I recently started using the “Interleave Lessons” feature. I like it, but I think the way it rounds can and should be improved to both increase interleaving and to give a slight preference to items necessary for leveling up.
I’ll use my lesson batch from today as an example. Here is what was in my lesson queue this morning:
I had a total of 89 lessons in my queue, consisting of 6 radicals, 17 kanji, and 66 vocabulary words. I requested an interleaved batch of 10 items, and so if you do the math, a perfectly even mix of all three item types would be 0.67 radicals, 1.91 kanji, and 7.41 vocabulary words. Of course, lessons are indivisible units, so some rounding is necessary here. 7 vocab and 1 kanji is a given; what the last two items are depends on how we round.
In this specific situation, simple rounding to the nearest whole number would result in 1 radical, 2 kanji, and 7 vocab. However, this is the batch Wanikani gave me:
8 vocab, 2 kanji, and no radicals whatsoever. Hmm. Giving me one radical before an eighth vocabulary item in this situation would (a) produce a mix whose proportions match the input set better; (b) produce greater interleaving, which is the whole point of the feature; and (c) prioritize an item that is more important for leveling up.
I recognize that “round to the nearest” can sometimes result in a batch that is one item bigger or smaller than requested, so it’s not an acceptable solution. Nonetheless, there is room for improvement in your rounding logic. I have two proposed solutions. The first is to always round so that kanji are preferred over vocab, and radicals are preferred over either. The second is to add items one at a time, always adding an item from the group that is currently the most underrepresented. I’m happy to explain either solution in more depth, or even just write the logic for you.
Not sure if this is helpful, but after around level 20 I always manually chose my subjects for the day. E.g. if I knew I wanted a 10 day level, pick 1/7 of the total kanji and fill the rest in with vocab (while still prioritizing radicals above everything else). Letting the app pick for me led to 2-3 day dead zones where I had no lessons at the end of every level.
I just started interleaving lessons, so I appreciate the warning about dead zones.
Naturally, the obvious solution for me as a user is to interleave manually, or to not interleave at all. I would bet there is a–may the Crabigator forgive me for uttering this vile phrase–user script that fixes this issue also. Still, I figured I would at least try to convince the WK devs to fix this on their own first.
I suspect there are quite a lot of options, making it difficult to make an automatic option that satisfies everyone. It would be nice a “visually close items in the same batch” option exists though. I started picking items manually around Lvl 3, due to having a quite specific ordering;
- For Radicals look for similarities, and take them together to allow me to focus on what is different. I try to do Radicals first that also look similar to previous levels but that is a lower priority.
- For Kanji basically the same as for Radicals. I always do all the Radicals first an then start on the Kanji.
- Vocab I always choose items that have the same Kanji or similar ‘vibe’, as an example, today was " I choose everything with the year Kanji in it", which was about 3 items.
The amount of items by default I am doing is 3 Radicals or 3 Kanji. And if I feel like I can handle it I take Vocab next to that. I also have the rule for myself that I have to schedule all Vocabs on the previous level before starting the Kanji.
I am not aware of user scripts that do this, but I have not been around on the forum for that long, so someone around longer might know. 
I suspect there are quite a lot of options, making it difficult to make an automatic option that satisfies everyone.
This reads as an excuse for them to do nothing. I don’t propose that they try to create a million options to try to please everybody; at the end of the day, you either install (or write!) a userscript to do what you want, or just manually select your preferred mix of lessons every day.
Nonetheless, I think my proposed change is worth advocating for. There are a few reasons for this. First, I would argue nobody prefers the current rounding behavior; I would wager it is the way it is because the programmer who implemented the rounding behavior simply didn’t put much thought into it. My proposed change would create behavior that almost everybody would either prefer, or be indifferent to, and that makes it worth doing.
Second, the philosophy of WaniKani seems to be “we’re going to decide how you do things, and you can either accept it or go somewhere else.” That may sound like a criticism, but it’s actually a perfectly fine philosophy. Successfully employing that philosophy, however, requires attention to detail and making good decisions almost all of the time. The point of this thread is to bring attention to an overlooked detail (that is nonetheless worthy of fixing.)
It would be nice a “visually close items in the same batch” option exists though.
A perfectly fine suggestion; I just think it’s distinct enough from my own suggestion that it ought to be in its own thread rather than piggybacking on my unrelated suggestion.
Making this thread got me thinking, and I think there’s an even better way to do things.
“Interleave lessons” should give you an even mix of current-level radicals and kanji and previous-level vocabulary. (In the event that there are stragglers from previous levels due to WK moving items around between levels or whatever other reason, include them as well.) Other items are pulled only if there aren’t enough to fill the batch.
This scheme is motivated by the observation that users frequently end up with two levels’ worth of vocabulary in their lesson queue, but one level’s worth of radicals and kanji. This causes vocab to be overrepresented in interleaved batches and therefore unduly prioritized over the items you actually need to level up.
Again, I think this proposal represents a clear and obvious improvement over the current scheme to the point where it’s worth replacing it entirely.
