Will this script generate intervals correctly using the SM-2 algorithm?
Bill Powell
First off, thanks so much for Mnemosyne. It's an excellent
program, and a joy to use.
But I spend so much time on the computer that I'm intrigued
at the possibility of doing a simpler version of all this on
paper.
This old SuperMemo article described a simple paper method:
http://www.supermemo.com/articles/paper.htm
but unless I'm mistaken, this method doesn't have the
finesse of rating the quality of your response from 0-5, and
scheduling your next interval based on that.
I'd like to generate a simple chart of intervals, so that if
I scored, say, a 3 on a card on repetition 4, I could look up
the right interval before repetition 5. I'm envisioning 6
charts, one for each of 0-5.
I've looked around but haven't been able to find charts like
this for Mnemosyne. The "scheduled cards feature" only shows
the next 7 days; I'd like to be able to pencil in up to 20
or 30 years in the future. Ideally, I could sit outside with
a couple notebooks, rate my responses, check the interval
charts, and jot the next interval in my schedule.
Mnemosyne links to the SM-2 algorithm, and so I worked up a
very simple bash script to generate a chart of intervals,
based on an original EF of 2.5, the default. You input a
quality response of 0 to 5.
Could someone take a quick look and tell me if I'm correct?
I have a feeling I may be missing something.
I know that, in reality, I'd have to _also_ save this new EF
and input it along with my next response. But for now I
just want to make sure I'm correctly determining the
intervals based on an EF of 2.5.
For instance, for a quality response of 4:
---
$ efintervals.sh 4
Quality of response: 4
Assumed old EF (Easiness Factor): 2.5
New EF based on response of 4 (EF'): 2.50
New Intervals:
(This script doesn't yet calculate anything based on which repetition
you're on, only a complete chart of intervals based on EF' (2.50)
n[1]: 1 days: 1 days
n[2]: 6 days: 6 days
n[3]: 15.00 days: 2 weeks 1.00 days
n[4]: 37.50 days: 1 months 7.50 days
n[5]: 93.75 days: 3 months 3.75 days
n[6]: 234.37 days: 7 months 3 weeks 3.37 days
n[7]: 585.92 days: 1 years 7 months 1 weeks 3.92 days
n[8]: 1464.80 days: 4 years 4.80 days
n[9]: 3662.00 days: 10 years 1 weeks 5.00 days
n[10]: 9155.00 days: 25 years 4 weeks 2.00 days
n[11]: 22887.50 days: 62 years 8 months 2 weeks 3.50 days
n[12]: 57218.75 days: 156 years 9 months 1 weeks 1.75 days
n[13]: 143046.87 days: 391 years 11 months 1.87 days
n[14]: 357617.17 days: 979 years 9 months 1 weeks 5.17 days
n[15]: 894042.92 days: 2449 years 5 months 7.92 days
---
$ efintervals.sh 1
Quality of response: 1
Assumed old EF (Easiness Factor): 2.5
New EF based on response of 1 (EF'): 1.96
New Intervals:
(This script doesn't yet calculate anything based on which repetition
you're on, only a complete chart of intervals based on EF' (1.96)
n[1]: 1 days: 1 days
n[2]: 6 days: 6 days
n[3]: 11.76 days: 1 weeks 4.76 days
n[4]: 23.04 days: 3 weeks 2.04 days
n[5]: 45.15 days: 1 months 2 weeks 1.15 days
n[6]: 88.49 days: 2 months 4 weeks .49 days
n[7]: 173.44 days: 5 months 3 weeks 2.44 days
n[8]: 339.94 days: 11 months 1 weeks 2.94 days
n[9]: 666.28 days: 1 years 10 months 1.28 days
n[10]: 1305.90 days: 3 years 7 months .90 days
n[11]: 2559.56 days: 7 years 4.56 days
n[12]: 5016.73 days: 13 years 9 months 1.73 days
n[13]: 9832.79 days: 26 years 11 months 1 weeks 5.79 days
n[14]: 19272.26 days: 52 years 9 months 3 weeks 1.26 days
n[15]: 37773.62 days: 103 years 5 months 4 weeks .62 days
---
Both these sets intervals seem a lot longer than the old intervals
given on that "paper SuperMemo" link given above, which are:
| Repetition | Interval |
| 1 | 4 days |
| 2 | 7 days |
| 3 | 12 days |
| 4 | 20 days |
| 5 | 1 month |
| 6 | 2 months |
| 7 | 3 months |
| 8 | 5 months |
| 9 | 9 months |
| 10 | 16 months |
| 11 | 2 years |
| 12 | 4 years |
| 13 | 6 years |
| 14 | 11 years |
| 15 | 18 years |
Which is why I wondered if I was making a mistake somewhere.
I've attached the script. I'll send it in plain text if it
doesn't post properly.
Thanks in advance for your help!
Bill Powell
--
_____________________________________________________________
http://pennyjustice.com : How do you spend your penny?
http://wineskinmedia.com : Books and sites crafted with care.
http://billpowellisalive.com : Man found alive with two legs.
_____________________________________________________________
Peter Bienstman
If you are not taking the easiness version into account, I don't think it
matters too much if you take the SM2 table or the paper based table.
Remember that this is not exact science, but rather an empirical set of rules.
Peter
--
------------------------------------------------
Peter Bienstman
Ghent University, Dept. of Information Technology
Sint-Pietersnieuwstraat 41, B-9000 Gent, Belgium
tel: +32 9 264 34 46, fax: +32 9 264 35 93
WWW: http://photonics.intec.UGent.be
email: Peter.B...@UGent.be
------------------------------------------------
Oisín Mac Fhearaí
I'd like to generate a simple chart of intervals, so that if
I scored, say, a 3 on a card on repetition 4, I could look up
the right interval before repetition 5. I'm envisioning 6
charts, one for each of 0-5.
I've looked around but haven't been able to find charts like
this for Mnemosyne. The "scheduled cards feature" only shows
the next 7 days; I'd like to be able to pencil in up to 20
or 30 years in the future. Ideally, I could sit outside with
a couple notebooks, rate my responses, check the interval
charts, and jot the next interval in my schedule.
This sounds like a lot of overhead each day. I assume you're going to have a daily calendar, with enough space to specify a few cards (id numbers?) on each day for the next 30 years?
So every day you'll:
* Eat yesterday's page in the calendar/diary to save space
* Go to today in your calendar and extract the scheduled cards (by id?) from your deck
* For every scheduled card, quiz yourself and calculate the review time and pencil in that card ID, n days in the future
* Do the same for some cards in your 'new' pile.
No offence intended, but it seems like a lot of donkey work on top of the job of learning your stuff (and studying, then adding new cards, of course - this essential task usually takes much more work than learning them in Mnemosyne).
Personally, I'm much less inclined to do a repeating task (even if I consider it a worthy one) if there is any significant/pointless difficulty involved.
For example, I go to the gym/pool regularly because they're within 5 minutes walking distance of my desk in university. If I had to get a bus and travel for 30 minutes, I would probably never have bothered...
Just a thought :)
Oisín
Bill Powell
[Peter:]
> If you are not taking the easiness version into account, I
> don't think it matters too much if you take the SM2 table
> or the paper based table.
>
> Remember that this is not exact science, but rather an
> empirical set of rules.
True. Perhaps my question boils down to this. The SM paper
algorithm from 1985 seems to have much shorter intervals
than my (incorrect?) attempts to generate tables based on
SM-2. So does this mean that the research in the last 20
years has indicated that longer, SM-2 length intervals work
pretty well?
[Oisín]
> This sounds like a lot of overhead each day. I assume you're going to have
> a daily calendar, with enough space to specify a few cards (id numbers?)
> on each day for the next 30 years?
>
> So every day you'll:
> * Eat yesterday's page in the calendar/diary to save space
> * Go to today in your calendar and extract the scheduled cards (by id?)
> from your deck
> * For every scheduled card, quiz yourself and calculate the review time
> and pencil in that card ID, n days in the future
> * Do the same for some cards in your 'new' pile.
Hmm. I should clarify. What I'm trying to do at the moment
is memorize texts, e.g., the Psalms. They're printed off, along
with mnemonic images that I handwrite in the margin (e.g, the Art
of Memory and the loci method). So it's not that hard to
jot "Ps 3" as an ID on my calendar, nor to look it up at a
later date.
Although SuperMemo (and Mnemosyne?) are based on breaking
information down to the smallest possible units, I've been
finding the opposite can be the case, at least with text
memorization. Dealing with chunks of text may mean I waste
time repeating verses I know well, but on the other hand,
the context and flow of the text seem to make memorizing
easier. So maybe it's a trade-off.
If I'm dealing with "chunks", I'll probably only do a
few "chunks" per day. Each "calendar day" need only be a line
on a page, so you can easily fit a month on one page.
Shuffling thousands of cards around does sound a bit
frightening, but citations to larger chunks seem less scary.
(Vocabulary, of course, would be a different matter.)
> * For every scheduled card, quiz yourself and calculate the review time
> and pencil in that card ID, n days in the future
Right, it seems like if I had a one-page chart of intervals,
it would be fairly easy to look up the right interval and
jot it in my calendar.
> Personally, I'm much less inclined to do a repeating task (even if I
> consider it a worthy one) if there is any significant/pointless difficulty
> involved.
I agree. Computers are supposed to do this sort of thing.
But if I can spend some time away from the computer, the
trade-off might be worth it. :)
Peter Bienstman
> So does this mean that the research in the last 20
> years has indicated that longer, SM-2 length intervals work
> pretty well?
Again, I wouldn't really call this research, but rather the empirical
observations of Wozniak.
Peter
Bill Powell
Ah, I see. Well, how about Mnemosyne? What sort of intervals
are they finding work well? Is it too early to tell?
Bill
Peter Bienstman
> Ah, I see. Well, how about Mnemosyne? What sort of intervals
> are they finding work well? Is it too early to tell?
Or current priority is working on 2.0 rather than analysing the logs. If you
want my personal gut feeling, I think that the details of the scheduling
algorithm will result in no big performance gains once you average them over a
large number of cards and a large number of users.
Peter