I actually think ignoring the first four or so reviews would definitely filter out the beginner items.
8 Likes
OMG love it!! The new stats page is so much better!
4 Likes
I actually just counted, and I apparently have 455 “troubled grammar” points, haha. I’d say that 75% accuracy is definitely far too high of a threshold.
I think it’s important to remember that forgetting (i.e. missing) reviews is an intended part of SRS. So, our accuracy shouldn’t be viewed as if it were a test score, especially since reviewing each grammar point 12 times over the course of a year (acing every single review) would be less beneficial to one’s long-term memory than missing lots of reviews and reinforcing them with extra (and more frequent!) reviews until the grammar finally starts to feel natural.
7 Likes
I think 75% is fine but if you’ve got a grammar point right the last 2/3 reviews in a row you probably know it well enough for it to not be troubled.
For instance I had trouble with amari months ago but now I know it well and have passed all its reviews since then, yet it still appears on troubled grammar.
7 Likes
agree completely…I struggle to remember just about everything…
I have 114 “troubled grammar” points and I’ve only unlocked 162…geez.
My average accuracy
I’m not great at WK either
Some of us are slower to learn and/or just don’t have great memories…
the stats need some tweaking … what would be more useful for me … rank the troubled grammar in terms of times wrong since last review or some other mathematical way of ranking them so they can be sorted or whatever so it’s easier to figure out the ones that I’m really struggling with and which ones I don’t feel like I’m struggling with anymore…also which ones are new … some way to filter/sort them …etc…
5 Likes
@Pushindawood You may not want to use a formula this complicated (as it could be hard for users to follow), but this is what I use in my leech script on WaniKani:
(incorrect / currentStreak^1.5) > 1
(If your system allows the current streak to be 0 instead of always a minimum of 1 you’ll need to do something about the divide by zero.)
I actually use >= 1, but this has the downside that a single wrong answer makes that item a leech, which seems strict (WaniKani always gives you a streak of at least 1, so getting the first answer wrong means 1 / 1^1.5 = 1.
> 1 means it requires at least two in a row wrong at first (or some other combination of right/wrong later) to make the item a leech. Plus, by using currentStreak, you’ll consider how the user is doing right now and not just the past.
For example:
With your formula, any combination of 2 wrong and 3 right will make the item a leech. With my formula, 2 right then 2 wrong then 1 right is very different from 2 wrong then 3 right. The former would have a score of 2, which is a leech. The latter would have a score of 0.38, which is not.
It gives the user the benefit of the doubt that if they got an item right more recently, they’re more likely to get them right again. Of course, this only goes so far, and if you got an item wrong a bunch before, it could stay a leech despite a long current streak. And naturally it makes sense to ignore “burned” items and never consider them leeches.
P.S. I wish you had the math plugin installed so I could use Latex.
11 Likes
Just came here to say, thanks for the updates, and I do agree with the others saying that the current way of calculating “troubled” points is too strict. I have 274 troubled points (out of 492 total).
I like @seanblue’s suggestion, plus perhaps some of the ideas about creating a cutoff of “worst 20” or something like that.
5 Likes
I am completely new to Bunpro and am loving it.
I just wanted to add how much I love the bookmarks (had no idea it was a new feature, since I’m new but I am getting addicted to them already!) and actually came to ask about having them available during reviews, but I see that’s already been covered in conversation, which makes me even more happy. I am really encouraged by all that you are doing. A lot of things in this update already has me hooked. Thanks for the hard work.
4 Likes
Oh, speaking of available during reviews… I’d like to be able to copy both English and Japanese during reviews. I usually need to copy sentences right after failing a review and I want to post them on the forums with a question. Right now I use a script, but it would be nice to have native support there.
7 Likes
Very good update! Thanks
2 Likes
Perhaps just let us set the percentage ourselves? With so many set to troubled, I don’t believe I’ll ever take a look at them. I think being able to set it to something like 50% (so you know which ones really really need checked) would be helpful.
4 Likes
I don’t really mind the cut-off being 75%, because I feel like if you get it wrong 1 in 4 times, then you’re already on the edge of not knowing it that well, but I agree maybe this should only be applied after a certain number of initial reviews so that you’ve gotten past the ‘I just started learning this’ part and it looks more and more like a leech. And then maybe, in addition, weigh newer reviews more heavily, as @seanblue suggests.
However, it would be nice, like @d11 and @ApolloFortyNine mention, to be able to tell which are the worst ones. However, I’d personally still want to be able to see all of the ones below the threshold.
May I suggest that instead of only showing some of them, you could add the percentages for each point in the corner of the box for that point in the troubled section?
This way I could just scan for the lowest percentages (or even sort?) and work from there if I want, but I can also gauge my overall performance a bit better because I can see more than just the worst x points.
4 Likes
It’s easily possible to get a particularly hard grammar point stuck in ‘low percentage hell’. If you had missed a point 8 times in a row, and then got it right 12 times in a row, now it would be eliminated from all further reviews, it would still be listed as ‘troubled’.
So at least adding a less than SRS level 8 would make sense to me. If you hit SRS 8 then you were able to recall after 2 week, so it’s likely not ‘troubled’ any more. I have at least one point that is on SRS 9 and still listed as troubled, and looking at it again now, it certainly isn’t trouble any more.
The issue seems to be that if you miss a card 3 times while learning, it would take over a month of getting it right every time to drop off the troubled list. And getting a grammar point wrong 3 times while learning isn’t that hard to do.
4 Likes
Hmm, yeah, I see what you mean. I think adding thresholds would be good, both minimum (e.g. only applied after you’ve seen it x number of times, avoiding your last example) and maximum (e.g. once you’ve gotten it to SRS level x, it goes away no matter percentage so far, avoiding your first example).
But I still think most issues with grammar points getting stuck as troubled even when you’ve improved on them can be eliminated by weighing newer reviews more heavily like suggested above. I think this more accurately reflects your actual performance while still keeping the troubled grammar feature, which I really like. Actually, this would probably make the whole feature even more useful in another way, too: With newer reviews being weighted more heavily would also mean that something you knew before but are starting to slip up on will be flagged more quickly.
3 Likes
Did you change the troubled grammar calculation? Cause yesterday I had 52 items across n5/4/3, and today I only have 1.
Honestly I’d prefer the old calculation, cause when I was doing that cram session, I only got 60% right, so obviously those are problem points and should still be on my list.
I just started trying it out a couple days ago and was quite happy with the results. But the way it is now… it’s not really of any use to me anymore.
So if you could change it to give us control over what it shows please.
3 Likes
The funny thing is that I would prefer the current settings since before the hotfix I had like 150+ troubled points, now I have only 35 which is far more precise and useful.
Indeed, it looks like the only solution that would make everyone happy is the implementation of a custom troubled threshold setting, though it may be hard to implement. Well, the more customization we have the better since everyone has a different way of learning.
2 Likes
I definitely like the former calculation better. The new one is quite… uselessI think they’re removing the first attempts for the first four SRS levels rather than just nit showing troubled grammar when it hasn’t left the first four levels yet.
2 Likes
All of mine are past first 4 srs levels, so it’s not just that.
2 Likes
I think there should be a way to prioritize one level over another when selecting multiple JLPT levels, like getting mostly N4 with some N5 ones sprinkled throughout, etc. That would be useful