def testBuild(self):
kd = KDTree(dimensions = 3)
points = array([[5, 4, 3], [10, 2, 1], [4, 5, 6], [6, 1, 5], [1, 2, 3]])
labels = array([[1], [1], [1], [-1], [-1]])
kd.build(points, labels)
print(kd.searchKNN(([4, 4, 6]), 3)[0][0]._data)
pass
def maisProximo():
maxN = 1000
startN = 1
increment = 100
N = 40
dados = []
labels = []
f1 = open('labels.dat', 'w')
f2 = open('dados.dat', 'w')
for k in range(startN, maxN, increment):
b = 0.0
for j in range(N):
lista = []
inserirDados(lista, k)
arv = KDTree(k, 2)
construirArvore(arv, lista)
t1 = time.time()
arv.nearestNeighbour(lista[randint(0, len(lista))])
t2 = time.time()
b += t2-t1
f1.write(str(k))
f2.write(str(b/N))
print k, (b/N)
dados.append(b/N)
labels.append(k)
f1.close()
f2.close()
drawGraph(dados, labels, "titulo")
def test_point_types(self):
emptyTree = KDTree.create(dimensions=3)
point1 = (2, 3, 4)
point2 = [4, 5, 6]
Point = collections.namedtuple('Point', 'x y z')
point3 = Point(5, 3, 2)
tree = KDTree.create([point1, point2, point3])
res, dist = tree.search_nn((1, 2, 3))
self.assertEqual(res, KDTree.KDNode((2, 3, 4)))
def test_invalid_child(self):
""" Children on wrong subtree invalidate Tree """
child = KDTree.KDNode((3, 2))
child.axis = 2
tree = KDTree.create([(2, 3)])
tree.left = child
self.assertFalse(tree.is_valid())
tree = KDTree.create([(4, 1)])
tree.right = child
self.assertFalse(tree.is_valid())
def contact_residues(A, B, cutoff=4.0, pointing=45.0):
tree = KDTree([b.co for b in B])
atoms = []
for a in A:
#if a.anam in [" C "," N "," O "," CA "]: continue
d, co = tree.nearest(a.co)
if d < cutoff * cutoff:
b = tree.index_of(co)
ang = side_chain_angle_to_contact(a, A, B[b])
if ang != -1 and ang < pointing: atoms.append(a)
residues = [x.rnum for x in atoms]
residues = list(set(residues)) # remove duplicates
residues.sort()
return residues
def get_knn_e_dist_with_kdtree(train_samples, new_inst, k):
kdtree = get_kd_tree(train_samples)
knn_heaps = KDTree.find_knn_from_kd_tree(new_inst, kdtree, k)
idx = np.array([item[2] for item in knn_heaps.knns[1:]])
dist_sq = np.array([item[0] for item in knn_heaps.knns[1:]])
# print(idx)
# print(dist_sq)
return idx, np.sqrt(dist_sq)
def get_kd_tree(train_samples):
global kd_tree
if kd_tree is None:
# print(train_samples.dtype)
m, _ = train_samples.shape
ls = np.array(list(range(m)))
kd_tree = KDTree.construct_kd_tree(train_samples, ls)
return kd_tree
def testBalanced(self):
t = array([[0, 1, 2]])
t = t.T
samples = 5000
x1 = rand(samples, 2) + array([3, 3])
y1 = ones((samples, 1), int) * t[0][0]
x2 = rand(samples, 2) + array([6, 6])
y2 = ones((samples, 1), int) * t[1][0]
x3 = rand(samples, 2) + array([7, 2])
y3 = ones((samples, 1), int) * t[2][0]
x = append(append(x1, x2, axis = 0), x3, axis = 0)
y = append(append(y1, y2, axis = 0), y3, axis = 0)
testPoint = array([3.2, 3])
kd = KDTree(dimensions = 2)
kd.build(x, y)
print(kd.isBlanced())
def build(self, points, labels):
'''
suppose n is samples' amount, m is features' amount.
points: array, training data, n * m
labels: array, n * 1
'''
dimensions = points.shape[1]
self._kdTree = KDTree(dimensions=dimensions)
self._kdTree.build(points, labels)
def test_search_nn(self, nodes=100):
points = list(islice(random_points(), 0, nodes))
tree = KDTree.create(points)
point = random_point()
nn, dist = tree.search_nn(point)
best, best_dist = self.find_best(tree, point)
self.assertEqual(best_dist,
dist,
msg=', '.join(repr(p)
for p in points) + ' / ' + repr(point))
def test_remove_duplicates(self):
""" creates a tree with only duplicate points, and removes them all """
points = [(1, 1)] * 100
tree = KDTree.create(points)
self.assertTrue(tree.is_valid())
random.shuffle(points)
while points:
point = points.pop(0)
tree = tree.remove(point)
# Check if the Tree is valid after the removal
self.assertTrue(tree.is_valid())
# Check if the removal reduced the number of nodes by 1 (not more, not less)
remaining_points = len(points)
nodes_in_tree = len(list(tree.inorder()))
self.assertEqual(nodes_in_tree, remaining_points)
def do_random_remove(self):
""" Creates a random tree, removes all points in random order """
points = list(set(islice(random_points(), 0, 20)))
tree = KDTree.create(points)
self.assertTrue(tree.is_valid())
random.shuffle(points)
while points:
point = points.pop(0)
tree = tree.remove(point)
# Check if the Tree is valid after the removal
self.assertTrue(tree.is_valid())
# Check if the point has actually been removed
self.assertTrue(point not in [n.data for n in tree.inorder()])
# Check if the removal reduced the number of nodes by 1 (not more, not less)
remaining_points = len(points)
nodes_in_tree = len(list(tree.inorder()))
self.assertEqual(nodes_in_tree, remaining_points)
def test_search_knn(self):
points = [(50, 20), (51, 19), (1, 80)]
tree = KDTree.create(points)
point = (48, 18)
all_dist = []
for p in tree.inorder():
dist = p.dist(point)
all_dist.append([p, dist])
all_dist = sorted(all_dist, key=lambda n: n[1])
result = tree.search_knn(point, 1)
self.assertEqual(result[0][1], all_dist[0][1])
result = tree.search_knn(point, 2)
self.assertEqual(result[0][1], all_dist[0][1])
self.assertEqual(result[1][1], all_dist[1][1])
result = tree.search_knn(point, 3)
self.assertEqual(result[0][1], all_dist[0][1])
self.assertEqual(result[1][1], all_dist[1][1])
self.assertEqual(result[2][1], all_dist[2][1])
def test_remove_empty_tree(self):
tree = KDTree.create(dimensions=2)
tree.remove((1, 2))
self.assertFalse(bool(tree))
def random_tree(nodes=20, dimensions=3, minval=0, maxval=100):
points = list(islice(random_points(), 0, nodes))
tree = KDTree.create(points)
return tree
def testAdd(self):
kd = KDTree(dimensions = 3)
points = array([[5, 4, 3], [10, 2, 1], [4, 5, 6], [6, 1, 5], [1, 2, 3]])
labels = array([[1], [1], [1], [-1], [-1]])
kd.add([5, 4, 3], 1)
class KNN(object):
def __init__(self):
self._kdTree = None
def __init__(self, points, labels):
'''
suppose n is samples' amount, m is features' amount.
points: array, training data, n * m
labels: array, n * 1
'''
self.build(points, labels)
def build(self, points, labels):
'''
suppose n is samples' amount, m is features' amount.
points: array, training data, n * m
labels: array, n * 1
'''
dimensions = points.shape[1]
self._kdTree = KDTree(dimensions=dimensions)
self._kdTree.build(points, labels)
def add(self, x, y):
'''
add a point
x: single array
y: int
'''
pass
def add(self, x, y):
'''
suppose n is samples' amount, m is features' amount.
x: array, training data, n * m
y: array, n * 1
'''
pass
def getKNN(self, point, k=5):
'''
get nearest n points
'''
return self._kdTree.searchKNN(point, k)
def getLabel(self, point, k=5):
'''
get label of point
'''
nns = self._kdTree.searchKNN(point, k)
votes = {}
for item in nns:
label = item[0]._label[0]
if label not in votes:
votes[label] = 1
else:
votes[label] += 1
return sorted(votes.items(), key=lambda x: x[1], reverse=True)[0][0]
#!/usr/bin/env python # coding=utf-8 T = [[2, 3], [5, 4], [9, 6], [4, 7], [8, 1], [7, 2]] T1 = [[0., 0., 0.], [0., .5, 0.], [1., 1., .5]] import KDTree kd_tree = KDTree.build_kdtree(T, 0, 2) print '-' * 15, '[9,4]', '-' * 15 print KDTree.search_kdtree(kd_tree, 0, [9, 4], 2) kd_tree = KDTree.build_kdtree(T1, 0, 3) print '-' * 15, '[1.,1.,1.]', '-' * 15 print KDTree.search_kdtree(kd_tree, 0, [1., 1., 1.], 3) #对照sklearn中实现的最近邻算法 from sklearn.neighbors import NearestNeighbors neigh = NearestNeighbors(n_neighbors=1) neigh.fit(T) print '-' * 15, '[9,4]', '-' * 15 print T print neigh.kneighbors([[9, 4]]) print '-' * 15, '[1.,1.,1.]', '-' * 15 print T1 neigh.fit(T1) print neigh.kneighbors([[1., 1., 1.]])
