Bump ndarray to 0.15.0

[YuhanLiin]

Addresses #124.
Having issues with new version of sprs. The compile error most likely has something to do with sparsemat/sprs#278 and was introduced by sparsemat/sprs#290.

[bytesnake]

replacing the operator with sprs::binop::csmat_binop(mat.view(), mat.transpose_view(), |x, y| x.add(*y)); seems to work as a workaround. The difference between making the explicit call to add and the operator makes the difference: https://docs.rs/sprs/0.11.0/src/sprs/sparse/binop.rs.html#63

[bytesnake]

the construction of the sparse matrix fails though in the line above and I'm not yet sure why

[bytesnake]

the new constructor assumes ordered indices, this solves it:

    // create CSR matrix from data, indptr and indices
    let mat = CsMatBase::new_from_unsorted((n_points, n_points), indptr, indices, data).unwrap();
    let mut mat = sprs::binop::csmat_binop(mat.view(), mat.transpose_view().to_other_storage().view(), |x, y| x.add(*y));

[mulimoen]

It seems there are some strange type inference going on when using linfa::Float in sprs. You can use the following code instead:

    // create CSR matrix from data, indptr and indices
    let mat = CsMat::<F>::new_from_unsorted((n_points, n_points), indptr, indices, data).unwrap();
    let transpose = mat.transpose_view().to_other_storage();
    sprs::binop::csmat_binop(mat.view(), transpose.view(), |_, _| F::one())

This does the same as we do behind the scene. I have assumed x + y != 0, and moved this computation into binop, which might be more performant

[bytesnake]

👍

[bytesnake]

the CI pipeline somehow failed to run here as well, I checked manually ..