Feat/flat power forgetting curve

[L-M-Sherlock]

Idea: open-spaced-repetition/fsrs4anki#461 (comment)

Python Implementation: open-spaced-repetition/fsrs-optimizer@8ac5d5b

Help wanted: There are duplicated function named power_forgetting_curve. Is it possible to refactor them into one function?

[asukaminato0721]

linux:

---- pre_training::tests::test_loss stdout ----
thread 'pre_training::tests::test_loss' panicked at src/pre_training.rs:327:9:
assertion `left == right` failed
  left: 14.5771
 right: 14.577101
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

win:

---- pre_training::tests::test_loss stdout ----
thread 'pre_training::tests::test_loss' panicked at src\pre_training.rs:327:9:
assertion `left == right` failed
  left: 14.5771
 right: 14.577101
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

floating point again :(

[dae]

I can't see an easy way to combine the three power_forgetting_curves, as you have one that uses raw floats, one that uses NdArray, and one that uses tensors. You could potentially just have a raw float one and then convert the tensors to floats and back again, but it might be a bit slower if it's called frequently.

[asukaminato0721]

Is it possible to refactor them into one function?

They have different types, not something like Tensor<B, 1> vs Tensor<B, 2>

instead, &Array1 vs Array1 vs f32 vs f64 vs Tensor, so it's hard to use trait to merge them.

Maybe first we need to unify the array type...

[asukaminato0721]

Is it possible to refactor them into one function?

They have different types, not something like Tensor<B, 1> vs Tensor<B, 2>

instead, &Array1 vs Array1 vs f32 vs f64 vs Tensor, so it's hard to use trait to merge them.

Maybe first we need to unify the array type...

Some small thought, maybe (I am not sure how to write the trait, this is a draft.)

fn power_forgetting_curve<T1, T2>(a: T1, b: T2) -> T1
where
    T1: Add<Rhs = xxx, Ouput = T1> + Mul<Rhs = xxx, Output = T1> + Pow<Rhs = xxx, Output = T1>,
    T2: Add<Rhs = xxx, Ouput = T2> + Mul<Rhs = xxx, Output = T2> + Pow<Rhs = xxx, Output = T2>,
{
    a.some_operator(b)
}