A special case where "Ignore reviews before" won't work.md

[Expertium]

No description provided.

[user1823]

Would it be too much to ask people to hit "forget" on all their cards to generate more learning steps if changing their review habits?

Ideally, the users should NOT do this because the estimates of initial stability won't be accurate in that case (resetting the card in Anki doesn't delete the memories in the brain).

The ideal solution is that they continue to learn similar new cards. But, I know that this is not possible in every case. I don't know what's the best solution.

Originally posted by @user1823 in ankitects/anki#2922 (comment)

[user1823]

Basically, in such cases, the user has to choose between two options:

1. Not using the ignore reviews before option and optimizing parameters using skewed review history
2. Using Reset and generating parameters with inaccurate initial stabilities. (Remember that the initial stabilities are among the most important parameters.)

I don't think that we should advise choosing one of these two. The option that is better for one user may not be better for another user.

[Expertium]

Yes, but we do have to advise something that isn't "Keep seeing an error that says 0 reviews found"

[user1823]

I really don't know what to advise. Suggestions are welcome.

[Expertium]

I see 3 possibilities. The ones you mentioned + keep using the default parameters. Using the default parameters is probably worse than either of the two options, since the parameters won't be personalized at all. So we could advise users to choose option 1, we could advise users to choose option 2, or we could tell them about both options and let them choose, but that introduces an extra cognitive load, which isn't great. Honestly, I'm not sure myself.
@L-M-Sherlock what do you think?
(option 0: rework this feature so that it actually ignores reviews, not cards)

[user1823]

By the way, there is a fourth option too. But, it is only for advanced users.

This option involves modifying the review history, as I advised a user here: #614 (comment)

[user1823]

(option 0: rework this feature so that it actually ignores reviews, not cards)

This is not possible because the optimizer can't work on incomplete revlogs.

[Expertium]

But memory_state_from_sm2 is a thing

[user1823]

But memory_state_from_sm2 is a thing

See open-spaced-repetition/fsrs-rs#114

[Expertium]

I think the crux of the problem is that we can't quantify which of the 4 options - default parameters, Reset, optimize on bad review history, optimize while using memory_state_from_sm2 - provides the best (well, least bad) result. If we could rank these options somehow, we could say "Well, they're all bad, but this one is less bad, so let's go with the lesser evil".

[Expertium]

@L-M-Sherlock thoughts?

[L-M-Sherlock]

It's impossible to ignores reviews, not cards because reviews are also used to construct the training features.

[Expertium]

So what do we recommend to people who reviewed all cards before the selected date and do not plan to add new cards?

[L-M-Sherlock]

If they misused hard button, I remember that @user1823 provides a solution to modify the database. Maybe I can port it into the helper add-on.

[Expertium]

That's complicated, I wouldn't recommend that to everyone. We need something for an average user.

[L-M-Sherlock]

If I port it into the helper add-on, it will be very simple. Just click a button. And I think this case is rare. Maybe we cannot treat them as an average user.

[Expertium]

Ok

[L-M-Sherlock]

LGTM