def test_kmeans(self):
labels, _centroids = kmeans.kmeans(DATA, 2, n_runs=5)
labels = list(labels)
# K-means is invariant by label permutations, hence we can always have two
# different classifications
assert labels == [0, 0, 0, 1, 1, 1] or labels == [1, 1, 1, 0, 0, 0]
def test_kmeans_convergence(self):
# If no guidance is given, convergence is unlikely to happen
labels, clusters = kmeans.kmeans(DATA, 2, max_iter=1, n_runs=2)
assert (labels != [0, 0, 0, 1, 1, 1]).any()
def test_trivial_classification_if_number_of_clusters_equal_to_number_of_samples(
self
):
labels, clusters = kmeans.kmeans(DATA, len(DATA), n_runs=5)
assert_equal(labels, range(len(DATA)))
assert_equal(clusters, DATA)
def test_kmeans_requires_k_gt_n(self):
with pytest.raises(ValueError):
kmeans.kmeans(DATA, len(DATA) + 1, n_runs=5)