Beta 4 is now available: Releases · ankitects/anki · GitHub
3 Likes
In beta 4 in the “Compute optimal retention” section there is a new parameter: Loss aversion.
What is it exactly and how to choose, what value to set? I am getting different retention values, from 0.88 to 0.93, depending on what I choose there.
@L.M.Sherlock
1 Like
It’s a multiplier that determines how much you don’t want to spend time on a card only to press “Again”. @L.M.Sherlock I really don’t think that this should be controlled by the user, it should be hidden.
Ideally, Anki should be user-friendly and not overwhelm the user with a dozen of settings while still providing the benefits of FSRS. I’m afraid that too many users will give up on FSRS just because of all the complicated looking stuff in the Advanced section. Other people in this thread have expressed this concern as well. If desired retention is the only setting (with the Beginner layout), it won’t scare off new users. Unfortunately, with the current approach, FSRS will likely be a feature that only a small number of advanced users use.
6 Likes
So far I haven’t found any available shortcuts in this mode, and I think it would be very practical if there are any.
E.g. ctrl-a select all, Del delete, some modes switching keys(something like v drag, c create image cloze) and so on.
Loss aversion means how many times worse you think it is to forget something compared to the reward of recalling it.
If you set loss aversion to 1, the optimizer will give you the retention with the fastest learning speed in theory, but its value might be low, which could make you less motivated to review. The default for loss aversion is 2.5. It doesn’t maximize your learning speed, but it gives you a higher chance to remember stuff, which can help keep you motivated to learn. It’s a subjective setting.
6 Likes
It’s not a bad idea, I just don’t think it should be controlled by the user. It adds yet another setting that only a handful of advanced users will understand.
2 Likes
It’s not a bad idea, I just don’t think it should be controlled by the user. It adds yet another setting that only a handful of advanced users will understand.
I second that. Even if I am considering myself quite advanced Anki and FSRS user and now I know what is a reasoning behind this parameter I still have doubts how to exactly calculate the “how many times worse you think it is to forget something compared to the reward of recalling it”. Especially, that you can set it with 2 decimal places precision.
5 Likes
It’s very hard to calculate the emotional/motivational cost of review. In the old version, the optimizer only considers the time cost of review. But the emotional/motivational cost is also important here. I don’t figure out how to set it automatically. So I leave it to the user.
@L.M.Sherlock I have some conceptual doubts regarding the ‘Desired Retention’ feature in the SRFS4. Why does it suggest a desired retention rate of 0.85 when the default is set at 0.90? What’s the rationale behind suggesting a lower retention rate, implying a higher forgetting rate?
The greatest overall knowledge acquisition rate is obtained for the forgetting index of about 20-30%. This results from the trade-off between reducing the repetition workload and increasing the relearning workload as the forgetting index progresses upward. In other words, high values of the forgetting index result in longer intervals, but the gain is offset by an additional workload coming from a greater number of forgotten items that have to be relearned.
https://supermemo.guru/wiki/First_steps_of_SuperMemo#Optimum_forgetting_index
5 Likes
“Compute optimal retention” tries to find retention that maximizes the number of cards you will learn and remember (strictly speaking, it maximizes the sum of all retrievabilities), given certain time constraints. IMO, the biggest problem is that just by changing “Days to simulate”, you can make it output pretty much any number. Suggested retention doesn’t converge to some constant as you increase “Days to simulate”, quite the opposite, actually. Personally, I don’t use that feature.
2 Likes
Simulator for longer “Days to simulate” tends to give higher retention. Which is logical, because if you give yourself more time to learn you will learn more cards, and with higher target retention you will know the material better.
I also don’t use it currently. Mainly because it assumes, that you are starting with a deck of all new cards, so simulation doesn’t consider, that I have already learned quite a lot of cards in a deck.
1 Like
I think “Predicted delay that you have a 90% chance of remembering.” is confusing, I suggest changing it to “A period of time in which 10% of cards are likely to be forgotten.” Also, change “Average interval” below the chart to “Average stability”.
Also, for retrievability, change “How likely you are to remember.” to “How likely you are to recall a card today.”
IMHO I like the current more. Because your version suggest, that it will be a single time, after which one will forget 10% of their learned cards. And Stability is individual for each card.
Maybe it’s because I’m not a native English speaker, but I find the sentence “Predicted delay that you have a 90% chance of remembering.” to be borderline nonsensical.
How about “A period of time in which a card has a 10% chance of being forgotten.”?
Or “The time required for the probability of recalling a card to decrease to 90%”, but that’s pretty long.
3 Likes
I apologize if I’m being too annoying, but both in this thread and on Reddit people agree that FSRS doesn’t have to be complicated to use, and that it’s entirely possible to make it accessible to new users if they only need to adjust desired retention. I believe that simplicity for new users should be top priority.
6 Likes
Even though I agree that hiding settings is good for new people. I think it’s far more important for fsrs to be on by default once it get’s to a certain point, and the weight to be recalculatd periodically. Because if everything works flawlessly there won’t be a need for beginners to even open the settings menu and be frigtened by the settings.
FSRS being the default algorithm would be good, but it won’t happen immediately after the stable version is released, and probably won’t happen for a while.
Only half joking, but a questionnaire about how demotivating the user finds it when they get an answer wrong, versus how optimally they want to minimize reviews over long time horizons, might be a decent solution. (Or a tooltip-ized version of it for the Loss Aversion field)
One of the abstruse parameters deep within the bowels of the algorithm’s optimizer’s algorithm doesn’t need to have perfect UX. The average user will probably just set the desired retention to the optimized value, and maybe the simple guidance version can be “Show me on the slider how much you hate getting things wrong versus learning efficiently.”
I’m glad the field is exposed to the user, and I think hiding it (and setting it to anything other than 1) would be doing users a disservice, as it’s something I personally want to tune and play with. I think maximal algorithmic control should be available to users in general, where possible.