Add cumprod

[NewBornRustacean]

Add cumprod() to match NumPy behavior

Hello ndarray community!

This is implementation of the cumprod() function to match NumPy's behavior regarding axis handling.

Related links

• Implement cumprod #1354
• numpy doc

Details

1. check if self.ndim() > 1 && axis.is_none() { panic!(...) }- removed since we now use Axis directly
2. Some(axis) branch to use input array values instead of result array values - removed since we now use Axis directly
3. tests to match NumPy's behavior

Example Usage

use ndarray::{array, Axis};

// 1D array - Axis(0) works
let a = array![1, 2, 3, 4];
assert_eq!(a.cumprod(Axis(0)), array![1, 2, 6, 24]);

// 2D array - must specify axis
let b = array![[1, 2], [3, 4]];
assert_eq!(b.cumprod(Axis(0)), array![[1, 2], [3, 8]]);
assert_eq!(b.cumprod(Axis(1)), array![[1, 2], [3, 12]]);

Testing

All existing tests pass,

• 1D array behavior with Axis(0)
• N-D array behavior requiring axis specification
• Error cases for invalid axis usage

The implementation maintains consistency with NumPy's behavior, but there might be more elegant way to implement this(especially for the nested loop). please give me any comment/feedback!

@akern40 what's up! I love the automated CI pipeline, thanks 👍

[NewBornRustacean]

@nilgoyette thanks for your quick review! I was seeking for what would be a better approach, your comment was very helpful! I tried to vectorize it a bit more with fold_axis()

[akern40]

@NewBornRustacean great to see you back so soon!

[NewBornRustacean]

@nilgoyette @akern40 sorry to belated reply! thanks to suggestions 🥇 most of the suggestions are included and passes all the tests.

[nilgoyette]

The code is really nice with accumulate_axis_inplace! Sorry for the detour. I guess we all learned something ;)

[NewBornRustacean]

The code is really nice with accumulate_axis_inplace! Sorry for the detour. I guess we all learned something ;)

yes indeed! thank you so much 😃

[akern40]

Thanks for your persistence! Last thing: I don't think Copy should be needed. I am assuming it prompted you for it when you wrote *curr = *curr * prev? If so, I think you can either a) use *curr = curr.clone() * prev, in which case Rust should elide the clone; or b) call *curr *= prev, which would require a MulAssign trait bound. I think I'm partial to option (a), since Rust should still just do a multiply-assign operation when possible, and you have fewer trait bounds. But I'm open to ideas.

[NewBornRustacean]

Thanks for your persistence! Last thing: I don't think Copy should be needed. I am assuming it prompted you for it when you wrote *curr = *curr * prev? If so, I think you can either a) use *curr = curr.clone() * prev, in which case Rust should elide the clone; or b) call *curr *= prev, which would require a MulAssign trait bound. I think I'm partial to option (a), since Rust should still just do a multiply-assign operation when possible, and you have fewer trait bounds. But I'm open to ideas.

@akern40 thanks for the suggestion! 😺 how about this one?

    #[track_caller]
    pub fn cumprod(&self, axis: Axis) -> Array<A, D>
    where
        A: Clone + Mul<Output = A> + MulAssign,
        D: Dimension + RemoveAxis,
    {
        if axis.0 >= self.ndim() {
            panic!("axis is out of bounds for array of dimension");
        }

        let mut result = self.to_owned();
        result.accumulate_axis_inplace(axis, |prev, curr| *curr *= prev.clone());
        result
    }

[akern40]

Looks good to me! Thanks again for your work 😄