Repository navigation
Expand file tree
/
Copy pathtest_graph.py
More file actions
962 lines (740 loc) · 36.4 KB
/
Copy pathtest_graph.py
File metadata and controls
962 lines (740 loc) · 36.4 KB
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
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
import os
import platform
import random
from typing import Optional
import pytest
import pathlib
import tempfile
import numpy as np
import deglib
from deglib.search import Filter
from deglib.distances import Metric, FloatSpace
IS_MACOS_M1 = platform.system() == "Darwin" and platform.machine() == "arm64"
def get_tmp_graph_file(samples: int, dims: int) -> pathlib.Path:
tmpdir = os.path.join(tempfile.gettempdir(), "deglib_test")
os.makedirs(tmpdir, exist_ok=True)
return pathlib.Path(os.path.join(tmpdir, "test_graph_S{}_D{}.deg".format(samples, dims)))
def get_ranking(features: np.ndarray, graph: deglib.DynamicExplorationGraph, query: np.ndarray) -> np.ndarray:
"""
Returns the ranking for each feature vector in the graph
"""
query = query.reshape(1, graph.get_feature_space().dim())
if features.dtype in (np.uint8, np.int8):
features = features.astype(np.float32)
if query.dtype in (np.uint8, np.int8):
query = query.astype(np.float32)
if graph.get_feature_space().metric() in (Metric.FP32_L2, Metric.Uint8_L2, Metric.Int8_L2):
distances = np.sum(np.square(features - query), axis=1)
elif graph.get_feature_space().metric() == Metric.FP32_InnerProduct:
distances = 1.0 - np.dot(features, query.T).flatten()
elif graph.get_feature_space().metric() in (Metric.Uint8_InnerProduct, Metric.Int8_InnerProduct):
distances = -np.dot(features, query.T).flatten()
else:
raise ValueError(f"unknown metric: {graph.get_feature_space().metric()}")
return np.argsort(distances)
class Configuration:
def __init__(
self,
edges_per_vertex: int,
samples: int,
dims: int,
data: np.ndarray,
graph: deglib.DynamicExplorationGraph,
graph_path: Optional[pathlib.Path],
query: np.ndarray,
metric: Metric,
):
self.edges_per_vertex = edges_per_vertex
self.samples = samples
self.dims = dims
self.data = data
self.graph = graph
self.graph_path = graph_path
self.query = query
self.metric = metric
@staticmethod
def generate(samples, dims, metric, edges_per_vertex):
rng = np.random.RandomState(42 + int(metric.value) * 17)
if metric == Metric.FP32_InnerProduct:
# normalize data
data = rng.random_sample((samples, dims)).astype(np.float32)
data /= np.linalg.norm(data, axis=1).reshape(-1, 1)
query = rng.random_sample((dims,)).astype(np.float32)
query /= np.linalg.norm(query)
elif metric == Metric.FP32_L2:
data = rng.normal(size=(samples, dims)).astype(np.float32)
query = rng.normal(size=(dims,)).astype(np.float32)
elif metric == Metric.Uint8_L2:
data = rng.randint(0, 256, size=(samples, dims)).astype(np.uint8)
query = rng.randint(0, 256, size=(dims,)).astype(np.uint8)
elif metric == Metric.Uint8_InnerProduct:
data = rng.randint(0, 256, size=(samples, dims)).astype(np.uint8)
query = rng.randint(0, 256, size=(dims,)).astype(np.uint8)
elif metric == Metric.Int8_L2:
data = rng.randint(-128, 128, size=(samples, dims)).astype(np.int8)
query = rng.randint(-128, 128, size=(dims,)).astype(np.int8)
elif metric == Metric.Int8_InnerProduct:
data = rng.randint(-128, 128, size=(samples, dims)).astype(np.int8)
query = rng.randint(-128, 128, size=(dims,)).astype(np.int8)
else:
raise ValueError(f"Unsupported metric: {metric}")
size_bounded_graph = deglib.build_from_data(
data,
edges_per_vertex=edges_per_vertex,
metric=metric,
optimization_target=deglib.builder.OptimizationTarget.LowLID,
seed=42,
)
graph_path = get_tmp_graph_file(samples, dims)
size_bounded_graph.save_graph(graph_path)
read_only_graph = deglib.load_readonly_graph(graph_path)
read_only_graph_converted = size_bounded_graph.to_readonly()
return [
Configuration(edges_per_vertex, samples, dims, data, size_bounded_graph, graph_path, query, metric),
Configuration(edges_per_vertex, samples, dims, data, read_only_graph, None, query, metric),
Configuration(edges_per_vertex, samples, dims, data, read_only_graph_converted, None, query, metric),
]
def create_new_size_bounded_graph(self):
return deglib.build_from_data(self.data, edges_per_vertex=self.edges_per_vertex, metric=self.metric)
def __repr__(self):
return f"Conf({type(self.graph).__name__}, metric={self.metric.name})"
configurations = [
*Configuration.generate(100, 128, Metric.FP32_L2, 16),
*Configuration.generate(100, 128, Metric.Uint8_L2, 16),
*Configuration.generate(100, 128, Metric.Int8_L2, 16),
*Configuration.generate(100, 128, Metric.FP32_InnerProduct, 16),
*Configuration.generate(100, 128, Metric.Uint8_InnerProduct, 16),
*Configuration.generate(100, 128, Metric.Int8_InnerProduct, 16),
]
large_configurations = [
*Configuration.generate(20_000, 2, Metric.Uint8_L2, 10),
]
mutable_configurations = [c for c in configurations if c.graph.is_mutable()]
@pytest.mark.parametrize("conf", configurations)
def test_search(conf: Configuration):
if IS_MACOS_M1 and conf.metric == Metric.FP32_InnerProduct:
pytest.skip("This test is skipped on macOS with M1 chip, as avx2 is not supported on m1 chip.")
k = 10
graph_result, dists = conf.graph.search(conf.query, eps_or_ef=0.2, k=k)
graph_result = graph_result.flatten()
correct_result = get_ranking(conf.data, conf.graph, conf.query)[:k]
assert graph_result.shape[-1] == k, "expected {} results, but got {}".format(k, graph_result.shape[-1])
# test matches are good
matches = set(graph_result).intersection(set(correct_result))
assert len(matches) >= k - 2, "expected at least {} matching results, but got only {}".format(k - 2, len(matches))
# test result is sorted
last_distance = float("-inf")
for index, distance in enumerate(dists):
assert last_distance <= distance, (
"ResultSet is not sorted.\ndistance {} at index {} larger than\ndistance {} at index {}".format(
last_distance, index - 1, distance, index
)
)
@pytest.mark.parametrize("conf", configurations)
def test_search_eps_or_ef(conf: Configuration):
"""Test both eps (< 1.0) and ef (>= 1.0) modes of eps_or_ef parameter."""
if IS_MACOS_M1 and conf.metric == Metric.FP32_InnerProduct:
pytest.skip("This test is skipped on macOS with M1 chip, as avx2 is not supported on m1 chip.")
k = 10
correct_result = get_ranking(conf.data, conf.graph, conf.query)[:k]
# 1. eps mode (< 1.0)
res_eps, dists_eps = conf.graph.search(conf.query, eps_or_ef=0.2, k=k)
assert len(res_eps) == k
assert len(dists_eps) == k
assert np.all(np.diff(dists_eps) >= -1e-6)
# 2. ef mode (>= 1.0)
res_ef, dists_ef = conf.graph.search(conf.query, eps_or_ef=64.0, k=k)
assert len(res_ef) == k
assert len(dists_ef) == k
assert np.all(np.diff(dists_ef) >= -1e-6)
matches_ef = set(res_ef).intersection(set(correct_result))
assert len(matches_ef) >= k - 2, f"Expected at least {k - 2} matches with ef=64, got {len(matches_ef)}"
# 3. Batch search with ef mode and multiple threads
queries = np.vstack([conf.query, conf.query])
batch_res, batch_dists = conf.graph.search(queries, eps_or_ef=64.0, k=k, threads=2)
assert batch_res.shape == (2, k)
assert batch_dists.shape == (2, k)
assert np.array_equal(batch_res[0], res_ef)
@pytest.mark.parametrize("conf", mutable_configurations)
def test_prune_non_mrng_edges(conf: Configuration):
graph = conf.create_new_size_bounded_graph()
deglib.optimization.prune_non_mrng_edges(graph)
@pytest.mark.parametrize("conf", configurations)
def test_threaded_search(conf: Configuration):
k = 10
graph_result, dists = conf.graph.search(conf.query, eps_or_ef=0.1, k=k)
for n_threads in range(2, 8):
threaded_graph_result, threaded_dists = conf.graph.search(conf.query, eps_or_ef=0.1, k=k, threads=n_threads)
assert np.all(np.equal(threaded_graph_result, graph_result)), (
"Threaded and non threaded results differ (n_threads={})".format(n_threads)
)
assert np.allclose(threaded_dists, dists), "Threaded and non threaded dists differ (n_threads={})".format(
n_threads
)
@pytest.mark.parametrize("conf", configurations)
def test_explore(conf: Configuration):
k = 10
include_entry = True
entry_vertex_index = random.randint(0, conf.samples - 1)
indices, distances = conf.graph.explore(
entry_vertex_index, k, max_distance_computation_count=k * 10, include_entry=include_entry
)
assert len(indices) == k
assert len(distances) == k
# check that valid (non-NaN) distances are sorted (ascending)
valid_distances = distances[~np.isnan(distances)]
for i in range(len(valid_distances) - 1):
assert valid_distances[i] <= valid_distances[i + 1], "Distances should be sorted in ascending order"
@pytest.mark.parametrize("conf", configurations)
def test_get_edges_per_vertex(conf: Configuration):
assert conf.graph.get_edges_per_vertex() == conf.edges_per_vertex
@pytest.mark.parametrize("conf", configurations)
def test_has_vertex(conf: Configuration):
assert conf.graph.has_vertex(0)
assert not conf.graph.has_vertex(conf.graph.size())
@pytest.mark.parametrize("conf", mutable_configurations)
def test_modify_graph(conf: Configuration):
graph = conf.create_new_size_bounded_graph()
# remove vertex via builder
builder = deglib.GraphBuilder(graph)
builder.remove_entry(graph.size() - 1)
builder.build()
assert graph.size() == conf.samples - 1
# add vertex back via builder
builder2 = deglib.GraphBuilder(graph)
builder2.add_entry(conf.samples - 1, conf.data[-1])
builder2.build()
@pytest.mark.parametrize("conf", mutable_configurations)
def test_load_graph(conf: Configuration):
graph = deglib.load_readonly_graph(conf.graph_path)
assert not graph.is_mutable()
with pytest.raises(FileNotFoundError):
_graph = deglib.load_readonly_graph(pathlib.Path("path") / "does" / "not" / "exist")
@pytest.mark.parametrize("conf", mutable_configurations)
def test_save_graph(tmp_path, conf):
target_path = tmp_path / "save_path.deg"
if target_path.is_file():
os.remove(target_path)
conf.graph.save_graph(target_path)
assert target_path.is_file()
os.remove(target_path)
@pytest.mark.parametrize("conf", mutable_configurations)
def test_convert_graph(conf: Configuration):
rd_graph = conf.graph.to_readonly()
assert not rd_graph.is_mutable()
@pytest.mark.parametrize("conf", large_configurations)
def test_filters(conf: Configuration):
k = 400
valid_labels = np.random.choice(conf.graph.size(), size=12_000, replace=False)
results, _dists = conf.graph.search(conf.query, filter_labels=Filter(valid_labels), eps_or_ef=0.01, k=k)
if not np.all(np.isin(results, valid_labels)):
raise ValueError("Found results that should have been filtered out.")
@pytest.mark.parametrize("conf", large_configurations)
def test_small_filters(conf: Configuration):
for n_valid in [200, 400, 600]:
k = 400
valid_labels = np.random.choice(conf.graph.size(), size=n_valid, replace=False)
results, _dists = conf.graph.search(conf.query, filter_labels=Filter(valid_labels), eps_or_ef=0.01, k=k)
assert results.shape[-1] == k
valid_results = results[results != np.iinfo(np.uint32).max]
if not np.all(np.isin(valid_results, valid_labels)):
raise ValueError("Found results that should have been filtered out.")
if n_valid < k:
num_unfilled = k - n_valid
assert np.sum(results == np.iinfo(np.uint32).max) == num_unfilled
assert np.sum(np.isnan(_dists)) == num_unfilled
@pytest.mark.parametrize("conf", configurations[:1])
def test_filter_edge_cases(conf: Configuration):
k = 5
# All valid labels filter
all_labels = np.arange(conf.graph.size(), dtype=np.int32)
results_all, _ = conf.graph.search(conf.query, filter_labels=Filter(all_labels), eps_or_ef=0.1, k=k)
assert results_all.shape[-1] == k
# No valid labels filter with max_value specified
no_labels = np.array([], dtype=np.int32)
results_none, dists_none = conf.graph.search(conf.query, filter_labels=Filter(no_labels, max_value=0), eps_or_ef=0.1, k=k)
assert results_none.shape[-1] == k
assert np.all(results_none == np.iinfo(np.uint32).max)
assert np.all(np.isnan(dists_none))
def _build_test_graph_with_distinct_labels(samples=50, dims=16, edges_per_vertex=10):
"""Build a graph with external labels that differ from internal indices."""
data = np.random.default_rng(42).standard_normal((samples, dims)).astype(np.float32)
labels = np.arange(samples, dtype=np.uint32) * 10
graph = deglib.build_from_data(
data,
labels=labels,
edges_per_vertex=edges_per_vertex,
optimization_target=deglib.builder.OptimizationTarget.LowLID,
extend_k=10,
extend_eps=0.001,
)
return graph, data
def test_has_vertex_accepts_external_label():
"""has_vertex should accept external labels, not internal indices."""
g, _ = _build_test_graph_with_distinct_labels()
# External labels are 0, 10, 20, ...
assert g.has_vertex(0) is True
assert g.has_vertex(10) is True
assert g.has_vertex(1) is False # internal index, not external label
assert g.has_vertex(99999) is False
def test_search_returns_external_labels():
"""DynamicExplorationGraph.search returns external labels."""
g, data = _build_test_graph_with_distinct_labels()
indices, distances = g.search(data[0:1], eps_or_ef=0.1, k=5)
assert indices.shape == (1, 5)
# search_batch returns external labels (multiples of 10)
for idx in indices.flatten():
if idx != 0xFFFFFFFF:
assert idx % 10 == 0, f"Search result {idx} should be an external label (multiple of 10)"
def test_facade_has_vertex_accepts_external_label():
"""DynamicExplorationGraph facade has_vertex should accept external labels."""
g, _ = _build_test_graph_with_distinct_labels()
# External labels are 0, 10, 20, ...
assert g.has_vertex(0) is True
assert g.has_vertex(10) is True
assert g.has_vertex(1) is False # internal index, not external label
def test_facade_explore_uses_external_label():
"""DynamicExplorationGraph facade explore should use external labels."""
g, _ = _build_test_graph_with_distinct_labels()
indices, distances = g.explore(0, k=5, max_distance_computation_count=50)
assert len(indices) == 5
assert len(distances) == 5
for idx in indices:
if idx != 0xFFFFFFFF:
assert idx % 10 == 0, f"Facade explore result should be external label (multiple of 10), got {idx}"
def test_facade_get_neighbors_uses_external_label():
"""DynamicExplorationGraph facade get_neighbors should return external labels."""
g, _ = _build_test_graph_with_distinct_labels()
neighbors = g.get_neighbors(0)
assert len(neighbors) == g.get_edges_per_vertex()
for n in neighbors:
if n != 0xFFFFFFFF:
assert n % 10 == 0, f"Facade get_neighbors should return external labels, got {n}"
def test_facade_search_returns_external_labels():
"""DynamicExplorationGraph facade search should return external labels."""
g, data = _build_test_graph_with_distinct_labels()
indices, distances = g.search(data[0:1], eps_or_ef=0.1, k=5)
assert indices.shape == (1, 5)
for idx in indices.flatten():
if idx != 0xFFFFFFFF:
assert idx % 10 == 0, f"Facade search should return external labels, got {idx}"
def test_facade_to_readonly():
"""DynamicExplorationGraph.to_readonly() should return a read-only graph."""
g, _ = _build_test_graph_with_distinct_labels()
rd_graph = g.to_readonly()
assert not rd_graph.is_mutable()
assert rd_graph.size() == g.size()
def test_facade_explore_return_distances_false():
"""explore with return_distances=False should return only indices array."""
g, _ = _build_test_graph_with_distinct_labels()
indices = g.explore(0, k=5, max_distance_computation_count=50, return_distances=False)
assert isinstance(indices, np.ndarray)
assert len(indices) == 5
def test_facade_explore_unsorted():
"""explore with unsorted=True should return indices."""
g, _ = _build_test_graph_with_distinct_labels()
indices, distances = g.explore(0, k=5, max_distance_computation_count=50, unsorted=True)
assert len(indices) == 5
assert len(distances) == 5
def test_facade_explore_batch():
"""explore_batch with return_distances and unsorted options."""
g, _ = _build_test_graph_with_distinct_labels()
labels = np.array([0, 10], dtype=np.uint32)
# Default return_distances=True
indices, distances = g.explore(labels, k=5, max_distance_computation_count=50)
assert indices.shape == (2, 5)
assert distances.shape == (2, 5)
# return_distances=False
indices_only = g.explore(labels, k=5, max_distance_computation_count=50, return_distances=False)
assert indices_only.shape == (2, 5)
assert isinstance(indices_only, np.ndarray)
# unsorted=True
indices_unsorted, distances_unsorted = g.explore(labels, k=5, max_distance_computation_count=50, unsorted=True)
assert indices_unsorted.shape == (2, 5)
assert distances_unsorted.shape == (2, 5)
def test_facade_has_path_reaches_adjacent_target():
"""has_path should walk from the entry to an adjacent target and return external labels."""
g, _ = _build_test_graph_with_distinct_labels()
entry = 0
neighbors = [int(n) for n in g.get_neighbors(entry)]
target = next(n for n in neighbors if n != entry and n != 0xFFFFFFFF)
path, distances = g.has_path([entry], target, eps=0.0, k=1)
assert len(path) == 2
assert int(path[0]) == target
assert int(path[1]) == entry
assert len(distances) == 2
assert distances[0] == 0.0
assert distances[1] > 0.0
def test_facade_has_path_traverses_multiple_hops():
"""has_path should backtrack the whole traversal chain, not only direct neighbors."""
g, _ = _build_test_graph_with_distinct_labels()
entry = 0
neighbors = {int(n) for n in g.get_neighbors(entry)}
far = next(label for label in range(0, 500, 10) if label != entry and label not in neighbors)
path, distances = g.has_path([entry], far, eps=0.5, k=1)
assert len(path) > 2, "a non-adjacent target requires a multi-hop path"
assert int(path[0]) == far
assert int(path[-1]) == entry
assert len(distances) == len(path)
for label in path:
assert int(label) % 10 == 0, f"has_path should return external labels, got {label}"
def test_facade_has_path_accepts_scalar_entry_without_distances():
"""has_path should accept a single entry label and honor return_distances=False."""
g, _ = _build_test_graph_with_distinct_labels()
entry = 0
target = next(int(n) for n in g.get_neighbors(entry) if int(n) != entry and int(n) != 0xFFFFFFFF)
path = g.has_path(entry, target, return_distances=False)
assert isinstance(path, np.ndarray)
assert [int(label) for label in path] == [target, entry]
def test_facade_has_path_from_multiple_entries():
"""has_path should accept several entry vertices and end the path at the entry that reached the target."""
g, data = _build_test_graph_with_distinct_labels()
entries = [0, 10, 20, 30]
target = 250
# The traversal pops the entry closest to the target first, so that one must close the path.
target_feature = data[target // 10]
closest = min(entries, key=lambda label: float(np.sum((data[label // 10] - target_feature) ** 2)))
path, distances = g.has_path(entries, target, eps=0.5, k=1)
assert len(path) == len(distances)
assert len(path) >= 2
assert int(path[0]) == target
assert int(path[-1]) == closest
def test_facade_search_return_distances_false():
"""search with return_distances=False should return only indices."""
g, data = _build_test_graph_with_distinct_labels()
indices = g.search(data[0:1], eps_or_ef=0.1, k=5, return_distances=False)
assert isinstance(indices, np.ndarray)
assert indices.shape == (1, 5)
def test_facade_search_unsorted():
"""search with unsorted=True should return results."""
g, data = _build_test_graph_with_distinct_labels()
indices, distances = g.search(data[0:2], eps_or_ef=0.1, k=5, unsorted=True)
assert indices.shape == (2, 5)
assert distances.shape == (2, 5)
def test_to_readonly_custom_features():
"""to_readonly with custom feature space and FP16 custom feature buffer."""
N, d = 20, 16
data = np.random.randn(N, d).astype(np.float32)
# Build FP32 L2 graph
fs_fp32 = FloatSpace.create(d, Metric.FP32_L2)
g = deglib.create_empty(N, fs_fp32, edges_per_vertex=4)
builder = deglib.GraphBuilder(g, extend_k=10, extend_eps=0.2)
builder.add_entry(range(N), data)
builder.build()
# Convert features to FP16
fp16_data = deglib.distances.floats_to_fp16(data)
fs_fp16 = FloatSpace.create(d, Metric.FP16_InnerProduct)
# Swap to FP16 ReadOnlyGraph
ro_g = g.to_readonly(feature_space=fs_fp16, custom_features=fp16_data)
assert ro_g.size() == N
assert not ro_g.is_mutable()
assert ro_g.get_feature_space().metric() == Metric.FP16_InnerProduct
# Search on FP16 ReadOnlyGraph
query_fp16 = deglib.distances.floats_to_fp16(data[0:2])
res_indices, res_dists = ro_g.search(query_fp16, eps_or_ef=0.1, k=3)
assert res_indices.shape == (2, 3)
assert res_dists.shape == (2, 3)
def test_to_readonly_swap_fp16_l2():
"""Test swapping a mutable graph's feature space to FP16 L2 ReadOnlyGraph."""
rng = np.random.default_rng(42)
N, d = 20, 8
data = rng.standard_normal((N, d)).astype(np.float32)
# Build FP32 L2 graph
fs_fp32 = FloatSpace.create(d, Metric.FP32_L2)
g = deglib.create_empty(N, fs_fp32, edges_per_vertex=4)
builder = deglib.GraphBuilder(g, extend_k=10, extend_eps=0.2)
builder.add_entry(range(N), data)
builder.build()
# Convert features to FP16
fp16_data = deglib.distances.floats_to_fp16(data)
fs_fp16 = FloatSpace.create(d, Metric.FP16_L2)
# Swap to FP16 L2 ReadOnlyGraph
ro_g = g.to_readonly(feature_space=fs_fp16, custom_features=fp16_data)
assert ro_g.size() == N
assert not ro_g.is_mutable()
assert ro_g.get_feature_space().metric() == Metric.FP16_L2
# Search on FP16 ReadOnlyGraph
query_fp16 = deglib.distances.floats_to_fp16(data[0:2])
res_indices, res_dists = ro_g.search(query_fp16, eps_or_ef=0.1, k=3)
assert res_indices.shape == (2, 3)
assert res_dists.shape == (2, 3)
assert res_indices[0, 0] == 0
assert res_dists[0, 0] < 1e-4
def test_create_random_graph_fp32():
"""Test DynamicExplorationGraph.create_random_graph with FP32 L2 data."""
samples = 100
dims = 8
edges_per_vertex = 10
data = np.random.default_rng(42).standard_normal((samples, dims)).astype(np.float32)
feature_space = FloatSpace.create(dims, Metric.FP32_L2)
graph = deglib.create_random_graph(data, feature_space, edges_per_vertex=edges_per_vertex, seed=7)
assert graph.size() == samples
assert graph.get_edges_per_vertex() == edges_per_vertex
assert graph.is_mutable()
# Search should work
query = data[0:1]
indices, distances = graph.search(query, eps_or_ef=0.1, k=5)
assert indices.shape == (1, 5)
assert distances.shape == (1, 5)
def test_graph_optimize_preserves_search_results():
"""graph.optimize() tunes traversal prefetch parameters but must not change search results."""
samples = 200
dims = 8
data = np.random.default_rng(1).standard_normal((samples, dims)).astype(np.float32)
feature_space = FloatSpace.create(dims, Metric.FP32_L2)
graph = deglib.create_random_graph(data, feature_space, edges_per_vertex=16, seed=7)
query = data[0:1]
before, _ = graph.search(query, eps_or_ef=0.1, k=5)
graph.optimize(sample_count=32, k=5, ef=50, seed=3)
after, _ = graph.search(query, eps_or_ef=0.1, k=5)
np.testing.assert_array_equal(before, after)
def test_create_random_graph_uint8():
"""Test DynamicExplorationGraph.create_random_graph with Uint8 L2 data."""
samples = 50
dims = 16
edges_per_vertex = 8
data = np.random.default_rng(42).integers(0, 256, size=(samples, dims), dtype=np.uint8)
feature_space = FloatSpace.create(dims, Metric.Uint8_L2)
graph = deglib.create_random_graph(data, feature_space, edges_per_vertex=edges_per_vertex, seed=42)
assert graph.size() == samples
assert graph.get_edges_per_vertex() == edges_per_vertex
def test_create_random_graph_deterministic():
"""Test that create_random_graph with the same seed produces the same graph."""
samples = 30
dims = 4
edges_per_vertex = 6
data = np.random.default_rng(99).standard_normal((samples, dims)).astype(np.float32)
feature_space = FloatSpace.create(dims, Metric.FP32_L2)
graph1 = deglib.create_random_graph(data, feature_space, edges_per_vertex=edges_per_vertex, seed=123)
graph2 = deglib.create_random_graph(data, feature_space, edges_per_vertex=edges_per_vertex, seed=123)
assert graph1.size() == graph2.size()
for v in range(samples):
assert graph1.get_neighbors(v) == graph2.get_neighbors(v)
def test_create_random_graph_inner_product():
"""Test create_random_graph with FP32 InnerProduct metric."""
samples = 40
dims = 8
edges_per_vertex = 6
data = np.random.default_rng(77).standard_normal((samples, dims)).astype(np.float32)
# Normalize for inner product
data = data / np.linalg.norm(data, axis=1, keepdims=True)
feature_space = FloatSpace.create(dims, Metric.FP32_InnerProduct)
graph = deglib.create_random_graph(data, feature_space, edges_per_vertex=edges_per_vertex, seed=7)
assert graph.size() == samples
assert graph.get_edges_per_vertex() == edges_per_vertex
def test_create_random_graph_explore():
"""Test that explore works on a graph created via create_random_graph."""
samples = 50
dims = 4
edges_per_vertex = 10
data = np.random.default_rng(55).standard_normal((samples, dims)).astype(np.float32)
feature_space = FloatSpace.create(dims, Metric.FP32_L2)
graph = deglib.create_random_graph(data, feature_space, edges_per_vertex=edges_per_vertex, seed=7)
# Explore from vertex 0
indices, distances = graph.explore(0, k=5, max_distance_computation_count=50)
assert len(indices) == 5
assert len(distances) == 5
# All indices should be valid external labels (0..samples-1)
for idx in indices:
assert idx in range(samples), f"Index {idx} out of range [0, {samples})"
def test_sizebounded_graph_from_graph():
"""Test DynamicExplorationGraph.to_mutable() creates a mutable graph from a readonly graph."""
samples = 50
dims = 16
edges_per_vertex = 10
data = np.random.default_rng(42).standard_normal((samples, dims)).astype(np.float32)
feature_space = FloatSpace.create(dims, Metric.FP32_L2)
# Build a mutable graph, then convert to readonly
graph = deglib.build_from_data(data, edges_per_vertex=edges_per_vertex, metric=Metric.FP32_L2)
readonly_graph = graph.to_readonly()
assert not readonly_graph.is_mutable()
# Convert back to mutable
mutable_graph = readonly_graph.to_mutable()
assert mutable_graph.is_mutable()
assert mutable_graph.size() == samples
assert mutable_graph.get_edges_per_vertex() == edges_per_vertex
# Search should work
query = data[0:1]
indices, distances = mutable_graph.search(query, eps_or_ef=0.1, k=5)
assert indices.shape == (1, 5)
assert distances.shape == (1, 5)
def test_sizebounded_graph_from_graph_custom_features():
"""Test to_mutable() with custom features buffer."""
samples = 30
dims = 8
edges_per_vertex = 6
data = np.random.default_rng(42).standard_normal((samples, dims)).astype(np.float32)
feature_space = FloatSpace.create(dims, Metric.FP32_L2)
graph = deglib.build_from_data(data, edges_per_vertex=edges_per_vertex, metric=Metric.FP32_L2)
# Create scaled custom features
custom_features = data * 10.0
# Convert to mutable with custom features
mutable_graph = graph.to_mutable(feature_space=feature_space, custom_features=custom_features)
assert mutable_graph.is_mutable()
assert mutable_graph.size() == samples
# Topology should match
for i in range(samples):
orig_neighbors = sorted(graph.get_neighbors(i))
new_neighbors = sorted(mutable_graph.get_neighbors(i))
assert orig_neighbors == new_neighbors, f"Neighbor mismatch at vertex {i}"
def test_sizebounded_graph_from_graph_capacity():
"""Test to_mutable() with new_max_size (capacity) parameter."""
samples = 20
dims = 4
edges_per_vertex = 4
data = np.random.default_rng(42).standard_normal((samples, dims)).astype(np.float32)
feature_space = FloatSpace.create(dims, Metric.FP32_L2)
graph = deglib.build_from_data(data, edges_per_vertex=edges_per_vertex, metric=Metric.FP32_L2)
# Convert to mutable with larger capacity
mutable_graph = graph.to_mutable(feature_space=feature_space, capacity=100)
assert mutable_graph.is_mutable()
assert mutable_graph.size() == samples
# Should be able to add a vertex via builder
builder = deglib.GraphBuilder(mutable_graph)
builder.add_entry(samples, data[0])
builder.build()
assert mutable_graph.size() == samples + 1
assert mutable_graph.has_vertex(samples)
def test_sizebounded_graph_from_graph_search_matches():
"""Test that search on a graph converted via to_mutable returns same results as original."""
samples = 50
dims = 16
edges_per_vertex = 10
data = np.random.default_rng(42).standard_normal((samples, dims)).astype(np.float32)
feature_space = FloatSpace.create(dims, Metric.FP32_L2)
graph = deglib.build_from_data(data, edges_per_vertex=edges_per_vertex, metric=Metric.FP32_L2)
readonly_graph = graph.to_readonly()
mutable_graph = readonly_graph.to_mutable()
# Search on both should return the same results
query = data[0:1]
orig_indices, orig_dists = readonly_graph.search(query, eps_or_ef=0.1, k=5)
new_indices, new_dists = mutable_graph.search(query, eps_or_ef=0.1, k=5)
assert orig_indices.shape == new_indices.shape
assert np.allclose(orig_dists, new_dists, atol=1e-5)
def test_sizebounded_graph_from_graph_inner_product():
"""Test to_mutable() with InnerProduct metric."""
samples = 30
dims = 8
edges_per_vertex = 6
data = np.random.default_rng(42).standard_normal((samples, dims)).astype(np.float32)
data = data / np.linalg.norm(data, axis=1, keepdims=True)
feature_space = FloatSpace.create(dims, Metric.FP32_InnerProduct)
graph = deglib.build_from_data(data, edges_per_vertex=edges_per_vertex, metric=Metric.FP32_InnerProduct)
mutable_graph = graph.to_mutable()
assert mutable_graph.is_mutable()
assert mutable_graph.size() == samples
assert mutable_graph.get_feature_space().metric() == Metric.FP32_InnerProduct
# Search should work
query = data[0:1]
indices, distances = mutable_graph.search(query, eps_or_ef=0.1, k=5)
assert indices.shape == (1, 5)
def test_dynamic_graph_create_empty_and_build():
"""Test creating an empty DynamicGraph and populating it with GraphBuilder."""
samples = 50
dims = 16
edges_per_vertex = 8
chunk_size = 32
data = np.random.default_rng(42).standard_normal((samples, dims)).astype(np.float32)
feature_space = FloatSpace.create(dims, Metric.FP32_L2)
graph = deglib.create_dynamic_empty(
feature_space=feature_space, edges_per_vertex=edges_per_vertex, chunk_size=chunk_size
)
assert graph.is_mutable()
assert graph.size() == 0
assert graph.get_edges_per_vertex() == edges_per_vertex
builder = deglib.GraphBuilder(graph)
labels = np.arange(samples, dtype=np.uint32)
builder.add_entry(labels, data)
builder.build()
assert graph.size() == samples
for i in range(samples):
assert graph.has_vertex(i)
# Search
query = data[0:1]
indices, distances = graph.search(query, eps_or_ef=0.1, k=5)
assert indices.shape == (1, 5)
assert indices[0, 0] == 0
assert np.isclose(distances[0, 0], 0.0, atol=1e-5)
def test_to_dynamic_conversion_and_mutation():
"""Test converting a graph to DynamicGraph via to_dynamic() and modifying it."""
samples = 40
dims = 16
edges_per_vertex = 8
data = np.random.default_rng(42).standard_normal((samples, dims)).astype(np.float32)
graph = deglib.build_from_data(data, edges_per_vertex=edges_per_vertex, metric=Metric.FP32_L2)
dyn_graph = graph.to_dynamic(chunk_size=64)
assert dyn_graph.is_mutable()
assert dyn_graph.size() == samples
# Search on converted graph
query = data[0:1]
orig_indices, orig_dists = graph.search(query, eps_or_ef=0.1, k=5)
dyn_indices, dyn_dists = dyn_graph.search(query, eps_or_ef=0.1, k=5)
assert np.array_equal(orig_indices, dyn_indices)
assert np.allclose(orig_dists, dyn_dists, atol=1e-5)
# Further mutation with GraphBuilder
new_data = np.random.default_rng(123).standard_normal((5, dims)).astype(np.float32)
new_labels = np.arange(samples, samples + 5, dtype=np.uint32)
builder = deglib.GraphBuilder(dyn_graph)
builder.add_entry(new_labels, new_data)
builder.build()
assert dyn_graph.size() == samples + 5
assert dyn_graph.has_vertex(samples + 4)
def test_save_and_load_dynamic_graph(tmp_path):
"""Test saving a graph and reloading it via load_dynamic_graph."""
samples = 30
dims = 8
edges_per_vertex = 6
data = np.random.default_rng(42).standard_normal((samples, dims)).astype(np.float32)
graph = deglib.build_from_data(data, edges_per_vertex=edges_per_vertex, metric=Metric.FP32_L2)
file_path = tmp_path / "test_dyn_graph.deg"
graph.save_graph(file_path)
loaded_graph = deglib.load_dynamic_graph(file_path, chunk_size=32)
assert loaded_graph.is_mutable()
assert loaded_graph.size() == samples
assert loaded_graph.get_edges_per_vertex() == edges_per_vertex
query = data[0:1]
orig_indices, orig_dists = graph.search(query, eps_or_ef=0.1, k=5)
loaded_indices, loaded_dists = loaded_graph.search(query, eps_or_ef=0.1, k=5)
assert np.array_equal(orig_indices, loaded_indices)
assert np.allclose(orig_dists, loaded_dists, atol=1e-5)
def test_save_and_load_mutable_graph(tmp_path):
"""Test saving a graph and reloading it via load_mutable_graph."""
samples = 30
dims = 8
edges_per_vertex = 6
data = np.random.default_rng(42).standard_normal((samples, dims)).astype(np.float32)
graph = deglib.build_from_data(data, edges_per_vertex=edges_per_vertex, metric=Metric.FP32_L2)
file_path = tmp_path / "test_mutable_graph.deg"
graph.save_graph(file_path)
# 1. Load with default capacity (size of file)
loaded_graph = deglib.load_mutable_graph(file_path)
assert loaded_graph.is_mutable()
assert loaded_graph.size() == samples
assert loaded_graph.get_edges_per_vertex() == edges_per_vertex
query = data[0:1]
orig_indices, orig_dists = graph.search(query, eps_or_ef=0.1, k=5)
loaded_indices, loaded_dists = loaded_graph.search(query, eps_or_ef=0.1, k=5)
assert np.array_equal(orig_indices, loaded_indices)
assert np.allclose(orig_dists, loaded_dists, atol=1e-5)
# 2. Load with extended capacity and add entries
loaded_with_cap = deglib.load_mutable_graph(file_path, capacity=samples + 10)
assert loaded_with_cap.is_mutable()
assert loaded_with_cap.size() == samples
builder = deglib.GraphBuilder(loaded_with_cap)
new_data = np.random.default_rng(99).standard_normal((5, dims)).astype(np.float32)
new_labels = np.arange(samples, samples + 5, dtype=np.uint32)
builder.add_entry(new_labels, new_data)
builder.build()
assert loaded_with_cap.size() == samples + 5
def test_create_helpers():
"""Test module-level and class-level graph creation helpers."""
space = deglib.FloatSpace.create(16, Metric.FP32_L2)
g1 = deglib.create_empty(100, space, 16)
assert g1.is_mutable()
assert g1.size() == 0
g2 = deglib.create_dynamic_empty(space, 16, 128)
assert g2.is_mutable()
assert g2.size() == 0
features = np.random.default_rng(42).standard_normal((20, 16)).astype(np.float32)
g3 = deglib.create_random_graph(features, space, 4)
assert g3.size() == 20