I’ve been optimizing about every month since I started Anki. My last optimization was 1.5 months ago. But today, when I clicked to optimize my fsrs parameters, fsrs told me my “parameters currently appear to be optimal” and didn’t change any of my parameters.
I read on a reddit forum that resetting the parameters and clicking optimize changed/updated the fsrs parameters for the author- which is what i did and i also did get slightly different parameters from my previous fsrs numbers.
Does anyone know why the initial fsrs optimization indicated my parameters were optimal but ‘resetting->optimization’ slightly changed my parameters(hopefully to “more updated” numbers?)
Which fsrs optimization result should i settle with?
FYI: My anki version is 26.05. Counting the fsrs number, there are 21 parameter values.
Thanks!
Hi yes. I’m grateful TheBB helped me understand that optimization algorithm can lead to different fsrs parameters based on my “initial” fsrs parameters before optimization, since this would explain why I got different fsrs-parameters after the ‘reset->optimize’.
However, there was a similar reddit forum about this issue few months ago: Reddit - Please wait for verification
It seems the issue for the author was the difference between FSRS5(19 parameter values) and FSRS6(21 parameters). However given my Anki version is 26.05 with FSRS6(21 FSRS parameters), I’m curious how common it is for a monthly FSRS optimization process to yield identical parameter values(assuming similar study habits)?
I don’t know if there is data on it. But I personally get the message, that the params appear to be optimal, most of the time.
It’s a good thing though. It means FSRS knows how you learn well enough that adjustments (different parameters) aren’t neccessary.
Thanks. I was hoping this is a good thing. It was my first time this happened so I wanted to make sure.
I also ran Evaluate for the RMSE values to compare my fsrs parameters:
With unchanged/current parameters: Log loss: 0.3496, RMSE(bins): 1.90%.
With reset->optimize”: Log loss: 0.3496, RMSE(bins): 1.99%.
I THINK fsrs kept my previous set of parameters because the RMSE was lower - which would be a good thing.
[I’m responding here as well to both your Reddit post and comment to me. But I agree with all of the answers you’ve gotten so far.]
Identical? Probably not very common. But the currently-optimal message is totally normal.
FSRS will only update your parameters if it finds a way to improve things (like log loss and RMSE) by a certain amount over your current parameters. You can see that happening in the Options window when it switches from counting through the reviews to –
prior to today, everytime I optimized on a monthly basis I was given updated/changed fsrs parameters.
If your study load is pretty consistent, each month you’re adding new review history, but it is a smaller and smaller portion of the total review history in the preset. All the message means is that FSRS looked at the additional review history you’ve accumulated and learned nothing new about you. That’s actually an excellent outcome.
Either set is fine. But it would be a waste of your time to blank out your parameters each month so you could re-optimize and guarantee you’ll get new ones.