long-term scheduling
I've mentioned this before, but when examining my scheduling of my notes, I'm noticing cards for which the recent run is 1incorrect, but are set to be scheduled for study again fo a couple of years. Likewise, I also see notes that have scheduled the next due date four years out from the last due date where they were graded correct. The former is certainly a problem. The latter is due to the expectation that the science behind the algorithm, which studied the effect of spaced repetition over the course of some months (I don't think they even studied a year) will continue to hold true for a period of years without the rate retention improvement ever diminishing or plateauing. I mean, we are not machines. I suppose if your app had been around to help me get an A on my Anthropology 101 course in college, I could have continue to study the notes and by now I would only need my next refresher 16 years from now and still get an A if retaking the exam. Just add an option for the user to put a maximum on the amount of time allowed to pass before a next due date. I'd rather cap it at around 15 months. There are other apps that have this option. Why not add it to yours?
Comments are currently closed for this discussion. You can start a new one.
Keyboard shortcuts
Generic
| ? | Show this help |
|---|---|
| ESC | Blurs the current field |
Comment Form
| r | Focus the comment reply box |
|---|---|
| ^ + ↩ | Submit the comment |
You can use Command ⌘ instead of Control ^ on Mac
Support Staff 1 Posted by drewmccormack on 16 Jan, 2019 08:46 AM
If you have 1 incorrect in the run, I would indeed expect that the due date would be quite soon. Sounds like that is a glitch. Perhaps you can try refreshing the due notes and see if that helps. Or maybe switching the schedule to something else, and then back again.
If you think you know better than the science, you are welcome to challenge the science. It's never perfect. But it is generally better than someone with a wet finger in the air.
The point with spaced repetition is this: Imagine you have a 2 year interval after studying for a long time. You study the card after those two years, and you know it. You grade it right. If you knew it after two years of not seeing it, you actually know it very well. The chance of forgetting it is very low — you knew it again after 2 years after all — and it can be scheduled even more years into the future. (Of course, if you are "lying" when you grade it right, you have done yourself a disservice.)
Drew
drewmccormack closed this discussion on 16 Jan, 2019 08:46 AM.
zeppo re-opened this discussion on 16 Jan, 2019 05:49 PM
2 Posted by zeppo on 16 Jan, 2019 05:49 PM
The point is there IS no science for space-repetition beyond a year (if that long). So you are putting your wet finger up in the air. To extrapolate this short term study out with a straight line (doubling every interval) or exponential curve is unlikely to be found a correct assumption if long-term studies were to be conducted.
On the other hand, a suggestion has been offered:
Do what other spaced-repetition apps provide and offer an option to alter or put a max on the long-term intervals.
Support Staff 3 Posted by drewmccormack on 16 Jan, 2019 07:14 PM
I think you will find the supermemo researchers have been studying this for 30-40 years. They have very many papers on various aspects of it.
But if you don’t believe in it, don’t use it. Do your own scheduling or something.
Kind regards,
Drew
Support Staff 4 Posted by drewmccormack on 16 Jan, 2019 07:15 PM
https://scholar.google.nl/scholar?hl=en&as_sdt=0%2C5&q=spaced+repetition&oq=spaced+repe
5 Posted by zeppo on 17 Jan, 2019 12:19 AM
I'm aware of the science behind spaced-repetition and believe in its benefits But supermemo researchers have not been following a single group of users for 30-40 years, nor anywhere close to it. To extrapolate a study on the effects of spaced-repetition study for a group of users monitored for a period of months to assume the same learning/retention curves hold true over a period of multiple years is not science. Anyone that said such short-term studies were all the proof you needed that the findings would hold true over much longer time periods would be laughed at by scientists -- that would be bad science. Now it would make for a good information with which to construct further long-term experiments. But beyond that it is pure hope and speculation. It stands to reason that longer periods of time will contain many more significant life changes and events that will distract and impact retention of something you last studied years before. There is also the factor of aging that may be a consideration. We don't know. I think the longest they've ever followed retention results for a group of users may have been 15-18 months in one of the early studies (I seem to recall also that they had a smaller subset of users that continued to be studied for the duration-- not the entire group). Anyhow, I had enough to convince me of the shorter term benefits of spaced-repetition, coupled with my own experience, but nothing to point to a steady or ever-increasing percentage of increase in the length of intervals between study events should continue apply, unwavering, to longer and longer terms.
The other thing to consider is I have used the app's long-term study mode on notes that go back years. Whereas I'm quite certain you do not have a significant amount of material loaded on your Studies app set for long-term study that have been scheduling daily notes for you for years-- your unfamiliarity to the sync issues of this app tell me this is true. But I try to keep my eye out for clues as to what may be the problem because I know you are never going to come across them yourself as you only test short-term use.
6 Posted by zeppo on 17 Jan, 2019 04:27 AM
In reviewing my Recently Incorrect cards, almost all of the have a long span of time until their next due date. My favorite is the one with the string of 1 incorrect last studied February 2018 and scheduled to become due again in January 2024.
I have to wait until I have due notes, but I'm going to see if perhaps a card that is initially marked wrong in a study session by accident , and then you swipe back to the card to correct it , somehow retains the status of 1 incorrect in that info panel statistic. That's the only thing I can think of. I've got over 400 notes that show up in my Recently Incorrect category (which considers something marked wrong in August as "recent" I guess ), so there is no shortage of examples.
Also, a few of them have due dates from 2018, as if they missed their due date somehow and got stuck in limbo. They aren't scheduling.
Switching the schedule of the stack off, or to short-term scheduling, and then back again doesn't affect it.
Support Staff 7 Posted by drewmccormack on 17 Jan, 2019 08:31 AM
My guess is that supermemo does have people who have been using the app a very long time, and they probably use that data. It is not really necessary for them to actually continually follow a group.
But this is a silly argument. I am not claiming to be a scientist in this area, and I have implemented a rough approximation of a standard supermemo algorithm in a consumer piece of software. If you really, really want all the bells and whistles, you are better with something like Anki. (Be prepared to leave your soul at the door, because that thing is awful to look at and use.)
All the stuff you describe sounds like problematic data. The algorithm is not supposed to work that way.
Drew
drewmccormack closed this discussion on 17 Jan, 2019 08:31 AM.
zeppo re-opened this discussion on 17 Jan, 2019 02:08 PM
8 Posted by zeppo on 17 Jan, 2019 02:08 PM
The best thing you've got going for your apps is the mac app's ability to manage notes (edit and arrange), and to also have that data on your computer to easily backup. But studying is best done with an iphone because you want to be able to fit in the studying where you would otherwise be killing time, like when standing in line or stopped at a red light. But when you can't sync the iphone and the mac app without that occur with long term use, that advantage is lost.
Yes, you are correct. Your app caused problematic data. It's not supposed to work that way, yet it does. That's how bugs work. They do the unexpected without a known explanation, and often with unexpected solutions. It happens. You might look into fixing it. It will make your app better for long-term users. On the other hand, for those that just want to use an app to study for a test in class, and delete the data once its over, I'm sure the app will work fine.
drewmccormack closed this discussion on 18 Jan, 2019 07:58 AM.