anki)
2.Multiple ‘again’ ratings
3.high D, low S
Changing the desired retention won’t solve the problem either, as the card has high D and low S (high review per day, even if you rated it good multiple times later) for the life time of this card.
To solve the problem you must reset the card as LM Sherlock said:
Here another example showing this I answered 4 again on the same day(See the dates) and the difficulty jumped to 98% (so hard to back the difficulty to normal value without reset it)
What do you mean by ‘not affect the further forgetting of memory’?
All I know is that multiple failures = high D and low S, = high reviews for the lifetime of this card, whatever you do, you simply can’t change it
To accurately estimate the DSR values, FSRS analyzes the user’s review history and uses machine learning to calculate parameters that provide the best fit to the review history. The most recent version of FSRS, FSRS-5, uses 19 parameters in the formulas for D and S (the formula for retrievability doesn’t require any parameters). If you are interested in the details, you can read the following wiki pages: The Algorithm and The mechanism of optimization. If a user doesn’t have enough reviews yet, the default parameters are used instead. They have been found by running the FSRS optimizer on several hundred million reviews from ~10k users. Even with the default parameters, FSRS is better than Anki’s default algorithm.
You are using FSRS parameters based on data from other users. If you stop deleting reviews and provide data for FSRS, it may adjust to your memory and your habits.
The parameter has nothing to do with this problem, and I don’t use anyone parameters. These are my parameters for all my card reviews, just changed the difficulty parameter to 0.0250 so the card difficulty can return to normal after some ‘good’.
And whatever your parameters are, after 4 “again,” your difficulty will hugely increase, and good luck decreasing it. And 4 “again” is not a big number when you’re learning the card for the first time. For harder cards, you may get 10+ “again”.
The problem is when we use Anki for learning (seeing the cards for the first time), and this is mandatory for anyone using pre-made decks, they will learn inside Anki and have no other choice.
I don’t fully understand how DSR is estimated, but in my experience, if you get multiple “again”, you will get a high DSR, and it will decrease slowly with “good”, as shown in the comparison in the topic or comparison by FSRS.
The parameters have a very important role in this problem. Even if all your cards have 100% difficulty (and it remains 100% forever), their intervals can still grow very fast if you have appropriate parameters.
But, for this, you have to provide FSRS with the data that “Hey FSRS. Look, I rated this card Again 4 times on the first day but I remember it reasonably well after that. So, I should see this card less often.” Eventually, FSRS will adapt and provide you with more optimal parameters and more sensible intervals. But, if you reset your cards, FSRS never gets the data it needs to give you proper intervals.
I already stopped deleting my reviewed/reset my cards and just change the difficulty parameter instead of resetting manually and still trying it, but seems good until now.
Though I usually don’t advice changing the parameters manually, I think that in your case, it is a good way to adjust until FSRS adapts to your learning/rating habits.
I’m not sure, but for sure you will need a lot of “good” (high reviews per day) to get high S, so it’s better to prevent FSRS from seeing multiple “again” or change the difficulty multiplier parameters to decrease the difficulty in the cards so you can have high intervals and low reviews per day.
I hope so.
The 7x case only matters when you use the default parameters. If you optimize FSRS on your review logs, it will be different.
If you see the formula, the stability after a review depends on many things. For an example, if w8 (9th parameter) is high enough, the next stability after a single review will be high even if the difficulty was high.
Also, remember that difficulty in FSRS is a relative value. If all of your cards have 100% difficulty, it is not much different from having all cards with 0% difficulty (provided that all parameters are well optimized).
Wow, I didn’t know this, All of my cards has 90%+ difficulty and I thought this was bad thing. so tried to decrease it.
Maybe some tests to confirm, but I think if FSRS saw only the review history, without learning steps (a lot of ‘again’), it will give better results (think of it like I studied the topic before using Anki) and no need to wait until parameters be optimized.
I also think the first review has a huge impact on DSR and can’t easily change without a reset even if the parameters well optimized.
While it is true that having all cards at D=100% and having all cards at D=0% is essentially the same because FSRS can just adjust a constant in the formulas, I don’t think it’s correct to say that D is a relative value.
A card’s D does not depend on D of other cards. If it did, then yes, it would be a relative value.
Thing is, your parameters are getting like this BECAUSE it is how FSRS was able to optimize the cost function.
Even if changing it so D would feel more intuitive, and getting back to “normal” after fewer Good, would result in worst cost function. If it could result to better cost function, the optimizer would have done that already.
Also, I don’t really see the problem, if you can’t remember something 4 times in a row at a few seconds/minutes interval, it definitely means your stability is also in a magnitude of a few seconds/minutes.
A lot of people bruteforce their way through reviews, and I’ve done the error for months in the past, but this is just what you get : High workload and very pessimistic function.
Be cautious to really learn/memorize the back WELL (spent time with it, connect it to others concepts …) before pressing “Again” and suddenly remembering it won’t require you >2 tries
I got about 1.25% higher in the cost function for changing the difficulty parameter to 0.0250. I believe this is a small number for in return I get fewer reviews per day and in the long term reviews, as you said, getting back to 'normal' after fewer Good, as I go with lower cards reviews over high accuracy.
And if I always reset cards or use FSRS through custom scheduling (prevent FSRS from seeing first ‘again’ reviews until I learn the card), I will get both low reviews per day and high accuracy.
It’s more about learning, not remembering, as learning new concepts isn’t about instant recall. It’s about building familiarity. You have to see the concept multiple times (multiple ‘again’) to understand it. After you understand the concept, you can remember it easily. Think of it like getting to know a new person you need to interact with them a few times before you really get them.
So there’s no need for FSRS to take effect (seeing first ‘again’ reviews) until I understand the card.
My average is about 29 seconds per card, so I’m not doing brute force (if you mean reviewing cards quickly).
Even one ‘again’ review will tell FSRS this leech card and you will get High D, low S for the lifetime of this card, and for premade decks maybe the average is 4 ‘again’ reviews (about 98% D, 0.02 S) until you get the card, especially if they are language decks.
