Learning Step and Review Interval Retention (Support thread)

This is the support thread for Learning Step and Review Interval Retention.

This add-on adds statistics showing the percentage of correct answers for each learning step and review interval.

2 Likes

I like your addon. However I cannot tell how well I do in the graduating interval because it gets mixed with lapsed cards. I mean cards that I might have added 3 months ago but failed 2-3 times a row and ended up having the same interval as the card who just graduated.

To combat this I created new settings that work like this 20 1441 7200 17280 (30 days Grad int). The downside is that if I fail the card then I have to go through these steps again. Also if the card is difficult then it will get stuck in a learning circle for months which has happened to me. This is why I donā€™t like these settings but I have no other choice.

It depends on your settings, difficulty of the cards, your ability to memorize them, etc., but I think that for your graduation interval, graduated cards should vastly outnumber other cards in review having the same interval. They should not affect the statistic that much.

I will probably add an option to show the graduation interval as a learning step. The data required for that is not readily available so it requires some trickery during the calculation. There are some edge cases that I have to take into an account first. Do you have any suggestions for these?

  • How to display the data in a table? The intervals are fuzzed and I donā€™t think the nominal intervals are stored anywhere. Should all graduated cards be displayed in a single ā€œGraduatedā€ step?
  • What do with relearning cards? The intervals can be practically anything. Does it make sense to group them?
  • What to do, if there are crammed reviews via filtered decks between the last learning step and first review?
  • While the review history of each card has to be scanned through anyway, should same-length learning steps be separated?

Well, I am a future programmer myself however I cannot think of a solution.

I guess first you need to study the fuzz. I once saw that python method. I donā€™t know how to find it again.

I think they have to be shown in a single graduating step because it wonā€™t do anything if you have 1 card on day 13, 5 on 15, 2 on 16, 3 on 17. You could get all cards by tracking the last two intervals. If interval Last - 1 = 5 && Last interval == 10 then display. Or some isolation through last review and latest review. If last first review > 30 days than last review display. (there should be some uncaught cards here because some get lost like mine 2 months in the learning steps).

It would be great if you could separate. The way that I separated them is also good with 1441, 1442, 1443, 5201, 7202, 7203.

I certainly have no clue about the cramming decks. I didnā€™t understand you. However I think that very few people use this so you could work by excluding these.

The learning steps should be separated yes.

I donā€™t know what to say but your add on has helped me a lot. When 3 months after I used Conaanas settings I realized that there was something wrong. I had a few cards trapped in learning steps, failing mature cards and my retention rate to 76%.

By using your add on I finally figured out that for my 3 decks Easy, Medium, Hard, 6 step days is too long for Hard and 5 days steps is too short for easy. I removed the graduation from 12 and put it on 30 so I could finally see what retention I got for 12days. Graduation of 30 increased by reviews by a lot though and changed the graph into days with low peaks and very high peaks.

I donā€™t know how these add ons are made. If this is Java then I could work with you for a solution.

The latest version supports showing the first review after graduation as a learning step and separating successive learning steps with the same interval. There are a couple of options to adjust those features.

Hey dude, I still donā€™t understand it. If I have 3 steps 20min 1day 6days and graduation 12. What is the last step calculating the 12 Ā± 0-3 days?

image

I would really appreciate your help and big thanks for making this add on.
In this scenario what is this supposed to mean on the graduation step(12) days the next one is 30. It means that 69.8% is the 30 day step?

The last step corresponds to the graduation interval (12d).

Ok then the 69.8% is the step after that step which is the 30days?

I am sorry to waste your time with questions like this but I have to know exactly the case.
Do you have a patreon?

No problem. And I think I might have misunderstood your question last time.

The last step (Graduation) is the 12d step. Your retention for learning cards that have graduated is 88.0%.

The 69.9% figure that might be confusing you is your retention for relearning cards after the last relearning step. I donā€™t think Anki has a term for those, but maybe you could call them regraduated cards. Their step length can be more or less anything depending on your ā€œnew intervalā€ and ā€œminimum intervalā€ settings. Your screenshot is missing the column headers, so it looks a bit confusing.

I donā€™t have a Patreon account but I appreciate the thought.

Sorry I didnā€™t post the whole profile because you would get stormed by 15 different percentage settings as it hasnā€™t been a month since I made the change.

The last relearning step? This means that a card that had 2 months interval and I failed 5 times when it gains back 12 days interval(graduating interval)?

Sorry for these questions. It would be great if you could explain them in the application page as you might have more people like me.

Relearning steps are the steps under the ā€œLapsesā€ tab in the deck options. The 69.8% is the retention for the first review step after the last relearning step no matter what the length is. It most likely is not 12 days unless you have ā€œnew intervalā€ set to 0% and ā€œminimum intervalā€ set to 12 days.

Iā€™ll update the add-on page once I figure out how to explain it more clearly.

I feel like this has to be obvious, but, how do you actually see the new stats?

Iā€™ve disabled every other add-on. Iā€™ve been looking at both deck stats and collection stats, but not seeing any ā€œRetentionā€ headings. Iā€™ve tried looking through menus too, but no luck so far. Iā€™m on Win 10 with Version āØ2.1.38 (355e4cd5)ā©
Python 3.8.6 Qt 5.14.2 PyQt 5.14.2.

My headings in stats are: Today, Future Due, Calendar, Reviews, Card Counts, Review Intervals ,Card Ease, Hourly Breakdown, Answer Buttons, Added

Hold shift while clicking the Stats link to open the old statistics view.

This add-on supports the old statistics only. The new statistics are a pain to extend, so as far as I know, thatā€™s the case for all the other add-ons too. Iā€™ll add a note about that to the add-on page.

2 Likes

Ah, yup, that did it. Thanks.

Hey Tossu,

Would it be possible to update the Add ons main page with the way that it functions?

Many thanks

Thank you very much for the addon!
You can add an option to show only current ā€œLapsesā€ in the deck options.
Another thing, you could have an option similar to ā€œmax_ivlā€ but where you would set a date (y / m / d), for counting. Or else archive / delete old histories (lapses).
example:
ā€œarchiveā€: 10, 20, 1440

This add-on does not show among the Anki charts.
Can I see it in previous Anki versions ?

Thank you for sharing.

@vit This add-on should work with the latest version of Anki (2.1.42). However, it does not support the new stats. To access the old stats, hold shift while opening the stats window.

@huandney I donā€™t quite understand what you mean. What would the setting in the deck options do? The add-on already respects the selected period (1 month, 1 year, or deck life) so I donā€™t see a reason for an option to limit the range further.

Also, your review history is what it is. Ignoring parts of it would only give you bogus statistics. If you want to remove old reviews from your review history, I believe there is an add-on for that.

@jeitiel Are you still having trouble with the add-on? Is there still something confusing?

I am good with the add on. I am very close to seeing the 2-month results of my last experiement.
I tried to do steps for following for my main programming deck.
25min 1d 4d 8d 20grad
32% 65% 80.1% 78%

Thanks to you now I can see the step after the graduation which is 50d at 59%. This is the lowest I have ever gotten to which makes me think even though I try to adjust the intervals to aim for 80% shorter intervals do not help with long term memory. Studies are made on these. Cards I hit correct on the 8th day maybe, I could get them correct on the 12th as well but with a bigger gain on long term.

In conclusion I might be hurting myself by lowering the intervals based on the info I got from the addon. What do you think? Should I aim for longer than 4-5 days on the first big step?Itā€™s been an interesting fun time though.

For General Info Deck it is ok though
26min 1d 6d 11d 28grad
35% 74% 73.4% 82% 78%
next step after grad 75%