It started failing one day ago or so in the scikit-learn tests using the development numpy wheels from https://travis-dev-wheels.scipy.org/.

I can reproduce the error on numpy master with this snippet:

import numpy as np

my_list = [[np.array([1, 2]),
            np.array([1, 2, 3])]]

np.testing.assert_array_equal(my_list, my_list)

The expected behaviour is that there is no exception since I compare my_list to itself. The actual behaviour is is an AssertionError with numpy master.

The same snippet runs fine with numpy 1.11.2.