Remove __array__() method

[sgillies]

This package shouldn't depend on Numpy. Instead, users should reshape their Numpy arrays created from Affine instances.

My second favorite approach would be to implement a .2d attribute (or a better name) so that users would do this: np.array(Affine().2d).

[mwtoews]

Sure, no big loss removing __array__. It's just a convenience when comparing against NumPy functions, as retained here in the tests.

This revert keeps the flat (dimensionless) structure the same as previous versions. I just have a suggestion to keep a test for the expected behavior of np.array(tfm) (i.e. that it doesn't return a 3x3 matrix).

[mwtoews]

Perhaps retain a simple test that describes what's expected of numpy, specifically something like this:

def test_array():
    six_args = (np.arange(6) + 1) / 10
    tfm = Affine(*six_args)
    ar = np.array(tfm)
    assert ar.shape == (9,)
    testing.assert_array_equal(ar, np.concat([six_args, (0, 0, 1)]))