Spaced repetition

[Guest]

Any chance of adding a more sophisticated SR system in the near future? Really like this app over the competition but wouldnt want to be stuck with all my cards for the future without an "up to date" algorithm. Just seems like a huge waste of time to handle every card identically in the long term although difficulty will differ greatly..

supermemo.com/english/ol/sm2.htm

[norbert]

What do you mean with sophisticated algorithm? Do you have any proposals what you want to improve? As far as I see that is one of the best algorithm I have seen, esp. since you can tweak it in very details to your liking.

[norbert]

Ahh, that one in the link. Well, there is one problem:
# After each repetition assess the quality of repetition response in 0-5 grade scale:
5 - perfect response
4 - correct response after a hesitation
3 - correct response recalled with serious difficulty
2 - incorrect response; where the correct one seemed easy to recall
1 - incorrect response; the correct one remembered
0 - complete blackout.

In FC-Deluxe there is only Strong accept, accept, and reject. And honestly I think that is good so.

The idea to have 6 different levels to choose from will cost me more time in answering the thinking aobut the answer itself.

OTOH, the above categories are actually a quite good description of what I am seeing myself. I am not sure if I would like to have these 6 options in FC, but indeed it seems to be a nice idea to play with.

[Guest]

Its not about more answer possibilities. Three are enough on a mobile phone in my opinion.. practically never use more than 5/3/1 on my desktop client anyway. More answer possibilities just speed up the difficulty estimation procedure.

Its about some flexibility in the in interval size (called e-factors in the link). So, instead of having constant factors for every single card to calculate the intervals, you estimate the difficulty of a card based on past answers and base the intervals for that card on that. A card you have responded to positively many times for example will have longer intervals in the future than a card you have answered to "wrong" more often than "correct". Difficult cards get shorter, easy cards longer intervals. Especially after you have forgotten a card with a long (years..) interval, the suggested algorithm would be a huge improvement.

The current algorithm is the simplest form of Spaced repetition. I agree that you can debate if the newest versions of the SM algorithm bring any measurable improvements (I dont think so), but the one in the link certainly would be an huge improvement in the long term over the one currently integrated in FCD.

[Ernie]

Hi Guest,

I don't have any plans right now to include e-factors in the logic. I wanted to keep things a little more simple and understandable, so I didn't use this idea. My factors don't have any memory of past performance like you said, but I generally like it that way. You say "huge waste of time", but I disagree. Cards you know well get higher and higher intervals, and those you don't won't. A downside that you mentioned is that after an incorrect response, two cards that might have a very different history but now have the same interval will behave identically with the same responses now. While there could be an improvement on incorporating this history in the calculation, my opinion is that the simplicity of my logic outweighs the relatively little benefit (in my opinion) you would gain in real-world practical use.

It's very probable that I'll be tweaking my logic or options in the future, but I don't have plans now for e-factors.

Anyway, that is my view and thoughts on this. Sorry about that.


Ernie

[Katja from Hamburg]

perhaps I'm on the wrong page but I set max-wrong value to 3 days and therefore wrong cards will have a short interval for the next time as my correct-interval-factor is 1.4.
In addition I use the feature "next review" if I like to overwrite what is calculated by FCD.