No, I find it okay, but it will take time to get used to it. I suppose these are stats that are not meant for beginner users anyways.
Hmm. So if stability changes with changing learning steps, does that mean there is no “optimal learning steps”
For example I imagine that if the stability for again interval is 0.05 min and then I go ahead and change it to 0.03 min, and the stability would then decrease further down to 0.03 → change again interval → stability decreases → change again interval…
So this creates a cycle, or am I understanding things a bit differently here
The main problem here is, not only the learning steps changed, the learning stuff also changed. We cannot control the variable of the learning stuff.
If the learning stuff doesn’t change, ideally, the stability will not change. If the stability changes, it’s caused by noise, rather than the changed learning steps.
Okay it is good that the stability is affected by noise at most, provided that the learning stuff is fixed. But do you mean by “changing learning stuff” the changing difficulties of what you are learning and interfering with what has been learned before? I think that is hard to quantify.
This line is a bit confusing. s
I think it would be better if you made what is meant by “first rating”, “the next reviews” and “delay since the first review” a bit clearer. They are a bit ambiguous.
In the first point, is it “for each first rating” or “after each first rating”? The second is clearer for the latter case.
The next three points can be improved. You can just describe what T25, T50, T75 includes instead of describing a process people aren’t aware of. Something like, “T50: This includes lowest 50% of the delays.” maybe.
Maybe this is just related to the way I (somewhat rarely) use hard on new cards combined with how Anki will mechanically give them a constant fraction of the good interval, which for me is always ending up as too short.
I think it’s also useful to add a bullet explaining how this information could be used, e.g. Again+Good stability inform first and second learning steps and Lapse stability informs you on the relearning step (scaled based on target retention to calculate intervals)
This table shows the average time you wait before rating each card the next time (Time Delay) based on your first rating of the day for each card in the deck (Again or Hard or Good or Lapse).
It also shows how well you remember a card after each subsequent rating (based on its first rating) on average.
The subsequent ratings after the first ratings of all cards in the deck are gathered and sorted by ascending order of the Time Delay (not shown on the table) and are then grouped into 4 groups (Time Delay 1<2<3<4).
The 4 groups are further split and assigned to whatever the first rating of the cards was (Again or Hard or Good or Lapse). Therefore, each First Rating has 4 groups of subsequent ratings (Groups 1,2,3,4).
Average Retention rates (R̅₁, R̅₂, R̅₃, R̅₄) for each group of subsequent ratings and the Average 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.
TX% means that X% of the cards in this deck with a first rating (Again or Hard or Good or Lapse) are delayed by this amount of time or less until the next rating on average.
For each first rating, the next reviews are sorted by the delay since the first review.
The next reviews are divided into four equal parts based on the delay, and the retention (R̄1, R̄2, R̄3, R̄4) is calculated for each part.
T25%, T50%, and T75% are the first, second, and third quartiles of the delay, respectively.
With the article L.M.Sherlock linked to, it seems that quartiles separate quarters of the sorted list.
If I understand it correctly, T50% is the median: 50% of the first reviews are followed by a shorter delay, and 50% by a longer delay.
T25%: 25% by a shorter delay, and 75% by a longer delay.
T75%: 75% by a shorter delay, and 25% by a longer delay.
Yes, but I fail to understand what is the problem here. What you are saying is basically what I am saying.
T25 means that 25% of the cards in this deck with a first rating (Again or Hard or Good or Lapse) are delayed by this amount of time (or less) till the next rating on average.
T50 means that 50% of the cards in this deck with a first rating (Again or Hard or Good or Lapse) are delayed by this amount of time (or less) till the next rating on average.
T75 means that 75% of the cards in this deck with a first rating (Again or Hard or Good or Lapse) are delayed by this amount of time (or less) till the next rating on average
The counts will be lower as they are only checking the button pressed the first time the card was ever seen. That graph includes reviews after the first time the card is seen (i.e. the learning steps).
You should see the review count for Again+Hard+Good+Easy add up to the number of cards you’ve learned in your deck (Easy isn’t in the table, so Again+Hard+Good should be a little lower than total cards learned).
I updated it to be a bit less wordy and corrected things a bit. Should there be no issues regarding its factuality, I suggest keeping this as the final version.
s
