class DistancesTest(unittest.TestCase):
"""
Tests the management of distances (in the dists matrix).
"""
def setUp(self):
self.vecs = [[10], [20], [0], [20]]
self.initial_vecs = self.vecs[:2]
self.h = Hierarchy(metric='euclidean',
lower_limit_scale=0.1,
upper_limit_scale=1.5)
self.h.fit(self.initial_vecs)
children = [self.h.create_node(vec=vec) for vec in self.vecs[2:]]
n = self.h.create_node(children=children)
self.h.g.add_child(2, n)
self.leaves = [0,1,3,4]
self.clusters = [2,5]
def test_distance(self):
node_k = self.h.create_node(vec=[20])
# Distances should be symmetric.
for n in self.leaves:
d = self.h.get_distance(n, node_k)
d_ = self.h.get_distance(node_k, n)
self.assertEqual(d, d_)
def test_update_distances(self):
# Create some extra nodes.
data = np.array([[1],[2],[4],[8],[12]])
nodes = [self.h.create_node(vec=center) for center in data]
# Calculate a distance matrix independently to compare to.
# We include the vector which initialized the hierarchy
# and the center of the initial cluster node.
old_data = self.initial_vecs + [self.h.centers[self.clusters[0]]] + self.vecs[2:] + [self.h.centers[self.clusters[1]]]
data = np.insert(data, 0, old_data, axis=0)
dist_mat = pairwise_distances(data, metric='euclidean')
self.assertTrue((dist_mat == self.h.dists).all())
def test_cdm(self):
# Expecting the matrix to have rows and columns 0,1,n (n=5)
# since those are the child nodes.
expected = [[ 0., 10., 0.],
[10., 0., 10.],
[ 0., 10., 0.]]
assert_array_equal(expected, self.h.cdm(2))
def test_get_closest_leaf(self):
node_k = self.h.create_node(vec=[11])
result, dist = self.h.get_closest_leaf(node_k)
self.assertEqual(result, self.leaves[0])
self.assertEqual(dist, 1)
def test_get_nearest_distances(self):
d = self.h.get_nearest_distances(2)
expected = [ 0., 10., 0.]
assert_array_equal(expected, d)
def test_get_nearest_child(self):
"""
2
+-+--+
0 1 5
+-+
3 4
"""
i, d = self.h.get_nearest_child(5, 1)
self.assertEqual(i, 4)
self.assertEqual(d, 0)
def test_get_nearest_children(self):
i, j, d = self.h.get_nearest_children(2)
self.assertEqual(i, 0)
self.assertEqual(j, 5)
self.assertEqual(d, 0)
def test_get_furthest_nearest_children(self):
i, j, d = self.h.get_furthest_nearest_children(2)
self.assertEqual(i, 0)
self.assertEqual(j, 1)
self.assertEqual(d, 10)
def test_get_representative(self):
r = self.h.get_representative(2)
self.assertEqual(r, 0)
def test_most_representative(self):
# Incorporating these vectors puts the center of all nodes around ~22
new_vecs = [[30], [40], [40]]
self.h.fit(new_vecs)
nodes = self.h.nodes
rep = self.h.most_representative(nodes)
# Expecting that the representative node is 1, w/ a center of [20]
self.assertEqual(rep, 1)
class DistancesTest(unittest.TestCase):
"""
Tests the management of distances (in the dists matrix).
"""
def setUp(self):
self.vecs = [[10], [20], [0], [20]]
self.initial_vecs = self.vecs[:2]
self.h = Hierarchy(metric='euclidean',
lower_limit_scale=0.1,
upper_limit_scale=1.5)
self.h.fit(self.initial_vecs)
children = [self.h.create_node(vec=vec) for vec in self.vecs[2:]]
n = self.h.create_node(children=children)
self.h.g.add_child(2, n)
self.leaves = [0, 1, 3, 4]
self.clusters = [2, 5]
def test_distance(self):
node_k = self.h.create_node(vec=[20])
# Distances should be symmetric.
for n in self.leaves:
d = self.h.get_distance(n, node_k)
d_ = self.h.get_distance(node_k, n)
self.assertEqual(d, d_)
def test_update_distances(self):
# Create some extra nodes.
data = np.array([[1], [2], [4], [8], [12]])
nodes = [self.h.create_node(vec=center) for center in data]
# Calculate a distance matrix independently to compare to.
# We include the vector which initialized the hierarchy
# and the center of the initial cluster node.
old_data = self.initial_vecs + [
self.h.centers[self.clusters[0]]
] + self.vecs[2:] + [self.h.centers[self.clusters[1]]]
data = np.insert(data, 0, old_data, axis=0)
dist_mat = pairwise_distances(data, metric='euclidean')
self.assertTrue((dist_mat == self.h.dists).all())
def test_cdm(self):
# Expecting the matrix to have rows and columns 0,1,n (n=5)
# since those are the child nodes.
expected = [[0., 10., 0.], [10., 0., 10.], [0., 10., 0.]]
assert_array_equal(expected, self.h.cdm(2))
def test_get_closest_leaf(self):
node_k = self.h.create_node(vec=[11])
result, dist = self.h.get_closest_leaf(node_k)
self.assertEqual(result, self.leaves[0])
self.assertEqual(dist, 1)
def test_get_nearest_distances(self):
d = self.h.get_nearest_distances(2)
expected = [0., 10., 0.]
assert_array_equal(expected, d)
def test_get_nearest_child(self):
"""
2
+-+--+
0 1 5
+-+
3 4
"""
i, d = self.h.get_nearest_child(5, 1)
self.assertEqual(i, 4)
self.assertEqual(d, 0)
def test_get_nearest_children(self):
i, j, d = self.h.get_nearest_children(2)
self.assertEqual(i, 0)
self.assertEqual(j, 5)
self.assertEqual(d, 0)
def test_get_furthest_nearest_children(self):
i, j, d = self.h.get_furthest_nearest_children(2)
self.assertEqual(i, 0)
self.assertEqual(j, 1)
self.assertEqual(d, 10)
def test_get_representative(self):
r = self.h.get_representative(2)
self.assertEqual(r, 0)
def test_most_representative(self):
# Incorporating these vectors puts the center of all nodes around ~22
new_vecs = [[30], [40], [40]]
self.h.fit(new_vecs)
nodes = self.h.nodes
rep = self.h.most_representative(nodes)
# Expecting that the representative node is 1, w/ a center of [20]
self.assertEqual(rep, 1)
