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
|
import numba
import numpy as np
import heapq
from scipy.sparse import coo_matrix
from scipy.sparse.csgraph import connected_components
from itertools import combinations
import pynndescent.distances as pynnd_dist
import joblib
from pynndescent.utils import (
rejection_sample,
make_heap,
deheap_sort,
simple_heap_push,
has_been_visited,
mark_visited,
)
FLOAT32_EPS = np.finfo(np.float32).eps
def create_component_search(index):
alternative_dot = pynnd_dist.alternative_dot
alternative_cosine = pynnd_dist.alternative_cosine
data = index._raw_data
indptr = index._search_graph.indptr
indices = index._search_graph.indices
dist = index._distance_func
@numba.njit(
fastmath=True,
nogil=True,
locals={
"current_query": numba.types.float32[::1],
"i": numba.types.uint32,
"j": numba.types.uint32,
"heap_priorities": numba.types.float32[::1],
"heap_indices": numba.types.int32[::1],
"candidate": numba.types.int32,
"vertex": numba.types.int32,
"d": numba.types.float32,
"d_vertex": numba.types.float32,
"visited": numba.types.uint8[::1],
"indices": numba.types.int32[::1],
"indptr": numba.types.int32[::1],
"data": numba.types.float32[:, ::1],
"heap_size": numba.types.int16,
"distance_scale": numba.types.float32,
"distance_bound": numba.types.float32,
"seed_scale": numba.types.float32,
},
)
def custom_search_closure(query_points, candidate_indices, k, epsilon, visited):
result = make_heap(query_points.shape[0], k)
distance_scale = 1.0 + epsilon
for i in range(query_points.shape[0]):
visited[:] = 0
if dist == alternative_dot or dist == alternative_cosine:
norm = np.sqrt((query_points[i] ** 2).sum())
if norm > 0.0:
current_query = query_points[i] / norm
else:
continue
else:
current_query = query_points[i]
heap_priorities = result[1][i]
heap_indices = result[0][i]
seed_set = [(np.float32(np.inf), np.int32(-1)) for j in range(0)]
############ Init ################
n_initial_points = candidate_indices.shape[0]
for j in range(n_initial_points):
candidate = np.int32(candidate_indices[j])
d = dist(data[candidate], current_query)
# indices are guaranteed different
simple_heap_push(heap_priorities, heap_indices, d, candidate)
heapq.heappush(seed_set, (d, candidate))
mark_visited(visited, candidate)
############ Search ##############
distance_bound = distance_scale * heap_priorities[0]
# Find smallest seed point
d_vertex, vertex = heapq.heappop(seed_set)
while d_vertex < distance_bound:
for j in range(indptr[vertex], indptr[vertex + 1]):
candidate = indices[j]
if has_been_visited(visited, candidate) == 0:
mark_visited(visited, candidate)
d = dist(data[candidate], current_query)
if d < distance_bound:
simple_heap_push(
heap_priorities, heap_indices, d, candidate
)
heapq.heappush(seed_set, (d, candidate))
# Update bound
distance_bound = distance_scale * heap_priorities[0]
# find new smallest seed point
if len(seed_set) == 0:
break
else:
d_vertex, vertex = heapq.heappop(seed_set)
return result
return custom_search_closure
# @numba.njit(nogil=True)
def find_component_connection_edge(
component1,
component2,
search_closure,
raw_data,
visited,
rng_state,
search_size=10,
epsilon=0.0,
):
indices = [np.zeros(1, dtype=np.int64) for i in range(2)]
indices[0] = component1[
rejection_sample(np.int64(search_size), component1.shape[0], rng_state)
]
indices[1] = component2[
rejection_sample(np.int64(search_size), component2.shape[0], rng_state)
]
query_side = 0
query_points = raw_data[indices[query_side]]
candidate_indices = indices[1 - query_side].copy()
changed = [True, True]
best_dist = np.inf
best_edge = (indices[0][0], indices[1][0])
while changed[0] or changed[1]:
inds, dists, _ = search_closure(
query_points, candidate_indices, search_size, epsilon, visited
)
inds, dists = deheap_sort(inds, dists)
for i in range(dists.shape[0]):
for j in range(dists.shape[1]):
if dists[i, j] < best_dist:
best_dist = dists[i, j]
best_edge = (indices[query_side][i], inds[i, j])
candidate_indices = indices[query_side]
new_indices = np.unique(inds[:, 0])
if indices[1 - query_side].shape[0] == new_indices.shape[0]:
changed[1 - query_side] = np.any(indices[1 - query_side] != new_indices)
indices[1 - query_side] = new_indices
query_points = raw_data[indices[1 - query_side]]
query_side = 1 - query_side
return best_edge[0], best_edge[1], best_dist
def adjacency_matrix_representation(neighbor_indices, neighbor_distances):
result = coo_matrix(
(neighbor_indices.shape[0], neighbor_indices.shape[0]), dtype=np.float32
)
# Preserve any distance 0 points
neighbor_distances[neighbor_distances == 0.0] = FLOAT32_EPS
result.row = np.repeat(
np.arange(neighbor_indices.shape[0], dtype=np.int32), neighbor_indices.shape[1]
)
result.col = neighbor_indices.ravel()
result.data = neighbor_distances.ravel()
# Get rid of any -1 index entries
result = result.tocsr()
result.data[result.indices == -1] = 0.0
result.eliminate_zeros()
# Symmetrize
result = result.maximum(result.T)
return result
def connect_graph(graph, index, search_size=10, n_jobs=None):
search_closure = create_component_search(index)
n_components, component_ids = connected_components(graph)
result = graph.tolil()
# Translate component ids into internal vertex order
component_ids = component_ids[index._vertex_order]
def new_edge(c1, c2):
component1 = np.where(component_ids == c1)[0]
component2 = np.where(component_ids == c2)[0]
i, j, d = find_component_connection_edge(
component1,
component2,
search_closure,
index._raw_data,
index._visited,
index.rng_state,
search_size=search_size,
)
# Correct the distance if required
if index._distance_correction is not None:
d = index._distance_correction(d)
# Convert indices to original data order
i = index._vertex_order[i]
j = index._vertex_order[j]
return i, j, d
new_edges = joblib.Parallel(n_jobs=n_jobs, prefer="threads")(
joblib.delayed(new_edge)(c1, c2)
for c1, c2 in combinations(range(n_components), 2)
)
for i, j, d in new_edges:
result[i, j] = d
result[j, i] = d
return result.tocsr()
|
