Support things like matrix > scalar? Or matrix.greater_than(scalar)?

[fzyzcjy]

Hi thanks for the lib! After looking at doc, it seems that we do not have the things mentioned in the title yet. Can we implement it?

I guess > must return bool in Rust so we cannot do that easily. But can we have a function like ge/gt/greater_than?

[jturner314]

map and mapv can be used for this purpose. I'm not sure that it would be worthwhile to add a bunch of specific methods.

[fzyzcjy]

Thank you! The use case: I am porting python numpy code to rust (you know, from research experiment to industry product). So there are a lot of things like a > 50/a != 100 etc where a is np array.

[jturner314]

I understand, although I'm still not sure these should be added to ndarray itself. Keep in mind that you can always create your own extension traits, e.g.:

use ndarray::prelude::*;
use ndarray::Data;

pub trait CompareScalarExt<Rhs> {
    type Elem: PartialOrd<Rhs>;
    type Dim;

    fn lt(&self, scalar: &Rhs) -> Array<bool, Self::Dim>;
    
    // ...
}

impl<A, S, D, Rhs> CompareScalarExt<Rhs> for ArrayBase<S, D>
where
    A: PartialOrd<Rhs>,
    S: Data<Elem = A>,
    D: Dimension,
{
    type Elem = A;
    type Dim = D;

    fn lt(&self, scalar: &Rhs) -> Array<bool, D> {
        self.map(move |elem| elem < scalar)
    }
    
    // ...
}

fn main() {
    let a = array![1, 2, 3];
    assert_eq!(a.lt(&2), array![true, false, false]);
}

You may also be interested in PR #1042, which is similar in spirit to this issue, but for float math functions.

[fzyzcjy]

Thank you! After my reply, I have made a proc macro. Now I can use:ops!{ (matrix_one > 42) & (matrix_two != matrix_three) } etc, and the operators such as >, ==, ... will be translated correctly. If you are interested, I can open-source this.