lack of functionality in "unravel" function

[tort-dla-psa]

In numpy we have this:

>>>np.unravel_index(0, [1,32,32,1])
(0, 0, 0, 0)

>>>np.unravel_index([0,1,2], [1,32,32,1])
(array([0, 0, 0]), array([0, 0, 0]), array([0, 1, 2]), array([0, 0, 0]))

But in xtensor we can only use unravel with one index. Also we'll get by this not an xarray of
dimensions, but this:

"xt::get_strides_t<std::vector<int, std::allocator>>"

Why is it so?

[tort-dla-psa]

also there's nothing about unravel in the docs

[tort-dla-psa]

there is a closed issue about unravel docs, but still no docs for it #1309

[tdegeus]

As far as I can see there is an xt::unravel_indices(...) to do what you want. It's output can be converted to a 'normal' matrix using xt::from_indices(...). E.g.

xt::xtensor<size_t,2> a = xt::zeros<size_t>({5, 2});
xt::xtensor<size_t,1> i = {0, 1, 2, 3};
xt::xtensor<size_t,2> idx = xt::from_indices(xt::unravel_indices(i, a.shape()));
std::cout << idx << std::endl;

But I do agree that we need to work on consistency with NumPy here. Also I feel that the overloads are not super flexible.