How SRS Works (video)
Just saw this great video on SRS (spaced repetition system/software), which provides an illuminating visual explanation:
> The video shows a grid of factoids, where new factoids are being presented at a constant rate. Over time, the factoids begin to fade to black… the closer they get to black, the closer they are to being forgotten. However, if they’re “recharged” by being relearned, they advance up a tier (represented by the color and number of the cell). The higher the tier, the longer it takes for the factoid to be forgotten. If at any point, a factoid gets completely forgotten, it is sent back down to the lowest level.
Be sure to click on “Show more” under the video to see the full explanation.
Via @ajatt.
Brilliant! As an Anki evangelist, i’m going to make sure to show this to everyone. SRS is awesome, it’s just keeping up that is the biggest challenge, especially this semester when I am adding roughly 100 words and phrases to my Anki deck per week.
What an excellent visualization!
Any idea if SRS algorithms are used in Pleco, or whether there are any plans for such an addition? If not, do you have any advice about the best flashcard system (specifically for learning Chinese) which does have SRS built in?
Thanks!
Yes, Pleco does use SRS if you choose the “Spaced Repetition” profile. Algorithm is based on the same SM-2 used in Anki and Mnemosyne.
Michael – I don’t see a “Spaced Repetition” profile in Pleco flashcards. Is it enough to just select “Repetition-spaced” in the Card Selection settings?
“Repetition-spaced” card selection would do most of it, but it’s missing some tweaks to the math that better optimize the algorithm around that (higher score increases for correct answers, e.g.).
What platform are you using Pleco on? Do you only have the “Default” profile? Try creating a new profile called “Spaced Repetition,” go back to the Manage Profiles screen, tap on that new profile, tap “Reset Settings,” tap “Yes,” and tap “Spaced Repetition” – that should give you a new profile with those settings.
Thanks Michael. That worked perfectly. (I am using Pleco for the iPhone.) I notice that the default value for “Limit # of unlearned” cards is 200. Isn’t this a lot? On the subject of “Learned cards”, is there any point where Pleco stop testing a card completely when using spaced repetition?
I saw this recommended by Khatz on Twitter and recommended it to a couple of friends. None of the understood anything. I think it’s the coolest animation for quite a while, they didn’t understand anything. Basically, I think you have to be either very much into SRS or read the description quite carefully in order to understand. However, I have contacted the author and asked him if I he can add an intro and some commentary. He has replied and said that he will, so hopefully we’ll have a more accessible version before long.
Yeah, the video definitely isn’t totally self-explanatory, especially since you start out with the traditional review method, and most of us don’t really put much thought into what “forgetting some words while remembering others” would look like in a data visualization scheme.
Anyway, good work contacting the author. Looking forward to seeing the updated video!
I planned to have the video up sooner, but I’ve just been really busy. It’s finally up, though:
I wonder if there is enough historical data in Anki to illustrate this with actual data. It should look similar, although those annoying leeches would show up as bright 1’s hanging out among the sea of green 4’s.
That would be so cool! I don’t see why someone couldn’t whip up a script that could take a full Anki export file and convert it into this kind of visualization. (Of course, you’d definitely want to be able to “drill down” and take a look at which words certain dots are, which would be a lot more work, and not easy from a UI perspective, either…)
This was the first thing I thought of when seeing the video the first time: Why not use real data? It should be quite easy if one has the right skills (I don’t, unfortunately).
Unfortunately, it would be impossible to do the same thing with the non-SRS version, so for the sake of comparison, using real data won’t help very much. However, if the purpose is to illustrate how SRS works, using real data would be awesome. It would perhaps even be possible to design a program or script that generates an animated version of one’s own SRS work. That would be so awesome I would probably pay for it. 🙂
In this video, we cannot see the concept of “difficulty” of the card that is so important in Anki. All cards are forgotten at the same rate, which is not the case in practice.
I am quite sure a visualization generated from real Anki data would be much more messy than that.
All the information about each repetition are available in the .anki file, but I think it is still quite a bit of work to generate this video.
Ah, that’s a great idea. I’ll have to look at how the .anki files are formatted, but this may very well be possible to do. Thanks for the ideas, guys.
Nice idea in theory. However I have a small problem with things like this. They assume that Anki (and paper flash cards) are the only way a person is encountering the material.
For example, if you are listening to mp3’s and reading material, especially once you get past beginner level, are going to significantly alter what these “graphs” would look like.
And in some ways, if you believe that learning words in context by activities such as reading and listening is more powerful than answering isolated facts on a computer, then things like this are even more useless