Bunpro's bad SRS algorithm is discouraging

[Halobob]

I haven’t been able to go through all of the post history, but please leave an option to continue using the current SRS system if you add FSRS. Long story short I have found that FSRS doesn’t work well for my particular study habits, and it certainly wouldn’t work for how I like to use Bunpro personally. The existing functioning of the website suit my workflow perfectly.

Just wanted to add that data point. I know giving customers more options can be confusing, you can hide it in some preferences menu, as long as I can keep using it I’ll be happy.

[Jake]

We will 100% keep the default our SRS system as it is the one that best supports our learning philosophy and the way we set content up.

The FSRS implementation will be an opt in setting that users can use if they so choose.

[rustx]

Chiming into say I also prefer the current SRS system you have for grammar, as you’re not receiving the same word (or sentence) each time so FSRS would make it extremely difficult if you get a fast 4 month waiting gap on a sentence you haven’t even seen before.

[lkjaerozhang]

Would the FSRS setting be global or could it be set for only grammar or vocabulary? Given that rote memorization is more effective on vocab than on grammar, it might make sense to only use FSRS on that since that’s what FSRS is optimized for.

[Saffina]

It does seem like an issue, because here you can see i learned something months ago, and im struggling on it, and yet bunpro would rather just lead me long the slow path, almost in a way to just keep me subbed, rather than just to help me learn by showing me the problematic point at sooner intervals, why should it take a month to see something im 50/50 at? i clearly dont know it

If this was Anki, id be seeing this problematic grammar point every day

[scarecrowfield]

It’s not a permanent fix, but I’d do in your case is put all example sentences on ghost manually, it helped me immensely to learn some grammar points

[lashette]

For the time being I’ve decided to use the setting to drop my missed points to beginner 0. All my other SRS methods do this so I figured it’ll help a bit more

[davebp]

@Asher

‘Again’ is the only one that drops the time, while ‘Hard’ generally keeps it around the same. ‘Good’ and ‘Easy’ then have their own separate timers. Implementing this would require a Bunpro user to not only know how FSRS is supposed to work, it would also require them to evaluate for every single sentence the reason why they didn’t get the answer correct. Did they completely forget it, did they have too many better known grammar patterns in their head, did they just make a simple typo?

I think that bunpro should do this, automatically. As you say, there are lots of reasons why people get things wrong, and the platform’s response should vary depending on what happened.

Typos are the obvious example, but if a user always gives the same wrong answer, then it probably reflects a confusion (with a different grammar point).

One common example for me is that I get the てform of a verb wrong, while knowing the correct form of the actual grammar point. This should not penalize the grammar point, but the てform grammar point.

[davebp]

@Jake

We will 100% keep the default our SRS system as it is the one that best supports our learning philosophy and the way we set content up.

The FSRS implementation will be an opt in setting that users can use if they so choose.

I don’t understand. Why bother. You have better things to do that support things you don’t believe in.

[Jake]

I can see the value in having it for plain old JP → EN flashcards that will always be the same information.

[Hf]

Would love to see FSRS for plain vocab Flashcards. I’ve been add vocab cards that I run across in the wild to Bunpro instead of another SRS app, and i think FSRS would enhance that experience.

[Jampy]

Any chance on users getting to try FSRS in Bunpro anytime soon?

[distantflower]

FSRS for Bunpro Please

It’s closer than ever We have a slew of other settings options that it will go live with in the near-ish future. [image]

[Saraph]

The lack of FSRS (for grammar) is currently my biggest gripe with Bunpro. I recently moved my kanji writing from an app that used old SM-2 algorithm to Anki with FSRS and it was such a noticeable improvement. I really hope it can be implemented soon, and not just for vocab like I’ve seen mentioned here.

[Xelbas]

My main observation on SRS is that I get a lot of cards jumping between beginner and adept.
It feels like they get “promoted” too quickly in relation to my real retention.
Now I have tried increasing the SRS strictness to 2. This should hopefully give me more frequent reviews for something that I get wrong and hopefully limit this flapping back and forth in the long term.
Thoughts?
Or shall I just wait for FSRS? Would that solve it?

Thank you

[Saraph]

I’ve been using SRS strictness 2, with ghosts turned off, for a long time now. With the SRS options we have available now I think it’s a good balance between reviewing what needs to be reviewed and not being buried in reviews.

[Xelbas]

Thank you @Saraph . Yes, I turned off ghosts too.

[okayfrog]

@ggw1776

Instead of being drilled on the grammar, I’m drilled on the sentence I got wrong. This, in my mind, is not good. I find myself memorizing sentences instead of grammar.

oh man do I know that feeling all too well.

[Alerean]

@ggw1776 @okayfrog

I have an even worse habit…sometimes I find myself not even memorising sentences but just a keyword of it, or even just the structure/length and I can remember the word Bunpro is looking for. 0 real understanding of the nuance or thinking it out…

[Noxsora]

Bunpro Algorithm shower thought.
So interval “i”
If answer TRUE i=i×1.5
If FALSE{ i=i×2/3
New Ghost }
I’m chill with this… I think that’s basically the bunpro algorithm.
The problem is when the variable stored as i doesn’t match the actual interval. When people review early (cram) or late (backlog).
TSLR time since last review.

… Then what.
If TRUE i= max(TSLRx1.5 or i)
If FALSE i= max (i-TSLR or ix2/3)
New Ghost
So it’s not a worse interval than it currently is, and is more accurate when TSLR and i are different.