def prediction_rankings(self):
"""
Computes rank of predictions and if they are statistically different
"""
overall_data = filter_nan(nanmeanw(self.rest, 1)).reshape(
-1, self.rest.shape[2]
)
overall = rank_test_2d(overall_data)
steps = rank_test_3d(self.rest)
return overall, steps
def test_rank__singlestep__3series_friedman__different_data(dataset_1d):
arr = np.stack([dataset_1d, dataset_1d, dataset_1d - 1], 1)
res = rank_test_2d(arr)
assert res.ranks.shape[0] == 3
assert res.mean_ranks.shape[0] == 3
assert res.equality_bool_mask.shape[0] == 3
assert np.array_equal(res.ranks, np.array([3, 1, 2]))
assert np.allclose(res.equality_bool_mask, np.array([True, True, False]))
def test_rank__singlestep__3series_friedman__identical_data(dataset_1d):
arr = np.stack([dataset_1d, dataset_1d, dataset_1d], 1)
res = rank_test_2d(arr)
assert res.ranks.shape[0] == 3
assert res.mean_ranks.shape[0] == 3
assert res.equality_bool_mask.shape[0] == 3
assert np.array_equal(res.ranks, np.array([1, 2, 3]))
assert np.allclose(res.equality_bool_mask,
np.full_like(res.equality_bool_mask, True))
def test_rank__singlestep__2series_mw__different_data(dataset_1d):
arr = np.stack([dataset_1d + 1, dataset_1d], 1)
res = rank_test_2d(arr)
assert res.ranks.shape[0] == 2
assert res.mean_ranks.shape[0] == 2
assert res.equality_bool_mask.shape[0] == 2
assert np.array_equal(res.ranks, np.array([2, 1]))
assert np.allclose(res.equality_bool_mask,
np.full_like(res.equality_bool_mask, False))
def test_rank__1d_fail(dataset_1d):
with pytest.raises(AssertionError):
rank_test_2d(dataset_1d)
with pytest.raises(AssertionError):
rank_test_3d(dataset_1d)