class TestSoinn(unittest.TestCase):
def setUp(self):
self.soinn = Soinn()
self.soinn.nodes = np.array([[0, 0], [1, 0], [1, 1], [0, 1]]
, dtype=np.float64)
self.soinn.dim = 2
self.soinn.adjacent_mat = dok_matrix((4, 4))
self.soinn.winning_times = [1] * 4
def test_skleran_api(self):
check_clustering('Soinn', Soinn)
def test_fit(self):
pass
def test_check_signal(self):
self.soinn = Soinn()
signal = 'hoge'
self.assertRaises(TypeError, self.soinn._Soinn__check_signal, signal)
signal = np.arange(6).reshape(2, 3)
self.assertRaises(TypeError, self.soinn._Soinn__check_signal, signal)
d = 6
signal = np.arange(d)
self.soinn._Soinn__check_signal(signal)
self.assertTrue(hasattr(self.soinn, 'dim'))
self.assertEqual(self.soinn.dim, d)
signal = np.arange(d + 1)
self.assertRaises(TypeError, self.soinn._Soinn__check_signal, signal)
signal = [i for i in range(d)]
self.soinn._Soinn__check_signal(signal)
def test_add_node(self):
self.soinn = Soinn()
self.soinn.dim = 2 # dim have to be set before calling __add_node()
signals = np.random.random((4, 2))
for n in range(signals.shape[0]):
signal = signals[n, :]
self.soinn._Soinn__add_node(signal)
self.assertEqual(len(self.soinn.winning_times), n + 1)
self.assertEqual(self.soinn.nodes.shape, (n + 1, self.soinn.dim))
np.testing.assert_array_equal(self.soinn.nodes[n, :], signal)
self.assertEqual(self.soinn.adjacent_mat.shape, (n + 1, n + 1))
self.assertEqual(self.soinn.adjacent_mat.nnz, 0)
def test_add_edge(self):
s1, s2 = 1, 2
self.soinn._Soinn__add_edge((s1, s2))
self.assertEqual(self.soinn.adjacent_mat.nnz, 2)
self.assertEqual(self.soinn.adjacent_mat[s1, s2], 1)
def test_increment_edge_ages(self):
self.soinn.adjacent_mat[0, 1:3] = 1
self.soinn.adjacent_mat[1:3, 0] = 1
self.soinn._Soinn__increment_edge_ages(0)
expected = dok_matrix([[0, 2, 2, 0], [2, 0, 0, 0], [2, 0, 0, 0], [0, 0, 0, 0]])
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(), expected.toarray())
self.soinn._Soinn__increment_edge_ages(1)
expected = dok_matrix([[0, 3, 2, 0], [3, 0, 0, 0], [2, 0, 0, 0], [0, 0, 0, 0]])
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(), expected.toarray())
def test_delete_old_edges(self):
self.soinn.winning_times = [i for i in range(4)]
m = self.soinn.max_edge_age
self.soinn.adjacent_mat[[0, 1], [1, 0]] = m + 2
self.soinn.adjacent_mat[[0, 2], [2, 0]] = m + 1
self.soinn._Soinn__delete_old_edges(0)
actual = self.soinn.adjacent_mat.toarray()
expected = dok_matrix([[0, m+1, 0], [m+1, 0, 0], [0, 0, 0]]).toarray()
np.testing.assert_array_equal(actual, expected)
expected = np.array([[0, 0], [1, 1], [0, 1]], dtype=np.float64)
np.testing.assert_array_equal(self.soinn.nodes, expected)
self.assertEqual(self.soinn.winning_times, [0, 2, 3])
def test_delete_old_edges_with_deleting_no_node(self):
# No node is deleted by the function
self.soinn.winning_times = [i for i in range(4)]
m = self.soinn.max_edge_age
self.soinn.adjacent_mat[[0, 1], [1, 0]] = m + 2
self.soinn.adjacent_mat[[1, 2], [2, 1]] = 1
previous_nodes = self.soinn.nodes
previous_winning_times = self.soinn.winning_times
self.soinn._Soinn__delete_old_edges(0)
actual = self.soinn.adjacent_mat.toarray()
expected = dok_matrix([[0, 0, 0, 0], [0, 0, 1, 0], [0, 1, 0, 0], [0, 0, 0, 0]]).toarray()
np.testing.assert_array_equal(actual, expected)
np.testing.assert_array_equal(self.soinn.nodes, previous_nodes)
self.assertEqual(self.soinn.winning_times, previous_winning_times)
def test_delete_old_edges_with_deleting_several_nodes(self):
# delete several nodes simultaneously
self.soinn.winning_times = [i for i in range(4)]
m = self.soinn.max_edge_age
self.soinn.adjacent_mat[[0, 1, 0, 3], [1, 0, 3, 0]] = m + 2
self.soinn.adjacent_mat[[0, 2], [2, 0]] = m + 1
self.soinn._Soinn__delete_old_edges(0)
actual = self.soinn.adjacent_mat.toarray()
expected = dok_matrix([[0, m+1], [m+1, 0]]).toarray()
np.testing.assert_array_equal(actual, expected)
self.assertEqual(self.soinn.winning_times, [0, 2])
def test_delete_old_edges_with_changing_winner_index(self):
# the winner index is changed by deleting nodes
self.soinn.winning_times = [i for i in range(4)]
expected = self.soinn.nodes[3, :]
m = self.soinn.max_edge_age
self.soinn.adjacent_mat[[3, 0, 3, 2], [0, 3, 2, 3]] = m + 2
self.soinn.adjacent_mat[[3, 1], [1, 3]] = m + 1
winner_index = self.soinn._Soinn__delete_old_edges(3)
np.testing.assert_array_equal(self.soinn.nodes[winner_index, :], expected)
def test_delete_nodes(self):
self.soinn.winning_times = [i for i in range(4)]
self.soinn.adjacent_mat[[0, 1], [1, 0]] = 1
self.soinn.adjacent_mat[[2, 3], [3, 2]] = 2
self.soinn._Soinn__delete_nodes([1])
expected = np.array([[0, 0], [1, 1], [0, 1]], dtype=np.float64)
np.testing.assert_array_equal(self.soinn.nodes, expected)
self.assertEqual(self.soinn.winning_times, [0, 2, 3])
expected = dok_matrix([[0, 0, 0], [0, 0, 2], [0, 2, 0]]).toarray()
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(), expected)
def test_delete_nodes_with_deleting_several_nodes(self):
# delete several nodes simultaneously
self.soinn.winning_times = [i for i in range(4)]
self.soinn.adjacent_mat[[0, 1], [1, 0]] = 1
self.soinn.adjacent_mat[[2, 3], [3, 2]] = 2
self.soinn._Soinn__delete_nodes([1, 3])
expected = np.array([[0, 0], [1, 1]], dtype=np.float64)
np.testing.assert_array_equal(self.soinn.nodes, expected)
self.assertEqual(self.soinn.winning_times, [0, 2])
expected = dok_matrix((2, 2)).toarray()
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(), expected)
def test_delete_noise_nodes(self):
self.soinn.adjacent_mat[0, 2:] = 1
self.soinn.adjacent_mat[2:, 0] = 1
self.soinn.winning_times = [2, 1, 2, 1]
self.soinn._Soinn__delete_noise_nodes()
np.testing.assert_array_equal(self.soinn.nodes, [[0, 0], [1, 1], [0, 1]])
self.assertEqual(self.soinn.winning_times, [2, 2, 1])
expected = [[0, 1, 1], [1, 0, 0], [1, 0, 0]]
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(), expected)
def test_delete_nodes_from_adjacent_mat(self):
self.soinn.adjacent_mat[0, 1] = 1
self.soinn.adjacent_mat[1, 0] = 1
self.soinn.adjacent_mat[0, 2] = 2
self.soinn.adjacent_mat[2, 0] = 2
self.soinn.adjacent_mat[2, 3] = 3
self.soinn.adjacent_mat[3, 2] = 3
indexes = [1]
expected1 = self.soinn.adjacent_mat[np.ix_([0, 2, 3], [0, 2, 3])]
self.soinn._Soinn__delete_nodes_from_adjacent_mat(indexes, 4, 3)
expected2 = [[0, 2, 0], [2, 0, 3], [0, 3, 0]]
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(), expected1.toarray())
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(), expected2)
def test_performance_of_update_adjacent_mat(self):
n = 500
p = 0.01
a = dok_matrix((n, n))
for c in range(n):
i = np.random.randint(n)
j = np.random.randint(n)
if i == j:
continue
a[i, j] += 1
indexes = np.random.randint(n, size=int(p * n)).tolist()
indexes = sorted(list(set(indexes)))
remain_indexes = list(set([i for i in range(n)]) - set(indexes))
self.soinn.adjacent_mat = a.copy()
start = time.time()
expected = a[np.ix_(remain_indexes, remain_indexes)]
expected_time = time.time() - start
start = time.time()
self.soinn._Soinn__delete_nodes_from_adjacent_mat(indexes, n, len(remain_indexes))
actual_time = time.time() - start
self.assertEqual(self.soinn.adjacent_mat.shape, expected.shape)
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(), expected.toarray())
#print('np._ix: ', expected_time)
#print('__update_adjacent_mat:', actual_time)
self.assertLess(actual_time, expected_time)
def test_find_nearest_nodes(self):
signal = np.array([-1, 0])
indexes, sq_dists = self.soinn._Soinn__find_nearest_nodes(1, signal)
self.assertEqual(indexes, [0])
self.assertEqual(sq_dists, [1])
indexes, sq_dists = self.soinn._Soinn__find_nearest_nodes(2, signal)
self.assertEqual(indexes, [0, 3])
self.assertEqual(sq_dists, [1, 2])
def test_calculate_similarity_thresholds(self):
self.soinn.adjacent_mat[0, 2:] = 1
self.soinn.adjacent_mat[2:, 0] = 1
self.assertEqual(self.soinn._Soinn__calculate_similarity_thresholds([0]), [2])
self.assertEqual(self.soinn._Soinn__calculate_similarity_thresholds([0, 1]), [2, 1])
def test_update_winner(self):
self.soinn.winning_times[0] = 2 - 1
np.testing.assert_array_equal(self.soinn.nodes[0, :], [0, 0])
self.soinn._Soinn__update_winner(0, np.array([-1, 0], dtype=np.float64))
np.testing.assert_array_equal(self.soinn.nodes[0, :], [-0.5, 0])
self.assertEqual(self.soinn.winning_times[0], 2)
self.soinn.winning_times[1] = 3 - 1
np.testing.assert_array_equal(self.soinn.nodes[1, :], [1, 0])
self.soinn._Soinn__update_winner(1, np.array([2, 0], dtype=np.float64))
np.testing.assert_array_equal(self.soinn.nodes[1, :], [1 + 1/3, 0])
self.assertEqual(self.soinn.winning_times[1], 3)
def test_update_adjacent_nodes(self):
self.soinn.adjacent_mat[0, 1:3] = 1
self.soinn.adjacent_mat[1:3, 0] = 1
np.testing.assert_array_equal(self.soinn.nodes[1], [1, 0])
np.testing.assert_array_equal(self.soinn.nodes[2], [1, 1])
self.soinn._Soinn__update_adjacent_nodes(0, np.array([-1, 0]))
np.testing.assert_array_equal(self.soinn.nodes[1], [1 - 2/100, 0])
np.testing.assert_array_equal(self.soinn.nodes[2], [1 - 2/100, 1 - 1/100])
np.testing.assert_array_equal(self.soinn.nodes[0], [0, 0])
np.testing.assert_array_equal(self.soinn.nodes[3], [0, 1])
@unittest.skip('skip test_input_signal() because it takes much time.')
def test_input_signal(self):
n = 5000
loop_num = 10
elapsed_time = 0
for l in range(loop_num):
data = np.random.rand(n, 2)
self.soinn = Soinn()
start = time.time()
for i in range(n):
self.soinn.input_signal(data[i])
elapsed_time += time.time() - start
elapsed_time /= loop_num
print('average time: ', str(elapsed_time), 'sec')
print(self.soinn)
def test_label_nodes(self):
self.soinn._Soinn__add_node([0, 2])
self.soinn._Soinn__add_node([1, 2])
self.soinn._Soinn__add_edge([0, 2])
self.soinn._Soinn__add_edge([0, 3])
self.soinn._Soinn__add_edge([1, 4])
labels = self.soinn._Soinn__label_nodes()
self.assertEqual(len(labels), self.soinn.nodes.shape[0])
self.assertEqual(labels[0], labels[2])
self.assertEqual(labels[0], labels[3])
self.assertNotEqual(labels[0], -1)
self.assertEqual(labels[1], -1)
self.assertEqual(labels[4], -1)
self.assertEqual(labels[5], -1)
labels = self.soinn._Soinn__label_nodes(min_cluster_size=2)
self.assertEqual(len(labels), self.soinn.nodes.shape[0])
self.assertEqual(labels[0], labels[2])
self.assertEqual(labels[0], labels[3])
self.assertNotEqual(labels[0], -1)
self.assertEqual(labels[1], labels[4])
self.assertNotEqual(labels[1], -1)
self.assertEqual(labels[5], -1)
def test_label_samples(self):
# There are 6 nodes and nodes indexed 0, 2 and 3 make a cluster.
self.soinn._Soinn__add_node([0, 2])
self.soinn._Soinn__add_node([1, 2])
self.soinn._Soinn__add_edge([0, 2])
self.soinn._Soinn__add_edge([0, 3])
self.soinn._Soinn__add_edge([1, 4])
actual = self.soinn._Soinn__label_samples(np.array([[-0.1, -0.1],
[1.1, 1.1],
[100, 100]]))
expected = np.array(
[self.soinn.node_labels[0]] * 2 + [Soinn.NOISE_LABEL])
assert_array_equal(actual, expected)
class TestSoinn(unittest.TestCase):
def setUp(self):
self.soinn = Soinn()
self.soinn.nodes = np.array([[0, 0], [1, 0], [1, 1], [0, 1]],
dtype=np.float64)
self.soinn.adjacent_mat = dok_matrix((4, 4))
self.soinn.winning_times = [1] * 4
def test_check_signal(self):
self.soinn = Soinn()
signal = [0, 1, 2]
self.assertRaises(TypeError, self.soinn._Soinn__check_signal, signal)
signal = np.arange(6).reshape(2, 3)
self.assertRaises(TypeError, self.soinn._Soinn__check_signal, signal)
d = 6
signal = np.arange(d)
self.soinn._Soinn__check_signal(signal)
self.assertTrue(hasattr(self.soinn, 'dim'))
self.assertEqual(self.soinn.dim, d)
signal = np.arange(d + 1)
self.assertRaises(TypeError, self.soinn._Soinn__check_signal, signal)
def test_add_node(self):
self.soinn = Soinn()
self.soinn.dim = 2 # dim have to be set before calling __add_node()
signals = np.random.random((4, 2))
for n in range(signals.shape[0]):
signal = signals[n, :]
self.soinn._Soinn__add_node(signal)
self.assertEqual(len(self.soinn.winning_times), n + 1)
self.assertEqual(self.soinn.nodes.shape, (n + 1, self.soinn.dim))
np.testing.assert_array_equal(self.soinn.nodes[n, :], signal)
self.assertEqual(self.soinn.adjacent_mat.shape, (n + 1, n + 1))
self.assertEqual(self.soinn.adjacent_mat.nnz, 0)
def test_add_edge(self):
s1, s2 = 1, 2
self.soinn._Soinn__add_edge((s1, s2))
self.assertEqual(self.soinn.adjacent_mat.nnz, 2)
self.assertEqual(self.soinn.adjacent_mat[s1, s2], 1)
def test_increment_edge_ages(self):
self.soinn.adjacent_mat[0, 1:3] = 1
self.soinn.adjacent_mat[1:3, 0] = 1
self.soinn._Soinn__increment_edge_ages(0)
expected = dok_matrix([[0, 2, 2, 0], [2, 0, 0, 0], [2, 0, 0, 0],
[0, 0, 0, 0]])
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(),
expected.toarray())
self.soinn._Soinn__increment_edge_ages(1)
expected = dok_matrix([[0, 3, 2, 0], [3, 0, 0, 0], [2, 0, 0, 0],
[0, 0, 0, 0]])
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(),
expected.toarray())
def test_delete_old_edges(self):
self.soinn.winning_times = [i for i in range(4)]
m = self.soinn.max_edge_age
self.soinn.adjacent_mat[[0, 1], [1, 0]] = m + 2
self.soinn.adjacent_mat[[0, 2], [2, 0]] = m + 1
self.soinn._Soinn__delete_old_edges(0)
actual = self.soinn.adjacent_mat.toarray()
expected = dok_matrix([[0, m + 1, 0], [m + 1, 0, 0], [0, 0,
0]]).toarray()
np.testing.assert_array_equal(actual, expected)
expected = np.array([[0, 0], [1, 1], [0, 1]], dtype=np.float64)
np.testing.assert_array_equal(self.soinn.nodes, expected)
self.assertEqual(self.soinn.winning_times, [0, 2, 3])
def test_delete_old_edges_with_deleting_no_node(self):
# No node is deleted by the function
self.soinn.winning_times = [i for i in range(4)]
m = self.soinn.max_edge_age
self.soinn.adjacent_mat[[0, 1], [1, 0]] = m + 2
self.soinn.adjacent_mat[[1, 2], [2, 1]] = 1
previous_nodes = self.soinn.nodes
previous_winning_times = self.soinn.winning_times
self.soinn._Soinn__delete_old_edges(0)
actual = self.soinn.adjacent_mat.toarray()
expected = dok_matrix([[0, 0, 0, 0], [0, 0, 1, 0], [0, 1, 0, 0],
[0, 0, 0, 0]]).toarray()
np.testing.assert_array_equal(actual, expected)
np.testing.assert_array_equal(self.soinn.nodes, previous_nodes)
self.assertEqual(self.soinn.winning_times, previous_winning_times)
def test_delete_old_edges_with_deleting_several_nodes(self):
# delete several nodes simultaneously
self.soinn.winning_times = [i for i in range(4)]
m = self.soinn.max_edge_age
self.soinn.adjacent_mat[[0, 1, 0, 3], [1, 0, 3, 0]] = m + 2
self.soinn.adjacent_mat[[0, 2], [2, 0]] = m + 1
self.soinn._Soinn__delete_old_edges(0)
actual = self.soinn.adjacent_mat.toarray()
expected = dok_matrix([[0, m + 1], [m + 1, 0]]).toarray()
np.testing.assert_array_equal(actual, expected)
self.assertEqual(self.soinn.winning_times, [0, 2])
def test_delete_old_edges_with_changing_winner_index(self):
# the winner index is changed by deleting nodes
self.soinn.winning_times = [i for i in range(4)]
expected = self.soinn.nodes[3, :]
m = self.soinn.max_edge_age
self.soinn.adjacent_mat[[3, 0, 3, 2], [0, 3, 2, 3]] = m + 2
self.soinn.adjacent_mat[[3, 1], [1, 3]] = m + 1
winner_index = self.soinn._Soinn__delete_old_edges(3)
np.testing.assert_array_equal(self.soinn.nodes[winner_index, :],
expected)
def test_delete_nodes(self):
self.soinn.winning_times = [i for i in range(4)]
self.soinn.adjacent_mat[[0, 1], [1, 0]] = 1
self.soinn.adjacent_mat[[2, 3], [3, 2]] = 2
self.soinn._Soinn__delete_nodes([1])
expected = np.array([[0, 0], [1, 1], [0, 1]], dtype=np.float64)
np.testing.assert_array_equal(self.soinn.nodes, expected)
self.assertEqual(self.soinn.winning_times, [0, 2, 3])
expected = dok_matrix([[0, 0, 0], [0, 0, 2], [0, 2, 0]]).toarray()
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(),
expected)
def test_delete_nodes_with_deleting_several_nodes(self):
# delete several nodes simultaneously
self.soinn.winning_times = [i for i in range(4)]
self.soinn.adjacent_mat[[0, 1], [1, 0]] = 1
self.soinn.adjacent_mat[[2, 3], [3, 2]] = 2
self.soinn._Soinn__delete_nodes([1, 3])
expected = np.array([[0, 0], [1, 1]], dtype=np.float64)
np.testing.assert_array_equal(self.soinn.nodes, expected)
self.assertEqual(self.soinn.winning_times, [0, 2])
expected = dok_matrix((2, 2)).toarray()
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(),
expected)
def test_delete_noise_nodes(self):
self.soinn.adjacent_mat[0, 2:] = 1
self.soinn.adjacent_mat[2:, 0] = 1
self.soinn.winning_times = [2, 1, 2, 1]
self.soinn._Soinn__delete_noise_nodes()
np.testing.assert_array_equal(self.soinn.nodes,
[[0, 0], [1, 1], [0, 1]])
self.assertEqual(self.soinn.winning_times, [2, 2, 1])
expected = [[0, 1, 1], [1, 0, 0], [1, 0, 0]]
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(),
expected)
def test_delete_nodes_from_adjacent_mat(self):
self.soinn.adjacent_mat[0, 1] = 1
self.soinn.adjacent_mat[1, 0] = 1
self.soinn.adjacent_mat[0, 2] = 2
self.soinn.adjacent_mat[2, 0] = 2
self.soinn.adjacent_mat[2, 3] = 3
self.soinn.adjacent_mat[3, 2] = 3
indexes = [1]
expected1 = self.soinn.adjacent_mat[np.ix_([0, 2, 3], [0, 2, 3])]
self.soinn._Soinn__delete_nodes_from_adjacent_mat(indexes, 4, 3)
expected2 = [[0, 2, 0], [2, 0, 3], [0, 3, 0]]
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(),
expected1.toarray())
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(),
expected2)
def test_performance_of_update_adjacent_mat(self):
n = 500
p = 0.01
a = dok_matrix((n, n))
for c in range(n):
i = np.random.randint(n)
j = np.random.randint(n)
if i == j:
continue
a[i, j] += 1
indexes = np.random.randint(n, size=int(p * n)).tolist()
indexes = sorted(list(set(indexes)))
remain_indexes = list(set([i for i in range(n)]) - set(indexes))
self.soinn.adjacent_mat = a.copy()
start = time.time()
expected = a[np.ix_(remain_indexes, remain_indexes)]
expected_time = time.time() - start
start = time.time()
self.soinn._Soinn__delete_nodes_from_adjacent_mat(
indexes, n, len(remain_indexes))
actual_time = time.time() - start
self.assertEqual(self.soinn.adjacent_mat.shape, expected.shape)
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(),
expected.toarray())
#print('np._ix: ', expected_time)
#print('__update_adjacent_mat:', actual_time)
self.assertLess(actual_time, expected_time)
def test_find_nearest_nodes(self):
signal = np.array([-1, 0])
indexes, sq_dists = self.soinn._Soinn__find_nearest_nodes(1, signal)
self.assertEqual(indexes, [0])
self.assertEqual(sq_dists, [1])
indexes, sq_dists = self.soinn._Soinn__find_nearest_nodes(2, signal)
self.assertEqual(indexes, [0, 3])
self.assertEqual(sq_dists, [1, 2])
def test_calculate_similarity_thresholds(self):
self.soinn.adjacent_mat[0, 2:] = 1
self.soinn.adjacent_mat[2:, 0] = 1
self.assertEqual(
self.soinn._Soinn__calculate_similarity_thresholds([0]), [2])
self.assertEqual(
self.soinn._Soinn__calculate_similarity_thresholds([0, 1]), [2, 1])
def test_update_winner(self):
self.soinn.winning_times[0] = 2 - 1
np.testing.assert_array_equal(self.soinn.nodes[0, :], [0, 0])
self.soinn._Soinn__update_winner(0, np.array([-1, 0],
dtype=np.float64))
np.testing.assert_array_equal(self.soinn.nodes[0, :], [-0.5, 0])
self.assertEqual(self.soinn.winning_times[0], 2)
self.soinn.winning_times[1] = 3 - 1
np.testing.assert_array_equal(self.soinn.nodes[1, :], [1, 0])
self.soinn._Soinn__update_winner(1, np.array([2, 0], dtype=np.float64))
np.testing.assert_array_equal(self.soinn.nodes[1, :], [1 + 1 / 3, 0])
self.assertEqual(self.soinn.winning_times[1], 3)
def test_update_adjacent_nodes(self):
self.soinn.adjacent_mat[0, 1:3] = 1
self.soinn.adjacent_mat[1:3, 0] = 1
np.testing.assert_array_equal(self.soinn.nodes[1], [1, 0])
np.testing.assert_array_equal(self.soinn.nodes[2], [1, 1])
self.soinn._Soinn__update_adjacent_nodes(0, np.array([-1, 0]))
np.testing.assert_array_equal(self.soinn.nodes[1], [1 - 2 / 100, 0])
np.testing.assert_array_equal(self.soinn.nodes[2],
[1 - 2 / 100, 1 - 1 / 100])
np.testing.assert_array_equal(self.soinn.nodes[0], [0, 0])
np.testing.assert_array_equal(self.soinn.nodes[3], [0, 1])
@unittest.skip('skip test_input_signal() because it takes much time.')
def test_input_signal(self):
n = 5000
loop_num = 10
elapsed_time = 0
for l in range(loop_num):
data = np.random.rand(n, 2)
self.soinn = Soinn()
start = time.time()
for i in range(n):
self.soinn.input_signal(data[i])
elapsed_time += time.time() - start
elapsed_time /= loop_num
print('average time: ', str(elapsed_time), 'sec')
print(self.soinn)
class TestSoinn(unittest.TestCase):
def setUp(self):
self.soinn = Soinn()
self.soinn.nodes = np.array([[0, 0], [1, 0], [1, 1], [0, 1]], dtype=np.float64)
self.soinn.adjacent_mat = dok_matrix((4, 4))
self.soinn.winning_times = [1] * 4
def test_check_signal(self):
self.soinn = Soinn()
signal = [0, 1, 2]
self.assertRaises(TypeError, self.soinn._Soinn__check_signal, signal)
signal = np.arange(6).reshape(2, 3)
self.assertRaises(TypeError, self.soinn._Soinn__check_signal, signal)
d = 6
signal = np.arange(d)
self.soinn._Soinn__check_signal(signal)
self.assertTrue(hasattr(self.soinn, 'dim'))
self.assertEqual(self.soinn.dim, d)
signal = np.arange(d + 1)
self.assertRaises(TypeError, self.soinn._Soinn__check_signal, signal)
def test_add_node(self):
self.soinn = Soinn()
self.soinn.dim = 2 # dim have to be set before calling __add_node()
signals = np.random.random((4, 2))
for n in range(signals.shape[0]):
signal = signals[n, :]
self.soinn._Soinn__add_node(signal)
self.assertEqual(len(self.soinn.winning_times), n + 1)
self.assertEqual(self.soinn.nodes.shape, (n + 1, self.soinn.dim))
np.testing.assert_array_equal(self.soinn.nodes[n, :], signal)
self.assertEqual(self.soinn.adjacent_mat.shape, (n + 1, n + 1))
self.assertEqual(self.soinn.adjacent_mat.nnz, 0)
def test_add_edge(self):
s1, s2 = 1, 2
self.soinn._Soinn__add_edge((s1, s2))
self.assertEqual(self.soinn.adjacent_mat.nnz, 2)
self.assertEqual(self.soinn.adjacent_mat[s1, s2], 1)
def test_increment_edge_ages(self):
self.soinn.adjacent_mat[0, 1:3] = 1
self.soinn.adjacent_mat[1:3, 0] = 1
self.soinn._Soinn__increment_edge_ages(0)
expected = dok_matrix([[0, 2, 2, 0], [2, 0, 0, 0], [2, 0, 0, 0], [0, 0, 0, 0]])
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(), expected.toarray())
self.soinn._Soinn__increment_edge_ages(1)
expected = dok_matrix([[0, 3, 2, 0], [3, 0, 0, 0], [2, 0, 0, 0], [0, 0, 0, 0]])
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(), expected.toarray())
def test_delete_old_edges(self):
self.soinn.winning_times = [i for i in range(4)]
m = self.soinn.max_edge_age
self.soinn.adjacent_mat[[0, 1], [1, 0]] = m + 2
self.soinn.adjacent_mat[[0, 2], [2, 0]] = m + 1
self.soinn._Soinn__delete_old_edges(0)
actual = self.soinn.adjacent_mat.toarray()
expected = dok_matrix([[0, m+1, 0], [m+1, 0, 0], [0, 0, 0]]).toarray()
np.testing.assert_array_equal(actual, expected)
expected = np.array([[0, 0], [1, 1], [0, 1]], dtype=np.float64)
np.testing.assert_array_equal(self.soinn.nodes, expected)
self.assertEqual(self.soinn.winning_times, [0, 2, 3])
def test_delete_old_edges_with_deleting_no_node(self):
# No node is deleted by the function
self.soinn.winning_times = [i for i in range(4)]
m = self.soinn.max_edge_age
self.soinn.adjacent_mat[[0, 1], [1, 0]] = m + 2
self.soinn.adjacent_mat[[1, 2], [2, 1]] = 1
previous_nodes = self.soinn.nodes
previous_winning_times = self.soinn.winning_times
self.soinn._Soinn__delete_old_edges(0)
actual = self.soinn.adjacent_mat.toarray()
expected = dok_matrix([[0, 0, 0, 0], [0, 0, 1, 0], [0, 1, 0, 0], [0, 0, 0, 0]]).toarray()
np.testing.assert_array_equal(actual, expected)
np.testing.assert_array_equal(self.soinn.nodes, previous_nodes)
self.assertEqual(self.soinn.winning_times, previous_winning_times)
def test_delete_old_edges_with_deleting_several_nodes(self):
# delete several nodes simultaneously
self.soinn.winning_times = [i for i in range(4)]
m = self.soinn.max_edge_age
self.soinn.adjacent_mat[[0, 1, 0, 3], [1, 0, 3, 0]] = m + 2
self.soinn.adjacent_mat[[0, 2], [2, 0]] = m + 1
self.soinn._Soinn__delete_old_edges(0)
actual = self.soinn.adjacent_mat.toarray()
expected = dok_matrix([[0, m+1], [m+1, 0]]).toarray()
np.testing.assert_array_equal(actual, expected)
self.assertEqual(self.soinn.winning_times, [0, 2])
def test_delete_old_edges_with_changing_winner_index(self):
# the winner index is changed by deleting nodes
self.soinn.winning_times = [i for i in range(4)]
expected = self.soinn.nodes[3, :]
m = self.soinn.max_edge_age
self.soinn.adjacent_mat[[3, 0, 3, 2], [0, 3, 2, 3]] = m + 2
self.soinn.adjacent_mat[[3, 1], [1, 3]] = m + 1
winner_index = self.soinn._Soinn__delete_old_edges(3)
np.testing.assert_array_equal(self.soinn.nodes[winner_index, :], expected)
def test_delete_nodes(self):
self.soinn.winning_times = [i for i in range(4)]
self.soinn.adjacent_mat[[0, 1], [1, 0]] = 1
self.soinn.adjacent_mat[[2, 3], [3, 2]] = 2
self.soinn._Soinn__delete_nodes([1])
expected = np.array([[0, 0], [1, 1], [0, 1]], dtype=np.float64)
np.testing.assert_array_equal(self.soinn.nodes, expected)
self.assertEqual(self.soinn.winning_times, [0, 2, 3])
expected = dok_matrix([[0, 0, 0], [0, 0, 2], [0, 2, 0]]).toarray()
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(), expected)
def test_delete_nodes_with_deleting_several_nodes(self):
# delete several nodes simultaneously
self.soinn.winning_times = [i for i in range(4)]
self.soinn.adjacent_mat[[0, 1], [1, 0]] = 1
self.soinn.adjacent_mat[[2, 3], [3, 2]] = 2
self.soinn._Soinn__delete_nodes([1, 3])
expected = np.array([[0, 0], [1, 1]], dtype=np.float64)
np.testing.assert_array_equal(self.soinn.nodes, expected)
self.assertEqual(self.soinn.winning_times, [0, 2])
expected = dok_matrix((2, 2)).toarray()
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(), expected)
def test_delete_noise_nodes(self):
self.soinn.adjacent_mat[0, 2:] = 1
self.soinn.adjacent_mat[2:, 0] = 1
self.soinn.winning_times = [2, 1, 2, 1]
self.soinn._Soinn__delete_noise_nodes()
np.testing.assert_array_equal(self.soinn.nodes, [[0, 0], [1, 1], [0, 1]])
self.assertEqual(self.soinn.winning_times, [2, 2, 1])
expected = [[0, 1, 1], [1, 0, 0], [1, 0, 0]]
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(), expected)
def test_find_nearest_nodes(self):
signal = np.array([-1, 0])
indexes, sq_dists = self.soinn._Soinn__find_nearest_nodes(1, signal)
self.assertEqual(indexes, [0])
self.assertEqual(sq_dists, [1])
indexes, sq_dists = self.soinn._Soinn__find_nearest_nodes(2, signal)
self.assertEqual(indexes, [0, 3])
self.assertEqual(sq_dists, [1, 2])
def test_calculate_similarity_thresholds(self):
self.soinn.adjacent_mat[0, 2:] = 1
self.soinn.adjacent_mat[2:, 0] = 1
self.assertEqual(self.soinn._Soinn__calculate_similarity_thresholds([0]), [2])
self.assertEqual(self.soinn._Soinn__calculate_similarity_thresholds([0, 1]), [2, 1])
def test_update_winner(self):
self.soinn.winning_times[0] = 2 - 1
np.testing.assert_array_equal(self.soinn.nodes[0, :], [0, 0])
self.soinn._Soinn__update_winner(0, np.array([-1, 0], dtype=np.float64))
np.testing.assert_array_equal(self.soinn.nodes[0, :], [-0.5, 0])
self.assertEqual(self.soinn.winning_times[0], 2)
self.soinn.winning_times[1] = 3 - 1
np.testing.assert_array_equal(self.soinn.nodes[1, :], [1, 0])
self.soinn._Soinn__update_winner(1, np.array([2, 0], dtype=np.float64))
np.testing.assert_array_equal(self.soinn.nodes[1, :], [1 + 1/3, 0])
self.assertEqual(self.soinn.winning_times[1], 3)
def test_update_adjacent_nodes(self):
self.soinn.adjacent_mat[0, 1:3] = 1
self.soinn.adjacent_mat[1:3, 0] = 1
np.testing.assert_array_equal(self.soinn.nodes[1], [1, 0])
np.testing.assert_array_equal(self.soinn.nodes[2], [1, 1])
self.soinn._Soinn__update_adjacent_nodes(0, np.array([-1, 0]))
np.testing.assert_array_equal(self.soinn.nodes[1], [1 - 2/100, 0])
np.testing.assert_array_equal(self.soinn.nodes[2], [1 - 2/100, 1 - 1/100])
np.testing.assert_array_equal(self.soinn.nodes[0], [0, 0])
np.testing.assert_array_equal(self.soinn.nodes[3], [0, 1])
def test_input_signal(self):
n = 500
data = np.random.rand(n, 2)
for i in range(n):
self.soinn.input_signal(data[i])
print(self.soinn)
class TestSoinn(unittest.TestCase):
def setUp(self):
self.soinn = Soinn()
self.soinn.nodes = np.array([[0, 0], [1, 0], [1, 1], [0, 1]], dtype=np.float64)
self.soinn.adjacent_mat = dok_matrix((4, 4))
self.soinn.winning_times = [1] * 4
def test_check_signal(self):
self.soinn = Soinn()
signal = [0, 1, 2]
self.assertRaises(TypeError, self.soinn._Soinn__check_signal, signal)
signal = np.arange(6).reshape(2, 3)
self.assertRaises(TypeError, self.soinn._Soinn__check_signal, signal)
d = 6
signal = np.arange(d)
self.soinn._Soinn__check_signal(signal)
self.assertTrue(hasattr(self.soinn, 'dim'))
self.assertEqual(self.soinn.dim, d)
signal = np.arange(d + 1)
self.assertRaises(TypeError, self.soinn._Soinn__check_signal, signal)
def test_add_node(self):
self.soinn = Soinn()
self.soinn.dim = 2 # dim have to be set before calling __add_node()
signals = np.random.random((4, 2))
for n in range(signals.shape[0]):
signal = signals[n, :]
self.soinn._Soinn__add_node(signal)
self.assertEqual(len(self.soinn.winning_times), n + 1)
self.assertEqual(self.soinn.nodes.shape, (n + 1, self.soinn.dim))
np.testing.assert_array_equal(self.soinn.nodes[n, :], signal)
self.assertEqual(self.soinn.adjacent_mat.shape, (n + 1, n + 1))
self.assertEqual(self.soinn.adjacent_mat.nnz, 0)
def test_add_edge(self):
s1, s2 = 1, 2
self.soinn._Soinn__add_edge((s1, s2))
self.assertEqual(self.soinn.adjacent_mat.nnz, 2)
self.assertEqual(self.soinn.adjacent_mat[s1, s2], 1)
def test_increment_edge_ages(self):
self.soinn.adjacent_mat[0, 1:3] = 1
self.soinn.adjacent_mat[1:3, 0] = 1
self.soinn._Soinn__increment_edge_ages(0)
expected = dok_matrix([[0, 2, 2, 0], [2, 0, 0, 0], [2, 0, 0, 0], [0, 0, 0, 0]])
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(), expected.toarray())
self.soinn._Soinn__increment_edge_ages(1)
expected = dok_matrix([[0, 3, 2, 0], [3, 0, 0, 0], [2, 0, 0, 0], [0, 0, 0, 0]])
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(), expected.toarray())
def test_delete_old_edges(self):
self.soinn.winning_times = [i for i in range(4)]
m = self.soinn.max_edge_age
self.soinn.adjacent_mat[[0, 1], [1, 0]] = m + 2
self.soinn.adjacent_mat[[0, 2], [2, 0]] = m + 1
self.soinn._Soinn__delete_old_edges(0)
actual = self.soinn.adjacent_mat.toarray()
expected = dok_matrix([[0, m+1, 0], [m+1, 0, 0], [0, 0, 0]]).toarray()
np.testing.assert_array_equal(actual, expected)
expected = np.array([[0, 0], [1, 1], [0, 1]], dtype=np.float64)
np.testing.assert_array_equal(self.soinn.nodes, expected)
self.assertEqual(self.soinn.winning_times, [0, 2, 3])
def test_delete_old_edges_with_deleting_no_node(self):
# No node is deleted by the function
self.soinn.winning_times = [i for i in range(4)]
m = self.soinn.max_edge_age
self.soinn.adjacent_mat[[0, 1], [1, 0]] = m + 2
self.soinn.adjacent_mat[[1, 2], [2, 1]] = 1
previous_nodes = self.soinn.nodes
previous_winning_times = self.soinn.winning_times
self.soinn._Soinn__delete_old_edges(0)
actual = self.soinn.adjacent_mat.toarray()
expected = dok_matrix([[0, 0, 0, 0], [0, 0, 1, 0], [0, 1, 0, 0], [0, 0, 0, 0]]).toarray()
np.testing.assert_array_equal(actual, expected)
np.testing.assert_array_equal(self.soinn.nodes, previous_nodes)
self.assertEqual(self.soinn.winning_times, previous_winning_times)
def test_delete_old_edges_with_deleting_several_nodes(self):
# delete several nodes simultaneously
self.soinn.winning_times = [i for i in range(4)]
m = self.soinn.max_edge_age
self.soinn.adjacent_mat[[0, 1, 0, 3], [1, 0, 3, 0]] = m + 2
self.soinn.adjacent_mat[[0, 2], [2, 0]] = m + 1
self.soinn._Soinn__delete_old_edges(0)
actual = self.soinn.adjacent_mat.toarray()
expected = dok_matrix([[0, m+1], [m+1, 0]]).toarray()
np.testing.assert_array_equal(actual, expected)
self.assertEqual(self.soinn.winning_times, [0, 2])
def test_delete_old_edges_with_changing_winner_index(self):
# the winner index is changed by deleting nodes
self.soinn.winning_times = [i for i in range(4)]
expected = self.soinn.nodes[3, :]
m = self.soinn.max_edge_age
self.soinn.adjacent_mat[[3, 0, 3, 2], [0, 3, 2, 3]] = m + 2
self.soinn.adjacent_mat[[3, 1], [1, 3]] = m + 1
winner_index = self.soinn._Soinn__delete_old_edges(3)
np.testing.assert_array_equal(self.soinn.nodes[winner_index, :], expected)
def test_delete_nodes(self):
self.soinn.winning_times = [i for i in range(4)]
self.soinn.adjacent_mat[[0, 1], [1, 0]] = 1
self.soinn.adjacent_mat[[2, 3], [3, 2]] = 2
self.soinn._Soinn__delete_nodes([1])
expected = np.array([[0, 0], [1, 1], [0, 1]], dtype=np.float64)
np.testing.assert_array_equal(self.soinn.nodes, expected)
self.assertEqual(self.soinn.winning_times, [0, 2, 3])
expected = dok_matrix([[0, 0, 0], [0, 0, 2], [0, 2, 0]]).toarray()
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(), expected)
def test_delete_nodes_with_deleting_several_nodes(self):
# delete several nodes simultaneously
self.soinn.winning_times = [i for i in range(4)]
self.soinn.adjacent_mat[[0, 1], [1, 0]] = 1
self.soinn.adjacent_mat[[2, 3], [3, 2]] = 2
self.soinn._Soinn__delete_nodes([1, 3])
expected = np.array([[0, 0], [1, 1]], dtype=np.float64)
np.testing.assert_array_equal(self.soinn.nodes, expected)
self.assertEqual(self.soinn.winning_times, [0, 2])
expected = dok_matrix((2, 2)).toarray()
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(), expected)
def test_delete_noise_nodes(self):
self.soinn.adjacent_mat[0, 2] = 1
self.soinn.adjacent_mat[0, 3] = 2
self.soinn.adjacent_mat[2, 0] = 1
self.soinn.adjacent_mat[3, 0] = 2
self.soinn.winning_times = [2, 1, 2, 1]
self.soinn._Soinn__delete_noise_nodes()
np.testing.assert_array_equal(self.soinn.nodes, [[0, 0], [1, 1], [0, 1]])
self.assertEqual(self.soinn.winning_times, [2, 2, 1])
expected = [[0, 1, 2], [1, 0, 0], [2, 0, 0]]
np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(), expected)
## def test_delete_noise_nodes2(self):
## self.soinn.adjacent_mat[0, 1] = 1
## self.soinn.adjacent_mat[0, 3] = 2
## self.soinn.adjacent_mat[1, 0] = 1
## self.soinn.adjacent_mat[3, 0] = 2
## self.soinn.winning_times = [3, 1, 2, 1]
## self.soinn._Soinn__delete_noise_nodes()
## np.testing.assert_array_equal(self.soinn.nodes, [[0, 0], [1, 1], [0, 1]])
## self.assertEqual(self.soinn.winning_times, [2, 2, 1])
## expected = [[0, 1, 1], [1, 0, 0], [1, 0, 0]]
## np.testing.assert_array_equal(self.soinn.adjacent_mat.toarray(), expected)
def test_find_nearest_nodes(self):
signal = np.array([-1, 0])
indexes, sq_dists = self.soinn._Soinn__find_nearest_nodes(1, signal)
self.assertEqual(indexes, [0])
self.assertEqual(sq_dists, [1])
indexes, sq_dists = self.soinn._Soinn__find_nearest_nodes(2, signal)
self.assertEqual(indexes, [0, 3])
self.assertEqual(sq_dists, [1, 2])
def test_calculate_similarity_thresholds(self):
self.soinn.adjacent_mat[0, 2:] = 1
self.soinn.adjacent_mat[2:, 0] = 1
self.assertEqual(self.soinn.calculate_similarity_thresholds([0]), [2])
self.assertEqual(self.soinn.calculate_similarity_thresholds([0, 1]), [2, 1])
def test_update_winner(self):
self.soinn.winning_times[0] = 2 - 1
np.testing.assert_array_equal(self.soinn.nodes[0, :], [0, 0])
self.soinn._Soinn__update_winner(0, np.array([-1, 0], dtype=np.float64))
np.testing.assert_array_equal(self.soinn.nodes[0, :], [-0.5, 0])
self.assertEqual(self.soinn.winning_times[0], 2)
self.soinn.winning_times[1] = 3 - 1
np.testing.assert_array_equal(self.soinn.nodes[1, :], [1, 0])
self.soinn._Soinn__update_winner(1, np.array([2, 0], dtype=np.float64))
np.testing.assert_array_equal(self.soinn.nodes[1, :], [1 + 1/3, 0])
self.assertEqual(self.soinn.winning_times[1], 3)
def test_update_adjacent_nodes(self):
self.soinn.adjacent_mat[0, 1:3] = 1
self.soinn.adjacent_mat[1:3, 0] = 1
np.testing.assert_array_equal(self.soinn.nodes[1], [1, 0])
np.testing.assert_array_equal(self.soinn.nodes[2], [1, 1])
self.soinn._Soinn__update_adjacent_nodes(0, np.array([-1, 0]))
np.testing.assert_array_equal(self.soinn.nodes[1], [1 - 2/100, 0])
np.testing.assert_array_equal(self.soinn.nodes[2], [1 - 2/100, 1 - 1/100])
np.testing.assert_array_equal(self.soinn.nodes[0], [0, 0])
np.testing.assert_array_equal(self.soinn.nodes[3], [0, 1])
def test_input_signal(self):
n = 500
data = np.random.rand(n, 2)
for i in range(n):
self.soinn.input_signal(data[i])
print(self.soinn)