Forum: Spaced Repetition / Algorithms
We receive a lot of contacts regarding the algorithm. Memory uses a fairly straight-forward approach to its SRS (Spaced Repetition System), loosely based on the principle of Supermemo and Leitner. On top of this, we use a lot of advanced methods to measure your accuracy. We have hundreds of variables, and at times perhaps our algorithm is considered overly complex. In these types of systems, typically after learning a question it is marked due for practice in a number of days based on how well you knew it. We are developing a modular approach where users can use different algorithms. Recently, we have added to user settings: - Ability to cap days next due to an arbitrary number of days - Ability to opt-in to our latest experimental algorithm If any users have suggestions, requirements or requests for algorithm tweaks, improvements, or even significant changes, please let us know here. Prefer the algorithm of another system? Have an idea for how we can improve our system? Let us know. We do have the ability to deploy custom algorithms to certain users to test-out, and if it is found useful we will move them into general release. Please note: This refers to the learning algorithm (i.e. the methodology used to determine when practice is due) specifically.
repeating a word every 1,2,3,5,8,12, etc.. days could be very effective, there is a learning curve in internet but i couldnt find it.
That is how it fundamentally works already, I think - unless I have misunderstood. We work to the learning curve and have our own analysis of it. Depending on if you skip it, get it wrong, or correct. If it's correct, it could end up being 1,2,3,5 etc as per your example. If you are a highly accurate it would be 1 , 4, 10, 25 etc Get it wrong, and it goes back to 0. That's the fundamentals of calculating when a question is due. Optimal learning curve performance is not necessarily the only objective though - it must also remain enjoyable to the users so they are motivated to keep working.
There's no a really import decision we overlooked in phase 1: After adjusting max consecutive days, questions could get STUCK in the future, e..g if you learned and were achieving 190 days in advance DND ("Days next due"), if you ten set it to 10 The older questions would stay on 192. So, because they are no in effect "never due" you wouldn't seem them for 6 months! The fix is simple: After adjusting the number of max consecutive days, you can "Resync" button to ensure that older questions are re calibrated to match your maximum If you play with algorithms a lot, and still find cards get stuck in unepected orders, you may have to reset your entire course prgress.
dev.memory now uses the latest algorithm by default
Forgot to say: the key primary change is that WRONG answers are now shuffled to the back of the pack. Previously, they would be set to 'due now' as per the original SM2 algorithm. In understandable terms: - In the new version, when you get a question wrong, it will be served much more quickly than before. A situation where this could go bad PRIOR to the update: - Create a course , add 20 questions. Learn 20. Come back 1 hour later. Add 20 more. Get questions wrong, and they are served AFTER the previous questions. So, it's fixed. Also, if you still don't like the way it works, there's a shuffle button in the new version to reset the current level.
You must be logged in to comment