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
|
import numpy as np
import pytest
from numpy.testing import assert_array_almost_equal
from sklearn.neighbors import KDTree
from sklearn.preprocessing import normalize
from umap import distances as dist
from umap.umap_ import (
nearest_neighbors,
smooth_knn_dist,
)
# ===================================================
# Nearest Neighbour Test cases
# ===================================================
# nearest_neighbours metric parameter validation
# -----------------------------------------------
def test_nn_bad_metric(nn_data):
with pytest.raises(ValueError):
nearest_neighbors(nn_data, 10, 42, {}, False, np.random)
def test_nn_bad_metric_sparse_data(sparse_nn_data):
with pytest.raises(ValueError):
nearest_neighbors(
sparse_nn_data,
10,
"seuclidean",
{},
False,
np.random,
)
# -------------------------------------------------
# Utility functions for Nearest Neighbour
# -------------------------------------------------
def knn(indices, nn_data): # pragma: no cover
tree = KDTree(nn_data)
true_indices = tree.query(nn_data, 10, return_distance=False)
num_correct = 0.0
for i in range(nn_data.shape[0]):
num_correct += np.sum(np.in1d(true_indices[i], indices[i]))
return num_correct / (nn_data.shape[0] * 10)
def smooth_knn(nn_data, local_connectivity=1.0):
knn_indices, knn_dists, _ = nearest_neighbors(
nn_data, 10, "euclidean", {}, False, np.random
)
sigmas, rhos = smooth_knn_dist(
knn_dists, 10.0, local_connectivity=local_connectivity
)
shifted_dists = knn_dists - rhos[:, np.newaxis]
shifted_dists[shifted_dists < 0.0] = 0.0
vals = np.exp(-(shifted_dists / sigmas[:, np.newaxis]))
norms = np.sum(vals, axis=1)
return norms
@pytest.mark.skip()
def test_nn_descent_neighbor_accuracy(nn_data): # pragma: no cover
knn_indices, knn_dists, _ = nearest_neighbors(
nn_data, 10, "euclidean", {}, False, np.random
)
percent_correct = knn(knn_indices, nn_data)
assert (
percent_correct >= 0.85
), "NN-descent did not get 89% accuracy on nearest neighbors"
@pytest.mark.skip()
def test_nn_descent_neighbor_accuracy_low_memory(nn_data): # pragma: no cover
knn_indices, knn_dists, _ = nearest_neighbors(
nn_data, 10, "euclidean", {}, False, np.random, low_memory=True
)
percent_correct = knn(knn_indices, nn_data)
assert (
percent_correct >= 0.89
), "NN-descent did not get 89% accuracy on nearest neighbors"
@pytest.mark.skip()
def test_angular_nn_descent_neighbor_accuracy(nn_data): # pragma: no cover
knn_indices, knn_dists, _ = nearest_neighbors(
nn_data, 10, "cosine", {}, True, np.random
)
angular_data = normalize(nn_data, norm="l2")
percent_correct = knn(knn_indices, angular_data)
assert (
percent_correct >= 0.85
), "NN-descent did not get 89% accuracy on nearest neighbors"
@pytest.mark.skip()
def test_sparse_nn_descent_neighbor_accuracy(sparse_nn_data): # pragma: no cover
knn_indices, knn_dists, _ = nearest_neighbors(
sparse_nn_data, 20, "euclidean", {}, False, np.random
)
percent_correct = knn(knn_indices, sparse_nn_data.todense())
assert (
percent_correct >= 0.75
), "Sparse NN-descent did not get 90% accuracy on nearest neighbors"
@pytest.mark.skip()
def test_sparse_nn_descent_neighbor_accuracy_low_memory(
sparse_nn_data,
): # pragma: no cover
knn_indices, knn_dists, _ = nearest_neighbors(
sparse_nn_data, 20, "euclidean", {}, False, np.random, low_memory=True
)
percent_correct = knn(knn_indices, sparse_nn_data.todense())
assert (
percent_correct >= 0.85
), "Sparse NN-descent did not get 90% accuracy on nearest neighbors"
@pytest.mark.skip()
def test_nn_descent_neighbor_accuracy_callable_metric(nn_data): # pragma: no cover
knn_indices, knn_dists, _ = nearest_neighbors(
nn_data, 10, dist.euclidean, {}, False, np.random
)
percent_correct = knn(knn_indices, nn_data)
assert (
percent_correct >= 0.95
), "NN-descent did not get 95% accuracy on nearest neighbors with callable metric"
@pytest.mark.skip()
def test_sparse_angular_nn_descent_neighbor_accuracy(
sparse_nn_data,
): # pragma: no cover
knn_indices, knn_dists, _ = nearest_neighbors(
sparse_nn_data, 20, "cosine", {}, True, np.random
)
angular_data = normalize(sparse_nn_data, norm="l2").toarray()
percent_correct = knn(knn_indices, angular_data)
assert (
percent_correct >= 0.90
), "Sparse NN-descent did not get 90% accuracy on nearest neighbors"
def test_smooth_knn_dist_l1norms(nn_data):
norms = smooth_knn(nn_data)
assert_array_almost_equal(
norms,
1.0 + np.log2(10) * np.ones(norms.shape[0]),
decimal=3,
err_msg="Smooth knn-dists does not give expected" "norms",
)
def test_smooth_knn_dist_l1norms_w_connectivity(nn_data):
norms = smooth_knn(nn_data, local_connectivity=1.75)
assert_array_almost_equal(
norms,
1.0 + np.log2(10) * np.ones(norms.shape[0]),
decimal=3,
err_msg="Smooth knn-dists does not give expected"
"norms for local_connectivity=1.75",
)
|
