Implement a convenience method DenseMatrix::from_ndarray2 (in case the ndarray feature is on)

[Mec-iS]

Example code (place in a module that depends on both smartcore and ndarray)

use std::fmt::{Debug, Display};
use ndarray::{ArrayBase, Ix2, OwnedRepr};
use smartcore::linalg::basic::arrays::Array2;
use smartcore::linalg::basic::matrix::DenseMatrix;

// Convert an ndarray owned 2D array into a smartcore DenseMatrix by copying.
pub fn from_ndarray2<T>(a: &ArrayBase<OwnedRepr<T>, Ix2>) -> DenseMatrix<T>
where
    T: Debug + Display + Copy + Sized,
{
    // SmartCore’s Array2::from_iterator(iter, nrows, ncols, axis)
    // axis == 0 means values are consumed in row-major and DenseMatrix will be created row-major=false internally,
    // which DenseMatrix::from_iterator translates correctly into its internal representation. [1]
    DenseMatrix::from_iterator(a.iter().copied(), a.nrows(), a.ncols(), 0)
}

Why this works

• smartcore defines a generic Array2::from_iterator for DenseMatrix that builds a matrix from any iterator of T with explicit shape and an axis flag to reconcile layout. Passing axis=0 matches the straightforward row-wise iteration over ndarray’s elements produced by a.iter().[1]
• DenseMatrix’s Array2::from_iterator is implemented as DenseMatrix::new(nrows, ncols, iter.collect(), axis != 0), and DenseMatrix::new enforces shape consistency; this path is the same used throughout the library for safe construction.[1]

If a borrowed view is needed

• For ArrayView<'_, T, Ix2> (or ArrayViewMut), convert to an owned array first: let owned = view.to_owned(); then call ndarray_to_densematrix(&owned).[1]

Optional: return Result instead of panicking

pub fn from_ndarray2_result<T>(
    a: &ArrayBase<OwnedRepr<T>, Ix2>,
) -> Result<DenseMatrix<T>, smartcore::error::Failed>
where
    T: Debug + Display + Copy + Sized,
{
    // Use the same constructor but avoid unwrap; if you prefer, replicate logic with DenseMatrix::new.
    Ok(DenseMatrix::from_iterator(a.iter().copied(), a.nrows(), a.ncols(), 0))
}

Notes

• This copies data; a zero-copy adapter would require exposing DenseMatrix internals or constructing a DenseMatrix over ndarray’s buffer with exact column-major constraints. That’s not exposed in the current public API, so copying is the safe route.
• smartcore already offers ndarray bindings to operate on ndarray-native matrices directly by implementing Array2 for ndarray::Array2, so in many cases DenseMatrix isn’t necessary — algorithms accept types implementing the Array2 trait. Prefer passing ndarray arrays directly to smartcore algorithms when possible.