def test5_filter_finite(self):
x_y_pairs = [(0, 0), (1, 1), (math.inf, 9), (9, math.inf),
(-math.inf, 2), (2, -math.inf), (3.2, -20),
(float("nan"), 0), (0, float("nan")), (7, 7),
(float("inf"), 100), (100, float("inf")), (4, 0),
(float("-inf"), 1), (1, float("inf")), (4, 0), (0.0, 0.1),
(-3.9, 27)]
computed = hac(x_y_pairs)
# hac should return an numpy array or matrix of the right shape
self.assertTrue(
isinstance(computed, np.ndarray)
or isinstance(computed, np.matrix))
self.assertEqual(np.shape(computed), (7, 4))
computed = np.array(computed)
# The third column should be increasing
for i in range(6):
self.assertGreaterEqual(computed[i + 1, 2], computed[i, 2])
# Verify hac operates exactly as linkage does
expected = linkage([(0, 0), (1, 1), (3.2, -20), (7, 7), (4, 0), (4, 0),
(0.0, 0.1), (-3.9, 27)])
self.assertTrue(np.all(np.isclose(computed, expected)))
def test_tiebreak(self):
x_y_pairs = get_x_y_pairs(tiebreak_csv_file)
computed = hac(x_y_pairs)
expected_cluster_sizes \
= [2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 4, 4, 4, 4, 4, 8, 8, 12, 20]
for i in range(np.shape(computed)[0]):
row = np.array(computed[i, :]).flatten()
self.assertEqual(row[0], 2 * i)
self.assertEqual(row[1], 2 * i + 1)
self.assertEqual(row[2], 0)
self.assertEqual(row[3], expected_cluster_sizes[i])
def test5_tiebreak(self):
x_y_pairs = get_x_y_pairs(tiebreak_csv_file)
computed = hac(x_y_pairs)
expected_cluster_sizes \
= [2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 4, 4, 4, 4, 4, 8, 8, 12, 20]
# chose lowest cluster index for the first position
# if still tied, chose lowest cluster index for the second position
for i in range(np.shape(computed)[0]):
row = np.array(computed[i, :]).flatten()
self.assertEqual(row[0], 2 * i)
self.assertEqual(row[1], 2 * i + 1)
self.assertEqual(row[2], 0)
self.assertEqual(row[3], expected_cluster_sizes[i])
def test_randomized(self):
x_y_pairs = get_x_y_pairs(random_csv_file)
computed = hac(x_y_pairs)
# hac should return an numpy array of the right shape
self.assertIsInstance(computed, np.ndarray)
self.assertEqual(np.shape(computed), (19, 4))
# The third column should be increasing
for i in range(18):
self.assertGreaterEqual(computed[i + 1, 2], computed[i, 2])
# Verify hac operates exactly as linkage does
expected = scipy.cluster.hierarchy.linkage(x_y_pairs)
self.assertTrue(np.all(np.isclose(computed, expected)))
def test7_more_than_20(self):
n_points = 80
x_y_pairs = [(x**2, x**2) for x in range(0, n_points)]
computed = hac(x_y_pairs)
computed = np.array(computed)
# The third column should be increasing
for i in range(n_points - 2):
self.assertGreaterEqual(computed[i + 1, 2], computed[i, 2])
# first row
self.assertTrue(np.allclose(computed[0], [0, 1, 2**.5, 2]))
for i, row in enumerate(computed[1:, :]):
self.assertEqual(row[0], i + 2)
self.assertEqual(row[1], i + n_points)
self.assertTrue(
np.isclose(row[2], ((i + i + 2)**2 + (i + i + 4)**2 - 2)**0.5))
self.assertEqual(row[3], i + 3)
def test3_pokemon_csv(self):
x_y_pairs = get_x_y_pairs(pokemon_csv_file)
computed = hac(x_y_pairs)
# hac should return an numpy array of the right shape
self.assertIsInstance(computed, np.ndarray)
self.assertEqual(np.shape(computed), (19, 4))
# The third column should be increasing
for i in range(18):
self.assertGreaterEqual(computed[i + 1, 2], computed[i, 2])
# Verify hac operates exactly as linkage does - giving leeway for tiebreaker
expected = linkage(x_y_pairs)
self.assertTrue(
np.allclose(computed[computed[:, 0].argsort()],
expected[expected[:, 0].argsort()]))
self.assertTrue(
np.allclose(computed[computed[:, 1].argsort()],
expected[expected[:, 1].argsort()]))