Lots of questions, and do they even matter?

(I am not thinking clearly, so this may not be very coherent or useful.)

What does “Optimize All Presets” evaluate with? What is “The FSRS parameters currently appear to be optimal.” about?

When clicking Optimize, I suppose that it evaluates the parameters for the same cards it optimizes on, that is, the same query as if I clicked Evaluate.
I guess the same thing happens when I click “Optimize All Presets”.

This means that if I choose a bad query, a query for the wrong cards (or a bad query that erroneously finds the wrong cards Anki 24.10 Release candidate - #64 by Aleksej), or the queried cards change, then “Optimize All Presets” can result in worse parameters, and since I have tens of presets, I will not know.
That could be silly and my fault, and I just haven’t had enough sleep, so I keep trying to post this. But I specified a query for a different deck because it was optimal for certain cards of this deck (perhaps based on date or interval length), and since there are no notes, I won’t know why I have exactly that query.

I guess using a separate preset/query for difficult cards makes them seem easier, and using a separate preset for easy cards makes them seem more difficult. So for finding leeches, it is better not to have a separate preset for difficult unimportant cards or easy important cards. But RMSE can be twice as low with a separate preset. Is the preset worth it with 3000 cards and 21000 reviews?

If I prioritize cards that are important, easy or retrievable, making difficult unimportant cards more difficult will make me abandon them, which is what should happen. And since there is a backlog, the few % of RMSE don’t matter.

But then the minimum recommended retention can differ 3-7 % for different unimportant decks.

I think the search you set in each preset is used.

If log-loss increases after optimising then we keep the old parameters and show this pop-up.

I understood the using preset:presetA in Preset B but not what you said after it. Clarify?

I also don’t think you should be selecting a different preset for one preset’s optimisation unless you have a really great reason.

What does “Optimize All Presets” evaluate with?

Each preset individually. It’s the same as clicking “Optimize” manually on each preset, one by one. You run “Optimize” on one preset, then on the next one, etc.

What is “The FSRS parameters currently appear to be optimal.” about?

Sometimes the optimizer cannot find better parameters than your current ones, so it just keeps the current ones.

then “Optimize All Presets” can result in worse parameters, and since I have tens of presets, I will not know.

Yes. Unfortunately, there are no hints that your search query is incorrect. I’ve mentioned this before, but it doesn’t seem like devs will do anything about it.

By “note” I mean specifying a reason or other info, similarly to deck description.
I had set “Ignore cards reviewed before” in some presets and forgot about it. But that setting means I don’t trust my old ratings to mean the same as later ratings. I don’t know why I chose the particular year.
Also, should I optimize on all cards, or would the extremely easy cards with 10 year intervals skew something? Did I choose a query because of a 0.01% decrease in RMSE? Or maybe I was just testing something and forgot?

When a small amount of cards has a different priority, but parameters from a larger amount fit it better.


What got me worried is, I wanted to merge some decks by replacing them with tags, and presets add perceived complexity and are the only reason to have different unimportant decks when I have a backlog and do not review cards by language. The minimum recommended retention currently varies between 77% and 86%. I guess presets with the same parameters would have the same MRR.

You just need to show them how it is a significant problem.

IIRC, dae’s last argument was that if the search query is incorrect, they would see 0 reviews after optimizing which would alert them that something is wrong.

However, when optimizing all presets, the user wouldn’t see that. So, we need a better way to suggest that the search query might be incorrect. @dae

(This is assuming that custom search queries are saved and used when optimizing all presets. The default search query can’t be incorrect anyway.)

2 Likes

I think it’s better done as a yellow warning message when the user inputs the search term rather than when optimising.