Create diagonal matrix in python::numpy

This is probably a feature request.
What is the correct way to create diagonal matrix in boost::python::numpy?

Of course, I can just create a usual 2D matrix then assign its diagonal. But is there a better way?

It seems that in numpy (in python), the diagonal matrix is stored in a compact format, e.g. only stores the diagonal data. This can be observed by `a = np.diag(np.random.rand(20000,1))` and watch the memory usage. (much smaller than 20000^2 * 8 / 2^20 MB)

It would be nice to have a method to build diagonal matrix in this "compact" format.