It seems that your fix does not work as intended (for released versions of scipy).

BTW, could you please report the problem upstream to notify scipy developers of the breaking change? It's probably not intended and maybe we shouldn't have to change anything in the scikit-learn code base.

For reference, here is the failure we observed with scipy-dev:

    def test_uniform_weights():
        # with uniform weights, results should be identical to stats.mode
        rng = np.random.RandomState(0)
        x = rng.randint(10, size=(10, 5))
        weights = np.ones(x.shape)
    
        for axis in (None, 0, 1):
            mode, score = stats.mode(x, axis)
            mode2, score2 = weighted_mode(x, weights, axis=axis)
    
>           assert_array_equal(mode, mode2)
E           AssertionError: 
E           Arrays are not equal
E           
E           (shapes (5,), (1, 5) mismatch)
E            x: array([3, 3, 7, 2, 1])
E            y: array([[3., 3., 7., 2., 1.]])

while this used to pass with previous versions of scipy.

I reported the change of shape in the return value of stats.mode as scipy/scipy#16418.

It seems that your fix does not work as intended (for released versions of scipy).

Adapted test to account for scipy-version along w/ a clarifying comment. I could have done it implicitly, e.g. by checking ndims but I figure this is more transparent.