def test_find_cluster(self):
x = np.array([1, 2, 9, 10]).reshape(-1, 1)
model = KMeans(x, 2)
model.mu = np.array([3, 10]).reshape(-1, 1)
cluster_index = model.find_cluster([[1]])
self.assertEqual(cluster_index, 0)
def test_find_random_centroid(self):
np.random.seed(0)
random_indexes = np.random.randint(low=0, high=self.n, size=self.k)
expected_centroids = self.x[random_indexes, :]
model = KMeans(self.x, self.k, random_state=0)
model.find_random_centroids()
np.testing.assert_array_equal(expected_centroids, model.mu)
def test_compute_cost(self):
x = np.array([1, 2, 9, 10]).reshape(-1, 1)
labels = np.array([0, 0, 1, 1])
mu = np.array([1.5, 9.5]).reshape(-1, 1)
k = 2
model = KMeans(x, k)
cost = model.compute_cost(mu, labels)
self.assertEqual(cost, 1)
def test_find_kpp_centroid(self):
expected_centroids = np.array([[13.87626757, 14.41023753],
[-0.7083392, -0.80309866],
[15.42080098, 15.77197743],
[0.59477214, -1.2883151]])
model = KMeans(self.x, self.k, random_state=0)
model.find_kpp_centroids()
np.testing.assert_allclose(model.mu, expected_centroids)
def test_get_cluster_data(self):
labels = np.zeros(self.n)
labels[np.random.randint(low=0, high=self.n, size=50)] = 1
expected_values = self.x[labels == 1]
k_means_kpp = KMeans(self.x, self.k, initialization=KMeans.K_MEANS_PP, eps=1e-5)
k_means_kpp.train()
x_cluster = k_means_kpp.get_cluster_data(1, labels)
np.testing.assert_array_equal(x_cluster, expected_values)
def test_train_with_init_mu(self):
init_mu = np.array([[1, 1],
[5, 5],
[10, 10],
[15, 15]])
model = KMeans(self.x, self.k)
model.find_kpp_centroids = MagicMock()
model.find_random_centroids = MagicMock()
model.train(init_mu)
self.assertFalse(model.find_kpp_centroids.called)
self.assertFalse(model.find_random_centroids.called)
def test_train_max_error_exceeded(self):
model = KMeans(self.x, self.k, max_iter=2)
model.train()
self.assertEqual(model.n_iter, 2)
def test_train_with_kpp(self):
k_means_kpp = KMeans(self.x, self.k, initialization=KMeans.K_MEANS_PP, eps=1e-5)
k_means_kpp.train()
self.assertEqual(len(np.unique(k_means_kpp.labels)), self.k)
def test_train_with_random(self):
k_means_random = KMeans(self.x, self.k, initialization=KMeans.RANDOM, eps=1e-5)
k_means_random.train()
self.assertEqual(len(np.unique(k_means_random.labels)), self.k)