Good ratio for a "leech"


#1

I’ve seen that we define “leeches” as elements with a high ratio of wrong/current_streak. What is a good threshold to identify an element as a leech?


#2

One user who created a tool for personal use and then shared it made that definition. I’m not sure they claimed it was “the” definition, though maybe it’s been refined since I last saw the discussion.


#3

I saw it in a few places so I assumed it was an accepted standard within the community :confused:. Guess I shouldn’t assume like that…


#4

Well, perhaps it came to be accepted since everyone liked that website. But it does have some flaws, I think, but I can’t necessarily propose a better formula off the top of my head.


#5

That’s fair enough. I honestly did not bother looking for a better formula since it seemed accepted. But perhaps I will. In the meanwhile, I will implement this definition though. gets something out there in the time being :slight_smile:.


#6

It doesn’t have to be complicated – Anki defines leech simply as a set number of lapses on a card that is in review mode (in WK-speak “review mode” is kind of like something being guru or above.)

I think it defaults to 8 – once you miss a card 8 times, it notifies you, and then tags or suspends the card.


#7

Are you writing a userscript? Or is this just about your own personal study?


#8

It’s for this: Android app WaniKanji

I think the Anki definition is too limited in that something that was a leech 40 years ago would still qualify if it ever went past Guru and then came back down to apprentice. I think something like the one offered in the community userscripts offer a better approximation, though I cannot say that I think the formula is “correct” just that it appears to me to be better. I think whether max_streak or current_streak is used is also subject to some debate an opinion.


#9

I’m not following – if something was marked “leech” 40 years ago, it’s a leech forever until you reset it. Then it would take 8 lapses to become a leech again.

I compared “review mode” in Anki to Guru+ in WK for the sake of brevity in my explanation, but they are different concepts. In Anki, when you are “learning” an item or “relearning” an item and you get it wrong, those incorrect answers don’t count toward the leech counter.

Also Anki deals with SRS differently than WK, allows you to make your own SRS settings, and doesn’t have the concept of burning anything.

I can say from my own personal experience that it would be very helpful to be notified by wanikani once I’ve entered a certain number of incorrect answers, and I’d like to be able to set that number myself. If you have the ability to show that data I think it would be super useful and easier to understand for the average user than a ratio.


#10

Ah, got it. It’s clearer now.
What would a typical number you’d like to be notified at be? I know you said you’d like to set it yourself, but at no point in the app do I ask for user input and for the sake off consistency of experience, I would like to keep it that way as much as possible.


#11

I really think it’s highly personal. What if there were multiple views?

  • list of items answered incorrectly on either reading/meaning 5 times or more
  • same, 10 times or more
  • same, 20 times or more

And then the same lists but for lapses on just meaning or just reading. Very often the “leech” part is just one or the other… you miss a meaning every third time but have a string of 40 right answers for the reading, for example.


#12

This is doable. I’m putting it on my list. Once I have the current leech thing I’m almost done with sorted, I’ll figure out the app flow for your suggestion. I quite like the idea of giving people the options.


#13

Cool! I wish I had android… :neutral_face:


#14

Oh no! I was hoping you would test it :scream:


#15

So just to be sure, the feature you are suggesting is:
Show items that 1. have X incorrect reading/meaning and 2. are Guru or above? or just 1?
I’ll implement 1. for now.


#16

Done :slight_smile:.
Now leeches are in the app:

  • Leech score >=2
  • Leech score >=3
  • Meaning or Reading >=5
  • Meaning or Reading >=10
  • Meaning or Reading >=20

#17

Items that have been missed on a review where they were guru or above more than a certain number of times are maybe traditionally more insidiously “leech”-ey but either stat could be useful, for sure.


#18

Given that the API doesn’t enable us to know how many times an item was wrong after being guru’d, the additional constraint would not mean that much, unfortunately, so I’ll leave it as number of incorrect regardless of srs level.
I know you don’t have Android, but feel free to checkpoints screenshots on the relevant page if out feel like it :slight_smile:


#19

And thanks for the help! :slight_smile:


#20

How about judging solely based on the number of occurrences that the item is in the system for an unburned card? If I’ve seen the card 30-50 times and it is still un-burned, it is probably a leech.