convert a vector<float> to ndarray does not work correctly

[bpkcl630]

Hi,

I have a trouble when using boost 1.64.0 to convert a vector to a 2D ndarray.

I try to use code like this

np::ndarray pyArray = np::from_data(vec.data(), np::dtype::get_builtin<float>(), np::tuple(rows, cols), py::make_tuple(cols, 1), py::object() );

when I return it to python the result is not same as C++. Then I change this to another way which is

np::ndarray result = np::zeros(np::tuple(rows, cols), np::dtype::get_builtin<float>());
std::copy(vec.begin(), vec.end(), reinterpret_cast<float*>(result.get_data()));

and it is works correctly now.

I am not sure it is a bug or I use the from_data() in a wrong way.

Thanks a lot for your help!

[bpkcl630]

forgot to mention I am on 64bit Windows 10.

[stefanseefeld]

I suspect the error is that you specify strides in units of elements, when it really needs to be units of bytes. Please replace py::make_tuple(cols, 1) by py::make_tuple(cols*sizeof(float), sizeof(float)), and see whether that gives you the right data.

[bpkcl630]

Thanks a lot! That do works.