1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
|
import pickle
import itertools
import numpy as np
import pytest
from numpy.testing import assert_array_almost_equal
from sklearn.neighbors.ball_tree import (BallTree, NeighborsHeap,
simultaneous_sort, kernel_norm,
nodeheap_sort, DTYPE, ITYPE)
from sklearn.neighbors.dist_metrics import DistanceMetric
from sklearn.utils import check_random_state
from sklearn.utils.testing import assert_allclose
rng = np.random.RandomState(10)
V_mahalanobis = rng.rand(3, 3)
V_mahalanobis = np.dot(V_mahalanobis, V_mahalanobis.T)
DIMENSION = 3
METRICS = {'euclidean': {},
'manhattan': {},
'minkowski': dict(p=3),
'chebyshev': {},
'seuclidean': dict(V=rng.random_sample(DIMENSION)),
'wminkowski': dict(p=3, w=rng.random_sample(DIMENSION)),
'mahalanobis': dict(V=V_mahalanobis)}
DISCRETE_METRICS = ['hamming',
'canberra',
'braycurtis']
BOOLEAN_METRICS = ['matching', 'jaccard', 'dice', 'kulsinski',
'rogerstanimoto', 'russellrao', 'sokalmichener',
'sokalsneath']
def dist_func(x1, x2, p):
return np.sum((x1 - x2) ** p) ** (1. / p)
def brute_force_neighbors(X, Y, k, metric, **kwargs):
D = DistanceMetric.get_metric(metric, **kwargs).pairwise(Y, X)
ind = np.argsort(D, axis=1)[:, :k]
dist = D[np.arange(Y.shape[0])[:, None], ind]
return dist, ind
@pytest.mark.parametrize('metric', METRICS)
@pytest.mark.parametrize('k', (1, 3, 5))
@pytest.mark.parametrize('dualtree', (True, False))
@pytest.mark.parametrize('breadth_first', (True, False))
def test_ball_tree_query(metric, k, dualtree, breadth_first):
rng = check_random_state(0)
X = rng.random_sample((40, DIMENSION))
Y = rng.random_sample((10, DIMENSION))
kwargs = METRICS[metric]
bt = BallTree(X, leaf_size=1, metric=metric, **kwargs)
dist1, ind1 = bt.query(Y, k, dualtree=dualtree,
breadth_first=breadth_first)
dist2, ind2 = brute_force_neighbors(X, Y, k, metric, **kwargs)
# don't check indices here: if there are any duplicate distances,
# the indices may not match. Distances should not have this problem.
assert_array_almost_equal(dist1, dist2)
@pytest.mark.parametrize('metric',
itertools.chain(BOOLEAN_METRICS, DISCRETE_METRICS))
def test_ball_tree_query_metrics(metric):
rng = check_random_state(0)
if metric in BOOLEAN_METRICS:
X = rng.random_sample((40, 10)).round(0)
Y = rng.random_sample((10, 10)).round(0)
elif metric in DISCRETE_METRICS:
X = (4 * rng.random_sample((40, 10))).round(0)
Y = (4 * rng.random_sample((10, 10))).round(0)
k = 5
bt = BallTree(X, leaf_size=1, metric=metric)
dist1, ind1 = bt.query(Y, k)
dist2, ind2 = brute_force_neighbors(X, Y, k, metric)
assert_array_almost_equal(dist1, dist2)
def test_ball_tree_query_radius(n_samples=100, n_features=10):
rng = check_random_state(0)
X = 2 * rng.random_sample(size=(n_samples, n_features)) - 1
query_pt = np.zeros(n_features, dtype=float)
eps = 1E-15 # roundoff error can cause test to fail
bt = BallTree(X, leaf_size=5)
rad = np.sqrt(((X - query_pt) ** 2).sum(1))
for r in np.linspace(rad[0], rad[-1], 100):
ind = bt.query_radius([query_pt], r + eps)[0]
i = np.where(rad <= r + eps)[0]
ind.sort()
i.sort()
assert_array_almost_equal(i, ind)
def test_ball_tree_query_radius_distance(n_samples=100, n_features=10):
rng = check_random_state(0)
X = 2 * rng.random_sample(size=(n_samples, n_features)) - 1
query_pt = np.zeros(n_features, dtype=float)
eps = 1E-15 # roundoff error can cause test to fail
bt = BallTree(X, leaf_size=5)
rad = np.sqrt(((X - query_pt) ** 2).sum(1))
for r in np.linspace(rad[0], rad[-1], 100):
ind, dist = bt.query_radius([query_pt], r + eps, return_distance=True)
ind = ind[0]
dist = dist[0]
d = np.sqrt(((query_pt - X[ind]) ** 2).sum(1))
assert_array_almost_equal(d, dist)
def compute_kernel_slow(Y, X, kernel, h):
d = np.sqrt(((Y[:, None, :] - X) ** 2).sum(-1))
norm = kernel_norm(h, X.shape[1], kernel)
if kernel == 'gaussian':
return norm * np.exp(-0.5 * (d * d) / (h * h)).sum(-1)
elif kernel == 'tophat':
return norm * (d < h).sum(-1)
elif kernel == 'epanechnikov':
return norm * ((1.0 - (d * d) / (h * h)) * (d < h)).sum(-1)
elif kernel == 'exponential':
return norm * (np.exp(-d / h)).sum(-1)
elif kernel == 'linear':
return norm * ((1 - d / h) * (d < h)).sum(-1)
elif kernel == 'cosine':
return norm * (np.cos(0.5 * np.pi * d / h) * (d < h)).sum(-1)
else:
raise ValueError('kernel not recognized')
@pytest.mark.parametrize("kernel", ['gaussian', 'tophat', 'epanechnikov',
'exponential', 'linear', 'cosine'])
@pytest.mark.parametrize("h", [0.01, 0.1, 1])
@pytest.mark.parametrize("rtol", [0, 1E-5])
@pytest.mark.parametrize("atol", [1E-6, 1E-2])
@pytest.mark.parametrize("breadth_first", [True, False])
def test_ball_tree_kde(kernel, h, rtol, atol, breadth_first, n_samples=100,
n_features=3):
np.random.seed(0)
X = np.random.random((n_samples, n_features))
Y = np.random.random((n_samples, n_features))
bt = BallTree(X, leaf_size=10)
dens_true = compute_kernel_slow(Y, X, kernel, h)
dens = bt.kernel_density(Y, h, atol=atol, rtol=rtol,
kernel=kernel,
breadth_first=breadth_first)
assert_allclose(dens, dens_true,
atol=atol, rtol=max(rtol, 1e-7))
def test_gaussian_kde(n_samples=1000):
# Compare gaussian KDE results to scipy.stats.gaussian_kde
from scipy.stats import gaussian_kde
rng = check_random_state(0)
x_in = rng.normal(0, 1, n_samples)
x_out = np.linspace(-5, 5, 30)
for h in [0.01, 0.1, 1]:
bt = BallTree(x_in[:, None])
gkde = gaussian_kde(x_in, bw_method=h / np.std(x_in))
dens_bt = bt.kernel_density(x_out[:, None], h) / n_samples
dens_gkde = gkde.evaluate(x_out)
assert_array_almost_equal(dens_bt, dens_gkde, decimal=3)
def test_ball_tree_two_point(n_samples=100, n_features=3):
rng = check_random_state(0)
X = rng.random_sample((n_samples, n_features))
Y = rng.random_sample((n_samples, n_features))
r = np.linspace(0, 1, 10)
bt = BallTree(X, leaf_size=10)
D = DistanceMetric.get_metric("euclidean").pairwise(Y, X)
counts_true = [(D <= ri).sum() for ri in r]
def check_two_point(r, dualtree):
counts = bt.two_point_correlation(Y, r=r, dualtree=dualtree)
assert_array_almost_equal(counts, counts_true)
for dualtree in (True, False):
check_two_point(r, dualtree)
def test_ball_tree_pickle():
rng = check_random_state(0)
X = rng.random_sample((10, 3))
bt1 = BallTree(X, leaf_size=1)
# Test if BallTree with callable metric is picklable
bt1_pyfunc = BallTree(X, metric=dist_func, leaf_size=1, p=2)
ind1, dist1 = bt1.query(X)
ind1_pyfunc, dist1_pyfunc = bt1_pyfunc.query(X)
def check_pickle_protocol(protocol):
s = pickle.dumps(bt1, protocol=protocol)
bt2 = pickle.loads(s)
s_pyfunc = pickle.dumps(bt1_pyfunc, protocol=protocol)
bt2_pyfunc = pickle.loads(s_pyfunc)
ind2, dist2 = bt2.query(X)
ind2_pyfunc, dist2_pyfunc = bt2_pyfunc.query(X)
assert_array_almost_equal(ind1, ind2)
assert_array_almost_equal(dist1, dist2)
assert_array_almost_equal(ind1_pyfunc, ind2_pyfunc)
assert_array_almost_equal(dist1_pyfunc, dist2_pyfunc)
assert isinstance(bt2, BallTree)
for protocol in (0, 1, 2):
check_pickle_protocol(protocol)
def test_neighbors_heap(n_pts=5, n_nbrs=10):
heap = NeighborsHeap(n_pts, n_nbrs)
for row in range(n_pts):
d_in = rng.random_sample(2 * n_nbrs).astype(DTYPE)
i_in = np.arange(2 * n_nbrs, dtype=ITYPE)
for d, i in zip(d_in, i_in):
heap.push(row, d, i)
ind = np.argsort(d_in)
d_in = d_in[ind]
i_in = i_in[ind]
d_heap, i_heap = heap.get_arrays(sort=True)
assert_array_almost_equal(d_in[:n_nbrs], d_heap[row])
assert_array_almost_equal(i_in[:n_nbrs], i_heap[row])
def test_node_heap(n_nodes=50):
vals = rng.random_sample(n_nodes).astype(DTYPE)
i1 = np.argsort(vals)
vals2, i2 = nodeheap_sort(vals)
assert_array_almost_equal(i1, i2)
assert_array_almost_equal(vals[i1], vals2)
def test_simultaneous_sort(n_rows=10, n_pts=201):
dist = rng.random_sample((n_rows, n_pts)).astype(DTYPE)
ind = (np.arange(n_pts) + np.zeros((n_rows, 1))).astype(ITYPE)
dist2 = dist.copy()
ind2 = ind.copy()
# simultaneous sort rows using function
simultaneous_sort(dist, ind)
# simultaneous sort rows using numpy
i = np.argsort(dist2, axis=1)
row_ind = np.arange(n_rows)[:, None]
dist2 = dist2[row_ind, i]
ind2 = ind2[row_ind, i]
assert_array_almost_equal(dist, dist2)
assert_array_almost_equal(ind, ind2)
def test_query_haversine():
rng = check_random_state(0)
X = 2 * np.pi * rng.random_sample((40, 2))
bt = BallTree(X, leaf_size=1, metric='haversine')
dist1, ind1 = bt.query(X, k=5)
dist2, ind2 = brute_force_neighbors(X, X, k=5, metric='haversine')
assert_array_almost_equal(dist1, dist2)
assert_array_almost_equal(ind1, ind2)
|
