If I run a simulation, everything looks good, but if I optimize my present and rerun, reviews increase drastically. Why is this?
Deck 1:
Deck 2:
In Deck 2, which has no new cards anymore, my review count never goes down even after 10 years. This seems bugged to me.
I’m no expert on the FSRS algorithm, but the biggest thing that stands out to me for deck 2 is the desired retention rate. Since your desired retention is 74%, you’re going to end up reviewing cards more often.
In deck 1, your review counts climb almost to 1000 around October but then drop off as the year progresses. This generally makes sense for a retention rate of 85%. Of course, other factors are involved, like stability of your cards, etc. But, in general, a card will be scheduled more frequently or with smaller intervals initially to help you get to an 85% retention rate. Once you’ve achieved that, then the intervals can get larger as it’s mostly a matter of maintenance.
With deck 2, no front loading is needed since the retention rate is lower. While the interval for any one card will still get larger over time (assuming you are always choosing a rating of 3), the rate of interval increase will be slower.
I hope that helps.
I assume the ones in the screenshot are your optimized parameters you used for the 2nd sim. What parameters did you use for the 1st sim?
Have you run Evaluate on each set of parameters? What’s your RMSE?
Deck 1:
First sim:
Second sim:
Time increases too:
Deck 2:
First sim:
Second sim:
Those old parameters aren’t the defaults, so I assume they are from a past optimization. It looks like a lot has changed between then and now. I think what you’re seeing here isn’t that the simulation got “significantly worse” – it got significantly more accurate.
When your RMSE drops from 12% to 7% – or even just from 6.86% to 6.09% – that’s FSRS telling you, knowing what it knows now, that the old way it was scheduling your cards isn’t as good a fit for you, and it’s going to schedule them more accurately now. That might mean you’ll have a different number of reps, differently growing intervals, and a different workload – but that doesn’t necessary mean that it’s worse.
[Every interpretation of FSRS Simulator results comes with a big asterisk – this is an experimental feature, so it also just might not be accurate.]
Okay that makes sense. I guess my ultimate question then is this: If I don’t change, will I have less reviews, vs changing will increase my reviews, according to the simulation?
I believe you’re saying the answer is yes. However, that also means it will actually get me closer to my desired retention. Is that correct?
If the simulation is correct – yes, it looks like it’s telling your you’ll have more reviews. The reason FSRS would be doing that would be to get you closer to your DR.
Okay, I’ll go ahead and optimize then, and just adjust the desired retention as needed if my reviews get out of hand. Thanks!
