test_ondisk_dataset.py 39.5 KB
Newer Older
1
import os
2
import re
3
4
import tempfile

5
6
import gb_test_utils as gbt

7
import numpy as np
8
import pandas as pd
9
10
11

import pydantic
import pytest
12
import torch
13
import yaml
14
15
16
17
18
19
from dgl import graphbolt as gb


def test_OnDiskDataset_TVTSet_exceptions():
    """Test excpetions thrown when parsing TVTSet."""
    with tempfile.TemporaryDirectory() as test_dir:
20
21
        os.makedirs(os.path.join(test_dir, "preprocessed"), exist_ok=True)
        yaml_file = os.path.join(test_dir, "preprocessed/metadata.yaml")
22
23
24

        # Case 1: ``format`` is invalid.
        yaml_content = """
25
26
27
28
29
30
31
        tasks:
          - name: node_classification
            train_set:
              - type: paper
                data:
                  - format: torch_invalid
                    path: set/paper-train.pt
32
        """
33
        yaml_file = os.path.join(test_dir, "preprocessed/metadata.yaml")
34
35
36
        with open(yaml_file, "w") as f:
            f.write(yaml_content)
        with pytest.raises(pydantic.ValidationError):
37
            _ = gb.OnDiskDataset(test_dir)
38

39
40
        # Case 2: ``type`` is not specified while multiple TVT sets are
        # specified.
41
        yaml_content = """
42
43
44
45
46
47
48
49
50
51
52
            tasks:
              - name: node_classification
                train_set:
                - type: null
                  data:
                    - format: numpy
                      path: set/train.npy
                - type: null
                  data:
                    - format: numpy
                      path: set/train.npy
53
54
55
56
57
        """
        with open(yaml_file, "w") as f:
            f.write(yaml_content)
        with pytest.raises(
            AssertionError,
58
            match=r"Only one TVT set is allowed if type is not specified.",
59
        ):
60
            _ = gb.OnDiskDataset(test_dir)
61
62
63
64
65
66


def test_OnDiskDataset_TVTSet_ItemSet_id_label():
    """Test TVTSet which returns ItemSet with IDs and labels."""
    with tempfile.TemporaryDirectory() as test_dir:
        train_ids = np.arange(1000)
67
68
        train_ids_path = os.path.join(test_dir, "train_ids.npy")
        np.save(train_ids_path, train_ids)
69
        train_labels = np.random.randint(0, 10, size=1000)
70
71
        train_labels_path = os.path.join(test_dir, "train_labels.npy")
        np.save(train_labels_path, train_labels)
72
73

        validation_ids = np.arange(1000, 2000)
74
75
        validation_ids_path = os.path.join(test_dir, "validation_ids.npy")
        np.save(validation_ids_path, validation_ids)
76
        validation_labels = np.random.randint(0, 10, size=1000)
77
78
        validation_labels_path = os.path.join(test_dir, "validation_labels.npy")
        np.save(validation_labels_path, validation_labels)
79
80

        test_ids = np.arange(2000, 3000)
81
82
        test_ids_path = os.path.join(test_dir, "test_ids.npy")
        np.save(test_ids_path, test_ids)
83
        test_labels = np.random.randint(0, 10, size=1000)
84
85
        test_labels_path = os.path.join(test_dir, "test_labels.npy")
        np.save(test_labels_path, test_labels)
86
87
88

        # Case 1:
        #   all TVT sets are specified.
89
        #   ``type`` is not specified or specified as ``null``.
90
91
        #   ``in_memory`` could be ``true`` and ``false``.
        yaml_content = f"""
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
            tasks:
              - name: node_classification
                num_classes: 10
                train_set:
                  - type: null
                    data:
                      - format: numpy
                        in_memory: true
                        path: {train_ids_path}
                      - format: numpy
                        in_memory: true
                        path: {train_labels_path}
                validation_set:
                  - data:
                      - format: numpy
                        in_memory: true
                        path: {validation_ids_path}
                      - format: numpy
                        in_memory: true
                        path: {validation_labels_path}
                test_set:
                  - type: null
                    data:
                      - format: numpy
                        in_memory: true
                        path: {test_ids_path}
                      - format: numpy
                        in_memory: true
                        path: {test_labels_path}
121
        """
122
123
        os.makedirs(os.path.join(test_dir, "preprocessed"), exist_ok=True)
        yaml_file = os.path.join(test_dir, "preprocessed/metadata.yaml")
124
125
126
        with open(yaml_file, "w") as f:
            f.write(yaml_content)

127
        dataset = gb.OnDiskDataset(test_dir)
128

129
130
131
132
133
        # Verify tasks.
        assert len(dataset.tasks) == 1
        assert dataset.tasks[0].metadata["name"] == "node_classification"
        assert dataset.tasks[0].metadata["num_classes"] == 10

134
        # Verify train set.
135
        train_set = dataset.tasks[0].train_set
136
137
138
139
140
141
        assert len(train_set) == 1000
        assert isinstance(train_set, gb.ItemSet)
        for i, (id, label) in enumerate(train_set):
            assert id == train_ids[i]
            assert label == train_labels[i]
        train_set = None
142
143

        # Verify validation set.
144
        validation_set = dataset.tasks[0].validation_set
145
146
147
148
149
150
        assert len(validation_set) == 1000
        assert isinstance(validation_set, gb.ItemSet)
        for i, (id, label) in enumerate(validation_set):
            assert id == validation_ids[i]
            assert label == validation_labels[i]
        validation_set = None
151
152

        # Verify test set.
153
        test_set = dataset.tasks[0].test_set
154
155
156
157
158
159
        assert len(test_set) == 1000
        assert isinstance(test_set, gb.ItemSet)
        for i, (id, label) in enumerate(test_set):
            assert id == test_ids[i]
            assert label == test_labels[i]
        test_set = None
160
161
162
163
        dataset = None

        # Case 2: Some TVT sets are None.
        yaml_content = f"""
164
165
166
167
168
169
170
            tasks:
              - name: node_classification
                train_set:
                  - type: null
                    data:
                      - format: numpy
                        path: {train_ids_path}
171
        """
172
        yaml_file = os.path.join(test_dir, "preprocessed/metadata.yaml")
173
174
175
        with open(yaml_file, "w") as f:
            f.write(yaml_content)

176
        dataset = gb.OnDiskDataset(test_dir)
177
178
179
        assert dataset.tasks[0].train_set is not None
        assert dataset.tasks[0].validation_set is None
        assert dataset.tasks[0].test_set is None
180
181
182
183
184
185
        dataset = None


def test_OnDiskDataset_TVTSet_ItemSet_node_pair_label():
    """Test TVTSet which returns ItemSet with IDs and labels."""
    with tempfile.TemporaryDirectory() as test_dir:
186
187
188
189
190
191
        train_src = np.arange(1000)
        train_src_path = os.path.join(test_dir, "train_src.npy")
        np.save(train_src_path, train_src)
        train_dst = np.arange(1000, 2000)
        train_dst_path = os.path.join(test_dir, "train_dst.npy")
        np.save(train_dst_path, train_dst)
192
        train_labels = np.random.randint(0, 10, size=1000)
193
194
195
196
197
198
199
200
201
        train_labels_path = os.path.join(test_dir, "train_labels.npy")
        np.save(train_labels_path, train_labels)

        validation_src = np.arange(1000, 2000)
        validation_src_path = os.path.join(test_dir, "validation_src.npy")
        np.save(validation_src_path, validation_src)
        validation_dst = np.arange(2000, 3000)
        validation_dst_path = os.path.join(test_dir, "validation_dst.npy")
        np.save(validation_dst_path, validation_dst)
202
        validation_labels = np.random.randint(0, 10, size=1000)
203
204
205
206
207
208
209
210
211
        validation_labels_path = os.path.join(test_dir, "validation_labels.npy")
        np.save(validation_labels_path, validation_labels)

        test_src = np.arange(2000, 3000)
        test_src_path = os.path.join(test_dir, "test_src.npy")
        np.save(test_src_path, test_src)
        test_dst = np.arange(3000, 4000)
        test_dst_path = os.path.join(test_dir, "test_dst.npy")
        np.save(test_dst_path, test_dst)
212
        test_labels = np.random.randint(0, 10, size=1000)
213
214
        test_labels_path = os.path.join(test_dir, "test_labels.npy")
        np.save(test_labels_path, test_labels)
215
216

        yaml_content = f"""
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
            tasks:
              - name: link_prediction
                train_set:
                  - type: null
                    data:
                      - format: numpy
                        in_memory: true
                        path: {train_src_path}
                      - format: numpy
                        in_memory: true
                        path: {train_dst_path}
                      - format: numpy
                        in_memory: true
                        path: {train_labels_path}
                validation_set:
                  - data:
                      - format: numpy
                        in_memory: true
                        path: {validation_src_path}
                      - format: numpy
                        in_memory: true
                        path: {validation_dst_path}
                      - format: numpy
                        in_memory: true
                        path: {validation_labels_path}
                test_set:
                  - type: null
                    data:
                      - format: numpy
                        in_memory: true
                        path: {test_src_path}
                      - format: numpy
                        in_memory: true
                        path: {test_dst_path}
                      - format: numpy
                        in_memory: true
                        path: {test_labels_path}
254
        """
255
256
        os.makedirs(os.path.join(test_dir, "preprocessed"), exist_ok=True)
        yaml_file = os.path.join(test_dir, "preprocessed/metadata.yaml")
257
258
259
        with open(yaml_file, "w") as f:
            f.write(yaml_content)

260
        dataset = gb.OnDiskDataset(test_dir)
261
262

        # Verify train set.
263
        train_set = dataset.tasks[0].train_set
264
265
266
267
268
269
270
        assert len(train_set) == 1000
        assert isinstance(train_set, gb.ItemSet)
        for i, (src, dst, label) in enumerate(train_set):
            assert src == train_src[i]
            assert dst == train_dst[i]
            assert label == train_labels[i]
        train_set = None
271
272

        # Verify validation set.
273
        validation_set = dataset.tasks[0].validation_set
274
275
276
277
278
279
280
        assert len(validation_set) == 1000
        assert isinstance(validation_set, gb.ItemSet)
        for i, (src, dst, label) in enumerate(validation_set):
            assert src == validation_src[i]
            assert dst == validation_dst[i]
            assert label == validation_labels[i]
        validation_set = None
281
282

        # Verify test set.
283
        test_set = dataset.tasks[0].test_set
284
285
286
287
288
289
290
        assert len(test_set) == 1000
        assert isinstance(test_set, gb.ItemSet)
        for i, (src, dst, label) in enumerate(test_set):
            assert src == test_src[i]
            assert dst == test_dst[i]
            assert label == test_labels[i]
        test_set = None
291
292
293
        dataset = None


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
def test_OnDiskDataset_TVTSet_ItemSet_node_pair_negs():
    """Test TVTSet which returns ItemSet with node pairs and negative ones."""
    with tempfile.TemporaryDirectory() as test_dir:
        train_src = np.arange(1000)
        train_src_path = os.path.join(test_dir, "train_src.npy")
        np.save(train_src_path, train_src)
        train_dst = np.arange(1000, 2000)
        train_dst_path = os.path.join(test_dir, "train_dst.npy")
        np.save(train_dst_path, train_dst)
        train_neg_dst = np.random.choice(1000 * 10, size=1000 * 10).reshape(
            1000, 10
        )
        train_neg_dst_path = os.path.join(test_dir, "train_neg_dst.npy")
        np.save(train_neg_dst_path, train_neg_dst)

        validation_src = np.arange(1000, 2000)
        validation_src_path = os.path.join(test_dir, "validation_src.npy")
        np.save(validation_src_path, validation_src)
        validation_dst = np.arange(2000, 3000)
        validation_dst_path = os.path.join(test_dir, "validation_dst.npy")
        np.save(validation_dst_path, validation_dst)
        validation_neg_dst = train_neg_dst + 1
        validation_neg_dst_path = os.path.join(
            test_dir, "validation_neg_dst.npy"
        )
        np.save(validation_neg_dst_path, validation_neg_dst)

        test_src = np.arange(2000, 3000)
        test_src_path = os.path.join(test_dir, "test_src.npy")
        np.save(test_src_path, test_src)
        test_dst = np.arange(3000, 4000)
        test_dst_path = os.path.join(test_dir, "test_dst.npy")
        np.save(test_dst_path, test_dst)
        test_neg_dst = train_neg_dst + 2
        test_neg_dst_path = os.path.join(test_dir, "test_neg_dst.npy")
        np.save(test_neg_dst_path, test_neg_dst)

        yaml_content = f"""
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
            tasks:
              - name: link_prediction
                train_set:
                  - type: null
                    data:
                      - format: numpy
                        in_memory: true
                        path: {train_src_path}
                      - format: numpy
                        in_memory: true
                        path: {train_dst_path}
                      - format: numpy
                        in_memory: true
                        path: {train_neg_dst_path}
                validation_set:
                  - data:
                      - format: numpy
                        in_memory: true
                        path: {validation_src_path}
                      - format: numpy
                        in_memory: true
                        path: {validation_dst_path}
                      - format: numpy
                        in_memory: true
                        path: {validation_neg_dst_path}
                test_set:
                  - type: null
                    data:
                      - format: numpy
                        in_memory: true
                        path: {test_src_path}
                      - format: numpy
                        in_memory: true
                        path: {test_dst_path}
                      - format: numpy
                        in_memory: true
                        path: {test_neg_dst_path}
369
        """
370
371
        os.makedirs(os.path.join(test_dir, "preprocessed"), exist_ok=True)
        yaml_file = os.path.join(test_dir, "preprocessed/metadata.yaml")
372
373
374
        with open(yaml_file, "w") as f:
            f.write(yaml_content)

375
        dataset = gb.OnDiskDataset(test_dir)
376
377

        # Verify train set.
378
        train_set = dataset.tasks[0].train_set
379
380
381
382
383
384
385
        assert len(train_set) == 1000
        assert isinstance(train_set, gb.ItemSet)
        for i, (src, dst, negs) in enumerate(train_set):
            assert src == train_src[i]
            assert dst == train_dst[i]
            assert torch.equal(negs, torch.from_numpy(train_neg_dst[i]))
        train_set = None
386
387

        # Verify validation set.
388
        validation_set = dataset.tasks[0].validation_set
389
390
391
392
393
394
395
        assert len(validation_set) == 1000
        assert isinstance(validation_set, gb.ItemSet)
        for i, (src, dst, negs) in enumerate(validation_set):
            assert src == validation_src[i]
            assert dst == validation_dst[i]
            assert torch.equal(negs, torch.from_numpy(validation_neg_dst[i]))
        validation_set = None
396
397

        # Verify test set.
398
        test_set = dataset.tasks[0].test_set
399
400
401
402
403
404
405
        assert len(test_set) == 1000
        assert isinstance(test_set, gb.ItemSet)
        for i, (src, dst, negs) in enumerate(test_set):
            assert src == test_src[i]
            assert dst == test_dst[i]
            assert torch.equal(negs, torch.from_numpy(test_neg_dst[i]))
        test_set = None
406
407
408
        dataset = None


409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
def test_OnDiskDataset_TVTSet_ItemSetDict_id_label():
    """Test TVTSet which returns ItemSetDict with IDs and labels."""
    with tempfile.TemporaryDirectory() as test_dir:
        train_ids = np.arange(1000)
        train_labels = np.random.randint(0, 10, size=1000)
        train_data = np.vstack([train_ids, train_labels]).T
        train_path = os.path.join(test_dir, "train.npy")
        np.save(train_path, train_data)

        validation_ids = np.arange(1000, 2000)
        validation_labels = np.random.randint(0, 10, size=1000)
        validation_data = np.vstack([validation_ids, validation_labels]).T
        validation_path = os.path.join(test_dir, "validation.npy")
        np.save(validation_path, validation_data)

        test_ids = np.arange(2000, 3000)
        test_labels = np.random.randint(0, 10, size=1000)
        test_data = np.vstack([test_ids, test_labels]).T
        test_path = os.path.join(test_dir, "test.npy")
        np.save(test_path, test_data)

        yaml_content = f"""
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
            tasks:
              - name: node_classification
                train_set:
                  - type: paper
                    data:
                      - format: numpy
                        in_memory: true
                        path: {train_path}
                  - type: author
                    data:
                      - format: numpy
                        path: {train_path}
                validation_set:
                  - type: paper
                    data:
                      - format: numpy
                        path: {validation_path}
                  - type: author
                    data:
                      - format: numpy
                        path: {validation_path}
                test_set:
                  - type: paper
                    data:
                      - format: numpy
                        in_memory: false
                        path: {test_path}
                  - type: author
                    data:
                      - format: numpy
                        path: {test_path}
462
        """
463
464
        os.makedirs(os.path.join(test_dir, "preprocessed"), exist_ok=True)
        yaml_file = os.path.join(test_dir, "preprocessed/metadata.yaml")
465
466
467
        with open(yaml_file, "w") as f:
            f.write(yaml_content)

468
        dataset = gb.OnDiskDataset(test_dir)
469
470

        # Verify train set.
471
        train_set = dataset.tasks[0].train_set
472
473
474
475
476
477
478
479
480
481
482
        assert len(train_set) == 2000
        assert isinstance(train_set, gb.ItemSetDict)
        for i, item in enumerate(train_set):
            assert isinstance(item, dict)
            assert len(item) == 1
            key = list(item.keys())[0]
            assert key in ["paper", "author"]
            id, label = item[key]
            assert id == train_ids[i % 1000]
            assert label == train_labels[i % 1000]
        train_set = None
483
484

        # Verify validation set.
485
        validation_set = dataset.tasks[0].validation_set
486
487
488
489
490
491
492
493
494
495
496
        assert len(validation_set) == 2000
        assert isinstance(validation_set, gb.ItemSetDict)
        for i, item in enumerate(validation_set):
            assert isinstance(item, dict)
            assert len(item) == 1
            key = list(item.keys())[0]
            assert key in ["paper", "author"]
            id, label = item[key]
            assert id == validation_ids[i % 1000]
            assert label == validation_labels[i % 1000]
        validation_set = None
497
498

        # Verify test set.
499
        test_set = dataset.tasks[0].test_set
500
501
502
503
504
505
506
507
508
509
510
        assert len(test_set) == 2000
        assert isinstance(test_set, gb.ItemSetDict)
        for i, item in enumerate(test_set):
            assert isinstance(item, dict)
            assert len(item) == 1
            key = list(item.keys())[0]
            assert key in ["paper", "author"]
            id, label = item[key]
            assert id == test_ids[i % 1000]
            assert label == test_labels[i % 1000]
        test_set = None
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
        dataset = None


def test_OnDiskDataset_TVTSet_ItemSetDict_node_pair_label():
    """Test TVTSet which returns ItemSetDict with node pairs and labels."""
    with tempfile.TemporaryDirectory() as test_dir:
        train_pairs = (np.arange(1000), np.arange(1000, 2000))
        train_labels = np.random.randint(0, 10, size=1000)
        train_data = np.vstack([train_pairs, train_labels]).T
        train_path = os.path.join(test_dir, "train.npy")
        np.save(train_path, train_data)

        validation_pairs = (np.arange(1000, 2000), np.arange(2000, 3000))
        validation_labels = np.random.randint(0, 10, size=1000)
        validation_data = np.vstack([validation_pairs, validation_labels]).T
        validation_path = os.path.join(test_dir, "validation.npy")
        np.save(validation_path, validation_data)

        test_pairs = (np.arange(2000, 3000), np.arange(3000, 4000))
        test_labels = np.random.randint(0, 10, size=1000)
        test_data = np.vstack([test_pairs, test_labels]).T
        test_path = os.path.join(test_dir, "test.npy")
        np.save(test_path, test_data)

        yaml_content = f"""
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
            tasks:
              - name: edge_classification
                train_set:
                  - type: paper
                    data:
                      - format: numpy
                        in_memory: true
                        path: {train_path}
                  - type: author
                    data:
                      - format: numpy
                        path: {train_path}
                validation_set:
                  - type: paper
                    data:
                      - format: numpy
                        path: {validation_path}
                  - type: author
                    data:
                      - format: numpy
                        path: {validation_path}
                test_set:
                  - type: paper
                    data:
                      - format: numpy
                        in_memory: false
                        path: {test_path}
                  - type: author
                    data:
                      - format: numpy
                        path: {test_path}
567
        """
568
569
        os.makedirs(os.path.join(test_dir, "preprocessed"), exist_ok=True)
        yaml_file = os.path.join(test_dir, "preprocessed/metadata.yaml")
570
571
572
        with open(yaml_file, "w") as f:
            f.write(yaml_content)

573
        dataset = gb.OnDiskDataset(test_dir)
574
575

        # Verify train set.
576
        train_set = dataset.tasks[0].train_set
577
578
579
580
581
582
583
584
585
586
587
588
        assert len(train_set) == 2000
        assert isinstance(train_set, gb.ItemSetDict)
        for i, item in enumerate(train_set):
            assert isinstance(item, dict)
            assert len(item) == 1
            key = list(item.keys())[0]
            assert key in ["paper", "author"]
            src, dst, label = item[key]
            assert src == train_pairs[0][i % 1000]
            assert dst == train_pairs[1][i % 1000]
            assert label == train_labels[i % 1000]
        train_set = None
589
590

        # Verify validation set.
591
        validation_set = dataset.tasks[0].validation_set
592
593
594
595
596
597
598
599
600
601
602
603
        assert len(validation_set) == 2000
        assert isinstance(validation_set, gb.ItemSetDict)
        for i, item in enumerate(validation_set):
            assert isinstance(item, dict)
            assert len(item) == 1
            key = list(item.keys())[0]
            assert key in ["paper", "author"]
            src, dst, label = item[key]
            assert src == validation_pairs[0][i % 1000]
            assert dst == validation_pairs[1][i % 1000]
            assert label == validation_labels[i % 1000]
        validation_set = None
604
605

        # Verify test set.
606
        test_set = dataset.tasks[0].test_set
607
608
609
610
611
612
613
614
615
616
617
618
        assert len(test_set) == 2000
        assert isinstance(test_set, gb.ItemSetDict)
        for i, item in enumerate(test_set):
            assert isinstance(item, dict)
            assert len(item) == 1
            key = list(item.keys())[0]
            assert key in ["paper", "author"]
            src, dst, label = item[key]
            assert src == test_pairs[0][i % 1000]
            assert dst == test_pairs[1][i % 1000]
            assert label == test_labels[i % 1000]
        test_set = None
619
        dataset = None
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


def test_OnDiskDataset_Feature_heterograph():
    """Test Feature storage."""
    with tempfile.TemporaryDirectory() as test_dir:
        # Generate node data.
        node_data_paper = np.random.rand(1000, 10)
        node_data_paper_path = os.path.join(test_dir, "node_data_paper.npy")
        np.save(node_data_paper_path, node_data_paper)
        node_data_label = np.random.randint(0, 10, size=1000)
        node_data_label_path = os.path.join(test_dir, "node_data_label.npy")
        np.save(node_data_label_path, node_data_label)

        # Generate edge data.
        edge_data_writes = np.random.rand(1000, 10)
        edge_data_writes_path = os.path.join(test_dir, "edge_writes_paper.npy")
        np.save(edge_data_writes_path, edge_data_writes)
        edge_data_label = np.random.randint(0, 10, size=1000)
        edge_data_label_path = os.path.join(test_dir, "edge_data_label.npy")
        np.save(edge_data_label_path, edge_data_label)

        # Generate YAML.
        yaml_content = f"""
            feature_data:
              - domain: node
                type: paper
                name: feat
                format: numpy
                in_memory: false
                path: {node_data_paper_path}
              - domain: node
                type: paper
                name: label
                format: numpy
                in_memory: true
                path: {node_data_label_path}
              - domain: edge
                type: "author:writes:paper"
                name: feat
                format: numpy
                in_memory: false
                path: {edge_data_writes_path}
              - domain: edge
                type: "author:writes:paper"
                name: label
                format: numpy
                in_memory: true
                path: {edge_data_label_path}
        """
669
670
        os.makedirs(os.path.join(test_dir, "preprocessed"), exist_ok=True)
        yaml_file = os.path.join(test_dir, "preprocessed/metadata.yaml")
671
672
673
        with open(yaml_file, "w") as f:
            f.write(yaml_content)

674
        dataset = gb.OnDiskDataset(test_dir)
675
676

        # Verify feature data storage.
677
        feature_data = dataset.feature
678
679
680
681
        assert len(feature_data) == 4

        # Verify node feature data.
        assert torch.equal(
682
683
            feature_data.read("node", "paper", "feat"),
            torch.tensor(node_data_paper),
684
685
        )
        assert torch.equal(
686
687
            feature_data.read("node", "paper", "label"),
            torch.tensor(node_data_label),
688
689
690
691
        )

        # Verify edge feature data.
        assert torch.equal(
692
693
            feature_data.read("edge", "author:writes:paper", "feat"),
            torch.tensor(edge_data_writes),
694
695
        )
        assert torch.equal(
696
697
            feature_data.read("edge", "author:writes:paper", "label"),
            torch.tensor(edge_data_label),
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
        )

        feature_data = None
        dataset = None


def test_OnDiskDataset_Feature_homograph():
    """Test Feature storage."""
    with tempfile.TemporaryDirectory() as test_dir:
        # Generate node data.
        node_data_feat = np.random.rand(1000, 10)
        node_data_feat_path = os.path.join(test_dir, "node_data_feat.npy")
        np.save(node_data_feat_path, node_data_feat)
        node_data_label = np.random.randint(0, 10, size=1000)
        node_data_label_path = os.path.join(test_dir, "node_data_label.npy")
        np.save(node_data_label_path, node_data_label)

        # Generate edge data.
        edge_data_feat = np.random.rand(1000, 10)
        edge_data_feat_path = os.path.join(test_dir, "edge_data_feat.npy")
        np.save(edge_data_feat_path, edge_data_feat)
        edge_data_label = np.random.randint(0, 10, size=1000)
        edge_data_label_path = os.path.join(test_dir, "edge_data_label.npy")
        np.save(edge_data_label_path, edge_data_label)

        # Generate YAML.
        # ``type`` is not specified in the YAML.
        yaml_content = f"""
            feature_data:
              - domain: node
                name: feat
                format: numpy
                in_memory: false
                path: {node_data_feat_path}
              - domain: node
                name: label
                format: numpy
                in_memory: true
                path: {node_data_label_path}
              - domain: edge
                name: feat
                format: numpy
                in_memory: false
                path: {edge_data_feat_path}
              - domain: edge
                name: label
                format: numpy
                in_memory: true
                path: {edge_data_label_path}
        """
748
749
        os.makedirs(os.path.join(test_dir, "preprocessed"), exist_ok=True)
        yaml_file = os.path.join(test_dir, "preprocessed/metadata.yaml")
750
751
752
        with open(yaml_file, "w") as f:
            f.write(yaml_content)

753
        dataset = gb.OnDiskDataset(test_dir)
754
755

        # Verify feature data storage.
756
        feature_data = dataset.feature
757
758
759
        assert len(feature_data) == 4

        # Verify node feature data.
760
761
762
763
764
765
766
767
        assert torch.equal(
            feature_data.read("node", None, "feat"),
            torch.tensor(node_data_feat),
        )
        assert torch.equal(
            feature_data.read("node", None, "label"),
            torch.tensor(node_data_label),
        )
768
769

        # Verify edge feature data.
770
771
772
773
774
775
776
777
778
        assert torch.equal(
            feature_data.read("edge", None, "feat"),
            torch.tensor(edge_data_feat),
        )
        assert torch.equal(
            feature_data.read("edge", None, "label"),
            torch.tensor(edge_data_label),
        )

779
780
        feature_data = None
        dataset = None
781
782
783
784
785
786
787
788
789
790
791


def test_OnDiskDataset_Graph_Exceptions():
    """Test exceptions in parsing graph topology."""
    with tempfile.TemporaryDirectory() as test_dir:
        # Invalid graph type.
        yaml_content = """
            graph_topology:
              type: CSRSamplingGraph
              path: /path/to/graph
        """
792
793
        os.makedirs(os.path.join(test_dir, "preprocessed"), exist_ok=True)
        yaml_file = os.path.join(test_dir, "preprocessed/metadata.yaml")
794
795
796
797
798
        with open(yaml_file, "w") as f:
            f.write(yaml_content)

        with pytest.raises(
            pydantic.ValidationError,
799
            match="1 validation error for OnDiskMetaData",
800
        ):
801
            _ = gb.OnDiskDataset(test_dir)
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817


def test_OnDiskDataset_Graph_homogeneous():
    """Test homogeneous graph topology."""
    csc_indptr, indices = gbt.random_homo_graph(1000, 10 * 1000)
    graph = gb.from_csc(csc_indptr, indices)

    with tempfile.TemporaryDirectory() as test_dir:
        graph_path = os.path.join(test_dir, "csc_sampling_graph.tar")
        gb.save_csc_sampling_graph(graph, graph_path)

        yaml_content = f"""
            graph_topology:
              type: CSCSamplingGraph
              path: {graph_path}
        """
818
819
        os.makedirs(os.path.join(test_dir, "preprocessed"), exist_ok=True)
        yaml_file = os.path.join(test_dir, "preprocessed/metadata.yaml")
820
821
822
        with open(yaml_file, "w") as f:
            f.write(yaml_content)

823
        dataset = gb.OnDiskDataset(test_dir)
824
        graph2 = dataset.graph
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848

        assert graph.num_nodes == graph2.num_nodes
        assert graph.num_edges == graph2.num_edges

        assert torch.equal(graph.csc_indptr, graph2.csc_indptr)
        assert torch.equal(graph.indices, graph2.indices)

        assert graph.metadata is None and graph2.metadata is None
        assert (
            graph.node_type_offset is None and graph2.node_type_offset is None
        )
        assert graph.type_per_edge is None and graph2.type_per_edge is None


def test_OnDiskDataset_Graph_heterogeneous():
    """Test heterogeneous graph topology."""
    (
        csc_indptr,
        indices,
        node_type_offset,
        type_per_edge,
        metadata,
    ) = gbt.random_hetero_graph(1000, 10 * 1000, 3, 4)
    graph = gb.from_csc(
849
        csc_indptr, indices, node_type_offset, type_per_edge, None, metadata
850
851
852
853
854
855
856
857
858
859
860
    )

    with tempfile.TemporaryDirectory() as test_dir:
        graph_path = os.path.join(test_dir, "csc_sampling_graph.tar")
        gb.save_csc_sampling_graph(graph, graph_path)

        yaml_content = f"""
            graph_topology:
              type: CSCSamplingGraph
              path: {graph_path}
        """
861
862
        os.makedirs(os.path.join(test_dir, "preprocessed"), exist_ok=True)
        yaml_file = os.path.join(test_dir, "preprocessed/metadata.yaml")
863
864
865
        with open(yaml_file, "w") as f:
            f.write(yaml_content)

866
        dataset = gb.OnDiskDataset(test_dir)
867
        graph2 = dataset.graph
868
869
870
871
872
873
874
875
876
877

        assert graph.num_nodes == graph2.num_nodes
        assert graph.num_edges == graph2.num_edges

        assert torch.equal(graph.csc_indptr, graph2.csc_indptr)
        assert torch.equal(graph.indices, graph2.indices)
        assert torch.equal(graph.node_type_offset, graph2.node_type_offset)
        assert torch.equal(graph.type_per_edge, graph2.type_per_edge)
        assert graph.metadata.node_type_to_id == graph2.metadata.node_type_to_id
        assert graph.metadata.edge_type_to_id == graph2.metadata.edge_type_to_id
878
879
880
881
882
883
884
885
886
887


def test_OnDiskDataset_Metadata():
    """Test metadata of OnDiskDataset."""
    with tempfile.TemporaryDirectory() as test_dir:
        # All metadata fields are specified.
        dataset_name = "graphbolt_test"
        yaml_content = f"""
            dataset_name: {dataset_name}
        """
888
889
        os.makedirs(os.path.join(test_dir, "preprocessed"), exist_ok=True)
        yaml_file = os.path.join(test_dir, "preprocessed/metadata.yaml")
890
891
892
        with open(yaml_file, "w") as f:
            f.write(yaml_content)

893
        dataset = gb.OnDiskDataset(test_dir)
894
895
896
897
898
899
        assert dataset.dataset_name == dataset_name

        # Only dataset_name is specified.
        yaml_content = f"""
            dataset_name: {dataset_name}
        """
900
        yaml_file = os.path.join(test_dir, "preprocessed/metadata.yaml")
901
902
903
        with open(yaml_file, "w") as f:
            f.write(yaml_content)

904
        dataset = gb.OnDiskDataset(test_dir)
905
        assert dataset.dataset_name == dataset_name
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
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980


def test_OnDiskDataset_preprocess_homogeneous():
    """Test preprocess of OnDiskDataset."""
    with tempfile.TemporaryDirectory() as test_dir:
        # All metadata fields are specified.
        dataset_name = "graphbolt_test"
        num_nodes = 4000
        num_edges = 20000
        num_classes = 10

        # Generate random edges.
        nodes = np.repeat(np.arange(num_nodes), 5)
        neighbors = np.random.randint(0, num_nodes, size=(num_edges))
        edges = np.stack([nodes, neighbors], axis=1)
        # Wrtie into edges/edge.csv
        os.makedirs(os.path.join(test_dir, "edges/"), exist_ok=True)
        edges = pd.DataFrame(edges, columns=["src", "dst"])
        edges.to_csv(
            os.path.join(test_dir, "edges/edge.csv"),
            index=False,
            header=False,
        )

        # Generate random graph edge-feats.
        edge_feats = np.random.rand(num_edges, 5)
        os.makedirs(os.path.join(test_dir, "data/"), exist_ok=True)
        np.save(os.path.join(test_dir, "data/edge-feat.npy"), edge_feats)

        # Generate random node-feats.
        node_feats = np.random.rand(num_nodes, 10)
        np.save(os.path.join(test_dir, "data/node-feat.npy"), node_feats)

        # Generate train/test/valid set.
        os.makedirs(os.path.join(test_dir, "set/"), exist_ok=True)
        train_pairs = (np.arange(1000), np.arange(1000, 2000))
        train_labels = np.random.randint(0, 10, size=1000)
        train_data = np.vstack([train_pairs, train_labels]).T
        train_path = os.path.join(test_dir, "set/train.npy")
        np.save(train_path, train_data)

        validation_pairs = (np.arange(1000, 2000), np.arange(2000, 3000))
        validation_labels = np.random.randint(0, 10, size=1000)
        validation_data = np.vstack([validation_pairs, validation_labels]).T
        validation_path = os.path.join(test_dir, "set/validation.npy")
        np.save(validation_path, validation_data)

        test_pairs = (np.arange(2000, 3000), np.arange(3000, 4000))
        test_labels = np.random.randint(0, 10, size=1000)
        test_data = np.vstack([test_pairs, test_labels]).T
        test_path = os.path.join(test_dir, "set/test.npy")
        np.save(test_path, test_data)

        yaml_content = f"""
            dataset_name: {dataset_name}
            graph: # graph structure and required attributes.
                nodes:
                    - num: {num_nodes}
                edges:
                    - format: csv
                      path: edges/edge.csv
                feature_data:
                    - domain: edge
                      type: null
                      name: feat
                      format: numpy
                      in_memory: true
                      path: data/edge-feat.npy
            feature_data:
                - domain: node
                  type: null
                  name: feat
                  format: numpy
                  in_memory: false
                  path: data/node-feat.npy
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
            tasks:
              - name: node_classification
                num_classes: {num_classes}
                train_set:
                  - type_name: null
                    data:
                      - format: numpy
                        path: set/train.npy
                validation_set:
                  - type_name: null
                    data:
                      - format: numpy
                        path: set/validation.npy
                test_set:
                  - type_name: null
                    data:
                      - format: numpy
                        path: set/test.npy
999
        """
1000
        yaml_file = os.path.join(test_dir, "metadata.yaml")
1001
1002
        with open(yaml_file, "w") as f:
            f.write(yaml_content)
1003
        output_file = gb.ondisk_dataset.preprocess_ondisk_dataset(test_dir)
1004
1005

        with open(output_file, "rb") as f:
1006
            processed_dataset = yaml.load(f, Loader=yaml.Loader)
1007
1008

        assert processed_dataset["dataset_name"] == dataset_name
1009
        assert processed_dataset["tasks"][0]["num_classes"] == num_classes
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
        assert "graph" not in processed_dataset
        assert "graph_topology" in processed_dataset

        csc_sampling_graph = gb.csc_sampling_graph.load_csc_sampling_graph(
            os.path.join(test_dir, processed_dataset["graph_topology"]["path"])
        )
        assert csc_sampling_graph.num_nodes == num_nodes
        assert csc_sampling_graph.num_edges == num_edges

        num_samples = 100
        fanout = 1
        subgraph = csc_sampling_graph.sample_neighbors(
            torch.arange(num_samples),
            torch.tensor([fanout]),
        )
        assert len(list(subgraph.node_pairs.values())[0][0]) <= num_samples
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065


def test_OnDiskDataset_preprocess_path():
    """Test if the preprocess function can catch the path error."""
    with tempfile.TemporaryDirectory() as test_dir:
        # All metadata fields are specified.
        dataset_name = "graphbolt_test"
        num_classes = 10

        yaml_content = f"""
            dataset_name: {dataset_name}
        """
        yaml_file = os.path.join(test_dir, "metadata.yaml")
        with open(yaml_file, "w") as f:
            f.write(yaml_content)

        # Case1. Test the passed in is the yaml file path.
        with pytest.raises(
            RuntimeError,
            match="The dataset must be a directory. "
            rf"But got {re.escape(yaml_file)}",
        ):
            _ = gb.OnDiskDataset(yaml_file)

        # Case2. Test the passed in is a fake directory.
        fake_dir = os.path.join(test_dir, "fake_dir")
        with pytest.raises(
            RuntimeError,
            match=rf"Invalid dataset path: {re.escape(fake_dir)}",
        ):
            _ = gb.OnDiskDataset(fake_dir)

        # Case3. Test the passed in is the dataset directory.
        # But the metadata.yaml is not in the directory.
        os.makedirs(os.path.join(test_dir, "fake_dir"), exist_ok=True)
        with pytest.raises(
            RuntimeError,
            match=r"metadata.yaml does not exist.",
        ):
            _ = gb.OnDiskDataset(fake_dir)