Suggestion about the learning steps after analyzing the Anki Dataset

Anki Scheduling
83

I don’t understand at all what you are getting at.

85
  • Have cards not formulated correctly.
  • Not know the material (well enough to “hook” into your brain, at least) before adding to Anki.
  • Why yes, German is not my mother tongue though I doubt that plays a role now after 2 years have passed.
  • I have spent a plenty good amount of time trying to hone my card making skills. I doubt they are optimal at all. I was using a cloze text format, copying chapters into anki and making clozes out of it. Now I am using a Q-A format made by Chatgpt but it seems that the same problem persists. I also have trouble piecing together information as part of a big picture with the new format just as well as the old format.
  • With me having tried a myriad of methods, I have come to accept that I am probably half-ret—ed, or young onset dementia or my memory just doesn’t have the knack for it. Either way, I know that there is a problem and it is showing in my extremely low stability. Woe is me, I guess.
86

I also suspect that the learning steps also has something to do with how the stability if influenced. For example, if I were to have 1s 5m learning steps instead of 1s 3m, I may have different stabilities, but theoretically it shouldnt have any influence since any change in the learning step would influence my retention so it should balance itself out, I suppose…

87

Jarrett, I’m really intrigued by the step stats. I haven’t seen anyone interpret this in a way that makes it actionable for a typical user yet, so I’m trying to understand it myself.

I’ll add the caveat that I’m not considering throwing steps out the window, and I probably won’t be using the new version until Spring. I’m just looking for how a typical user is going to be able to apply this data and improve.

From the description:

Overall Retention (R̅) for the first ratings are shown. Based on this, the average stability for cards after the first rating of the day (Again or Hard or Good or Lapse) is calculated.

  1. Is that average Stability column – (a) the average time it takes a card (with that first rating) to drop to that R̅, (b) the average time it takes a card to drop to 90% Retrievability, or (c) the average time it takes a card to drop to Retrievability that matches my desired retention?

image
image1123×399 41.3 KB

  1. What time-period does the chart pull data from? It doesn’t seem to change when I click from 1mo to 1y to deck life. I was really hoping I’d be able to get new data for comparison after a month of trying new steps based on what I see now.

  2. What set of cards does this pull from? I’m assuming each of my cards is only included in the top 3 rows once – and that it’s pulling the whole preset deck except suspended and New. But the total of cards in the top 3 rows is about 2200 fewer than the number of active cards I have. I know there is some winnowing as FSRS decides what cards it can count for optimization – is that same process happening here? Even so, leaving out 30% of my active collection is an unexpectedly high figure.

88

The stability in summary is the average time it takes a card to drop to 90%.

It doesn’t consider time-period. Maybe I need to add it.

It counts the cards which have two or more review logs at least.

Edit: the time period will be taken into account after this PR is merged. You can check the difference.

1 Like
89

Let me try:

  1. As we know, the first learning step is applied to cards whose first rating is again. And the second learning step is applied to cards whose first rating is good. The average of the 1st and 2nd steps is applied to card whose first rating is hard.
  2. The stability of Again is recommended setting as the 1st learning step if your desired retention is 90%. And if the stability of Good is less than 0.5d and you are using the default desired retention, the 2nd learning step can be set as the stability of Good.
  3. If the stability of Hard is less than the average of stabilities of Again and Good, I recommend lowering the 2nd learning step to 2x stability of Hard to ensure you have 90% retention on cards whose first rating is hard.
3 Likes
90

Okay, that’s really helpful!

In that case, since I’m aiming for 92% retention, I would interpret the chart as telling me [updating based on your last message] –

  • Shorten my 1st learning step to under 4.5m (currently 10m, with a 20m learn-ahead limit) to increase retention for first-Again cards.

  • I can’t independently control Hard, so that will be ~7m (currently 20m) for first-Hard cards. Those aren’t having any issues until something in the 30-90m range, but I’m not worried about a few extra reps with them.

  • It looks like I don’t need a 2nd learning step at all (currently 30m), since my retention on my first-Good cards is strong even a couple days later.

  • The other option would be to set a 2nd learning step of ~3.5h to make sure my first-Hard cards are seen at the right time. But since my study sessions don’t necessarily take that long, that would keep my first-Good cards from graduating most days, and there are 3x as many of those as their are first-Hard cards, I’m going to let the Hards be shorter and the Goods be longer.

  • Increase my relearning step to somewhere in the 15-90m range (currently 10m), which should be enough to keep my retention up on those.

I’ll give this a try and see where it takes me. Thanks!

3 Likes
91

You’re so fast! Thanks!

92

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.