Thanks for the info Expertium.
But it really looks like Ease Hell to me lol.
1 Like
The thing is, me and LMSherlock have tried every singe idea to improve the formula for D that we could think of. Nothing worked. We tried some simple changes, we tried completely scraping the curreent formula and re-defining D. Nothing worked, in the sense that nothing improved log loss/RMSE, only made them worse or didn’t change them. I have personally tested at least 30 changes. Nothing. This doesn’t mean that the current definition of D is perfect and it’s mathematically impossible to do better. It does mean, however, that virtually anything that makes you think “Obviously this would make difficulty better, why haven’t you tried that?!” is almost guaranteed to not work.
Have you reoptimized your parameters recently? After 3-26, for instance? Perhaps again after 4-12? (Maybe even after 3-18?) That’s what it looks like to me. It’s hard to compare one revlog entry/memory state to another if they are calculated based on different parameters.
Do you use any add-ons which may modify the Ease? It’s unexpected that the difficulty increases when you press good.
im optimizing the parameters nearly each day yes, or at least i click on the button and i often got told that the parameters are fine.
In my opinion since i moved to fsrs i dont use any add-ons any more that modifies the ease (in earlier days i used straight reward as u can see in the screen).
I now only use 2 buttons (again and good), months ago I occasionally pressed them lightly or heavily.
At least that explains why your D seems to be backtracking – the next time you study the card it’s being computed with different numbers – all of your memory state values are.
That is unnecessarily often to reoptimize. Through at least Feb/Mar, and into April (until whenever you upgraded to 24.04) – you may have been giving yourself parameters that were a worse fit for your memory curve. I have seen the devs putting a lot of work into making sure that isn’t possible anymore, but you still don’t need to re-optimize more than every month or two. If you stop doing that, then I think you’ll be able to start seeing progress in card info like this one.
You said your desired retention was 90% – are you keeping an eye on your true retention? That’s a more effective way to make sure FSRS is meeting your needs than constantly re-optimizing.
- Stats > Answer buttons > Mature % correct
- [If you have the FSRS Helper add-on] Shift-click Stats > True retention
1 Like