You can change the IM for new cards, cards with a few reps, but there is a danger with changing IM for review cards. It should not deviate 20% from the starting IM. And it is unfortunate that Anki does not log any of these.
Here's a rough example, StudentA divides his study into 3 decks:
Low priority deck: 200% IM
Mid priority deck: 100% IM
High priority deck: 55% IM
Suppose a card in LP was studied a couple of times hitting hard and ends up with 1.3 EF and the student moves the card from the LP deck to the HP deck because of an up coming test.
What we have is:
ivl * 1.3 EF * 55% IM = DOUBLE EASE HELL
And suppose a card with a 2.5 EF was moved from the HP deck to the LP deck. What we have is:
ivl * 2.5EF * 200% IM = DOUBLE LAPSE HELL
Each review changes the ease, but when you move a card between decks, it does not adjust the ease according to the new deck options. This make IM unsafe for newbies to use and must be capped or restricted.
Bumping up the IM in increments is good, but there is still a danger of going above the initial IM. So if you have a deck with 2 sibling and their starting IM set to 150%, the IM changes should be capped between 140% - 180% (-10%, +20%). 20% of 150.
I set a limit of 20% because the forgetting index ranges from 3% to 20%, exact numbers differ for each person. Remember the point of changing IM is to handle material overflow, at the cost of retention for low priority materials. If something important, you shouldn't change the IM, the default SM2 is good enough.
It has more to do with how the learning curve is mapped out. I don't modify the IM once the learning process has started as that skews the learning curve. New cards and lapsed cards can be changed, but review cards should only be altered in small adjustments. Hope that makes it clear?
I was under the assumption that we are discussing this for language learning. I am not familiar with anking's work as it targets medical students. (And where is the black guy on the team, is this a south park parody?)
If we map out the numbers, it's close to anki's default settings: very good for test taking, not so much for long term retention. So it's not useful for language learners.
Urgh, I finally found the time to jump into this discussion.
Hm, one of the things I don't get is why one would separate by importance alone. It has to be importance + topic, otherwise you are going to bias your retention rate. Also, the point of my discussion was adding a semi-automatic routine to adapting the IM, not how this workflow can be done manually. Additionally, the way the IM is designed as of now is very impractical for both the workflow, how add-ons could affect it and also contradicts the manual a little bit.
While saving the IM on each card - so it could be accessed on via tag/deck/whatever individually - would solve some problems, changing the IM group/IM settings (IM group = newly defined term for cards which belong together and have the same IM according to my idea) of an array of cards would still be troublesome. But only being able to change the IM per deck is very impractical in my opinion.
Hm, one of the things I don't get is why one would separate by importance alone.
Most experienced users don't. This is done just to illustrate the problem with IM by isolating it into an abstract scenario where the numbers are easier to see and predict. Some users might actually use it this way though... and not necessarily for SRS. I use Anki for my todo list as well.
Your idea of using tags to prioritize cards is pretty good. This is the way SM8 or SM11? decided to adapt. By using categories instead of rep counts to prioritize scheduling reviews. So you might have something there and should spend more time playing with this idea.
Actually, if Arthur Milchior's add-on works decent it would be possible to add a hidden card modifier field to each card which can be overwritten by an add-on via the tags and then still being able to reschedule mobile reviews.
Let's see if I can get my python to the needed fluency to tackle such a task. But the idea is obviously quite simple and the coding is - in theory - not that big of a deal.
So, I had some new ideas how an advanced, yet-SM2 scheduling could look like and wanted to share with you.
An add-on could analyse the performance data based on the following groups:
- decks + (sub)tags
- decks + notetype
- notetype + (sub)tags
The performance data which should be looked into is probably: ease factor (based on performance of mature cards), initial interval (SM users claim that well-formulated items can have up to 21 days as the initial interval, which is IMO the biggest performance difference), lapse new interval.
This could be optimised towards specific criteria, most likely on a per deck basis. (Standard to 90% retention performance).
Based on the performance data for each of those groups, every new card could be classified into those groups: the algorithm will search for the group the card belongs to with the least amount of performance scattering and adjust the lapse interval (when it is not a new card, but a failed review card), the starting ease and the starting learning step/graduation interval.
What do you guys think about this? (ATM it's just some thinking, I am afraid I lack the time/experience to start such a project. But while it will cost a lot of hours, it doesn't actually seem over-ambitious from an implementation standpoint)