A few days ago, Koohii focused my interest in better analyzing my Anki reviews. I already had a browser app to explore deck contents, and added review analysis to that. Please upload your collection.anki2 Anki database (found in Documents/Anki/name/) and give it a try!
http://fasiha.github.io/fuzzy-anki/ よろしく!
If you want to commit your browser to consuming your entire database (a fellow Koohiite didn't have (much) problems with their database of 130'000 reviews), change the limit to something large, like a million, and upload your Anki collection database.
You'll be able to export the review log as CSV (and view it in the browser), as well as run some visualizations on it.
The visualizations are fun. Select the deck(s) you want to run them on, then optionally select the fact fields you want each model therein to use in the display (to help you identify interesting and potentially troublesome cards), then click the "Visualize" button.
Here's the analysis for my RTK1/3 deck. Please don't publicly laugh at my terrible performance.
---
![[Image: 2LhTiVV.png]](http://i.imgur.com/2LhTiVV.png)
Lapses, i.e., number of times you've missed a card, versus how many days since you learned it. You expect some widening in lapses as cards get older, but here, some cards planted long ago have far higher lapse rates than peer cards of the same age. I need to fix these's stories right now! I can also see the use of an automated warning for cards that are drifting too high above their peers.
---
![[Image: IwQbBeb.png]](http://i.imgur.com/IwQbBeb.png)
A histogram of pass rates of cards (pass here means ease>=2: a non-lapse). I have a very curious bi-modal distribution: the histogram has about half its mass at 100% (50% of cards have never lapsed), and the rest of it is diffused over a bell curve between 80% and 96%. Does anyone else have something like this?
---
![[Image: Sz32dZP.png]](http://i.imgur.com/Sz32dZP.png)
Calendar view of when cards were planted. The slope of the curve helps me reminisce about times past when I was consistently planting cards. Each card is represented by a circle whose radius and opacity are tied to the number of lapses: large, transparent circles have no lapses; small, opaque circles have many lapses
---
![[Image: Po0jJkX.png]](http://i.imgur.com/Po0jJkX.png)
This is a quick-and-dirty plot showing the relationship between two highly correlated variables: the number of lapses and the total number of reviews for each card. I like the blocks the jitter makes in the lower left, but the data's meaning is more apparent and less misleading in the first plot (card lapses versus card age).
I'm not too fond of these last two plots. I think they're trying to capture something but I'm not sure what.
---
Please feel free to ask for visualizations that interest you and I can add them.
Geek notes:
I'm using C3.js, which is a plotting wrapper over D3.js (which is used by itself extensively here). It's surprisingly good, but not perfect, and I'm planning on replacing it with pure D3. Most annoying is that the tooltip will highlight only dots you don't care about.
I know I've use the word 'analysis' but there isn't any here yet. I was planning on fitting some scatter plots to least-squares lines but I haven't found a compelling-enough view into this database to warrant that.
There's a lot of todos and room for improvement. For purely technical things that you don't want to bother Koohii with, feel free to file a bug report on Github: https://github.com/fasiha/fuzzy-anki/issues (where you can see fixed issues too). I hope to clean up the user interface (put visualizations in their own tabs, with separate tabs for decks, etc.) and refactor the code to make it Javascript-object-oriented.
http://fasiha.github.io/fuzzy-anki/ よろしく!
If you want to commit your browser to consuming your entire database (a fellow Koohiite didn't have (much) problems with their database of 130'000 reviews), change the limit to something large, like a million, and upload your Anki collection database.
You'll be able to export the review log as CSV (and view it in the browser), as well as run some visualizations on it.
The visualizations are fun. Select the deck(s) you want to run them on, then optionally select the fact fields you want each model therein to use in the display (to help you identify interesting and potentially troublesome cards), then click the "Visualize" button.
Here's the analysis for my RTK1/3 deck. Please don't publicly laugh at my terrible performance.
---
Lapses, i.e., number of times you've missed a card, versus how many days since you learned it. You expect some widening in lapses as cards get older, but here, some cards planted long ago have far higher lapse rates than peer cards of the same age. I need to fix these's stories right now! I can also see the use of an automated warning for cards that are drifting too high above their peers.
---
A histogram of pass rates of cards (pass here means ease>=2: a non-lapse). I have a very curious bi-modal distribution: the histogram has about half its mass at 100% (50% of cards have never lapsed), and the rest of it is diffused over a bell curve between 80% and 96%. Does anyone else have something like this?
---
Calendar view of when cards were planted. The slope of the curve helps me reminisce about times past when I was consistently planting cards. Each card is represented by a circle whose radius and opacity are tied to the number of lapses: large, transparent circles have no lapses; small, opaque circles have many lapses
---
This is a quick-and-dirty plot showing the relationship between two highly correlated variables: the number of lapses and the total number of reviews for each card. I like the blocks the jitter makes in the lower left, but the data's meaning is more apparent and less misleading in the first plot (card lapses versus card age).
I'm not too fond of these last two plots. I think they're trying to capture something but I'm not sure what.
---
Please feel free to ask for visualizations that interest you and I can add them.
Geek notes:
I'm using C3.js, which is a plotting wrapper over D3.js (which is used by itself extensively here). It's surprisingly good, but not perfect, and I'm planning on replacing it with pure D3. Most annoying is that the tooltip will highlight only dots you don't care about.
I know I've use the word 'analysis' but there isn't any here yet. I was planning on fitting some scatter plots to least-squares lines but I haven't found a compelling-enough view into this database to warrant that.
There's a lot of todos and room for improvement. For purely technical things that you don't want to bother Koohii with, feel free to file a bug report on Github: https://github.com/fasiha/fuzzy-anki/issues (where you can see fixed issues too). I hope to clean up the user interface (put visualizations in their own tabs, with separate tabs for decks, etc.) and refactor the code to make it Javascript-object-oriented.
Edited: 2014-09-02, 3:06 am