Implementation of matrix norm for order greater than 2

When using numpy.linalg.norm with ord different from 2, for example 3, then I expected the norm to be computed according to the documentation, i.e. as sum(abs(x)**ord)**(1./ord)

But instead I got:

ValueError                                Traceback (most recent call last)
<ipython-input-784-3344f4d8d663> in <module>()
----> 1 nnorm((distpdfs_diverg['BC']/n_samples) - np.eye(n_bundles), 3)

/usr/lib/python3.5/site-packages/numpy/linalg/linalg.py in norm(x, ord, axis, keepdims)
   2247             ret = _multi_svd_norm(x, row_axis, col_axis, sum)
   2248         else:
-> 2249             raise ValueError("Invalid norm order for matrices.")
   2250         if keepdims:
   2251             ret_shape = list(x.shape)

ValueError: Invalid norm order for matrices.

Looking at the source code (numpy 1.13, numpy 1.14, and current) it seems it is simply the norm is simply not implemented for ord greater than 2 contrarily to what the doc says. Would that be something difficult to do?

The relevant implementation is numpy/linalg/linalg.py, identified in the traceback and linked source; start by reading norm's matrix-order handling and the surrounding documentation. Add coverage for matrix ord values above 2 and confirm the documented result against the reported formula.