search

open-spaced-repetition / fsrs-rs

FSRS for Rust, including Optimizer and Scheduler
https://crates.io/crates/fsrs
BSD 3-Clause "New" or "Revised" License
185 stars 19 forks source link

Change the R_MAX of CMRR.rs #219

Closed Expertium closed 2 months ago

Expertium commented 2 months ago

https://github.com/open-spaced-repetition/fsrs4anki/issues/686#issuecomment-2335482240 It appears that our previous estimates of optimal retention were underestimates. I think it makes sense to expand the range. image Note that here it says "workload", but it's not just workload, it's workload/knowledge

user1823 commented 2 months ago

I think that DR above 0.96 doesn't make much sense because the returns are diminishing.

image

In this image, see the drastic increase in slope above DR = 0.96. (There are occasional high slopes before DR = 0.96 too but after DR = 0.96, most of the values are too high.)

Data: https://github.com/open-spaced-repetition/fsrs4anki/issues/686#issuecomment-2336648397

L-M-Sherlock commented 2 months ago

The data is based on default parameters.

user1823 commented 2 months ago

Well, it is OK if you think that some people can have such a high value of MRR. But, I am not really convinced. Nonetheless, let's forget about this for now. We can always decrease the value if someone complains.

Expertium commented 2 months ago

We can change it slightly to 0.97 instead of 0.98

Expertium commented 2 months ago

@L-M-Sherlock do you think we should revert this (or set R_MAX to 0.96 or 0.97)? It seems like the idea that previous values were underestimates was based on the wrong data Btw, here's the most recent graph, but with workload divided by retention. This is with loss_aversion = 1. Figure_1