File size: 25,846 Bytes
24596ed
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""
Create a deployment-realistic VANET-IDS26 test split with a strict 99:1
benign-to-attack ratio.

The public VANET-IDS26 release is intentionally attack-heavy for research
coverage. A deployed NIDS usually sees benign safety traffic dominate the
stream, so this script builds an operational test split where attacks are rare
but still represented across all 26 attack families.

Examples
--------
Quick schema/sampler path for the public sample file:

    python scripts/create_realistic_vanet_ids26_test_split.py --sample

Chunked path for the full master file:

    python scripts/create_realistic_vanet_ids26_test_split.py ^
        --input research_ready/vanet_ids26_master.csv.gz ^
        --output research_ready/vanet_ids26_test_99_1.csv.gz ^
        --mode chunked ^
        --attack-rows 2600

Note: the bundled sample has only 1,000 benign rows. A strict 99:1 split that
also includes all 26 attack classes requires at least 2,574 benign rows
(26 attacks * 99 benign per attack), so use the master file for the final split.
"""

from __future__ import annotations

import argparse
import gzip
import json
from datetime import datetime, timezone
from pathlib import Path
from typing import Dict, Iterable, Mapping

import pandas as pd
from sklearn.utils import check_random_state


ROOT = Path(__file__).resolve().parents[1]
DEFAULT_MASTER = ROOT / "research_ready" / "vanet_ids26_master.csv.gz"
DEFAULT_SAMPLE = ROOT / "release" / "github" / "VANET-IDS26" / "samples" / "vanet_ids26_sample.csv.gz"
DEFAULT_OUTPUT = ROOT / "research_ready" / "vanet_ids26_test_99_1.csv.gz"
DEFAULT_MANIFEST_DIR = ROOT / "dataset" / "manifests"

BINARY_LABEL = "binary_label"
MULTICLASS_LABEL = "multiclass_label"
BENIGN_LABEL = 0
ATTACK_LABELS = tuple(range(1, 27))
BENIGN_TO_ATTACK_RATIO = 99

ATTACK_TYPES = {
    1: "constant_position",
    2: "position_offset",
    3: "random_position",
    4: "speed_manipulation",
    5: "acceleration_manipulation",
    6: "heading_manipulation",
    7: "lane_spoofing",
    8: "impossible_kinematics",
    9: "eventual_stop",
    10: "false_brake_event",
    11: "false_emergency_vehicle",
    12: "false_hazard_event",
    13: "replay",
    14: "delayed_message",
    15: "timestamp_shift",
    16: "stale_message_replay",
    17: "sybil",
    18: "impersonation",
    19: "pseudonym_abuse",
    20: "flooding_ddos",
    21: "beacon_rate_abuse",
    22: "gnss_spoofing",
    23: "map_location_spoofing",
    24: "ghost_vehicle",
    25: "false_object_injection",
    26: "object_position_shift",
}
ATTACK_TYPE_TO_LABEL = {name: label for label, name in ATTACK_TYPES.items()}


def parse_args() -> argparse.Namespace:
    parser = argparse.ArgumentParser(
        description="Build a strict 99:1 benign/attack VANET-IDS26 test split."
    )
    parser.add_argument(
        "--input",
        type=Path,
        default=DEFAULT_MASTER,
        help=f"Input CSV/CSV.GZ path. Default: {DEFAULT_MASTER}",
    )
    parser.add_argument(
        "--output",
        type=Path,
        default=DEFAULT_OUTPUT,
        help=f"Output CSV/CSV.GZ path. Default: {DEFAULT_OUTPUT}",
    )
    parser.add_argument(
        "--sample",
        action="store_true",
        help=f"Use the small public sample file with in-memory loading: {DEFAULT_SAMPLE}",
    )
    parser.add_argument(
        "--mode",
        choices=("chunked", "memory"),
        default="chunked",
        help="Use chunked two-pass sampling for massive files or memory mode for small files.",
    )
    parser.add_argument(
        "--chunksize",
        type=int,
        default=250_000,
        help="Rows per chunk for chunked mode.",
    )
    parser.add_argument(
        "--attack-rows",
        type=int,
        default=2_600,
        help=(
            "Exact number of attack rows to include. Benign rows are set to "
            "attack_rows * 99. Default: 2600, which is 100 rows per attack "
            "family with --attack-strategy balanced."
        ),
    )
    parser.add_argument(
        "--use-largest-feasible",
        action="store_true",
        help=(
            "Ignore --attack-rows and build the largest strict 99:1 split supported "
            "by the input. Use carefully on the full master file."
        ),
    )
    parser.add_argument(
        "--attack-strategy",
        choices=("balanced", "proportional"),
        default="balanced",
        help=(
            "balanced gives each attack family nearly equal support; proportional "
            "preserves the source attack-family distribution as closely as possible."
        ),
    )
    parser.add_argument(
        "--random-state",
        type=int,
        default=42,
        help="Random seed for reproducible sampling.",
    )
    parser.add_argument(
        "--count-source",
        choices=("auto", "manifests", "scan"),
        default="auto",
        help=(
            "Where to get source label counts. auto uses local manifests when present "
            "and falls back to scanning the input."
        ),
    )
    parser.add_argument(
        "--manifest-dir",
        type=Path,
        default=DEFAULT_MANIFEST_DIR,
        help=f"Directory containing local source manifests. Default: {DEFAULT_MANIFEST_DIR}",
    )
    return parser.parse_args()


def label_counts_from_memory(df: pd.DataFrame) -> Dict[int, int]:
    _require_label_columns(df.columns)
    counts = df[MULTICLASS_LABEL].astype("int64").value_counts().sort_index()
    return {int(label): int(count) for label, count in counts.items()}


def label_counts_from_chunks(path: Path, chunksize: int) -> Dict[int, int]:
    counts: Dict[int, int] = {}
    for chunk in pd.read_csv(
        path,
        usecols=[BINARY_LABEL, MULTICLASS_LABEL],
        chunksize=chunksize,
        low_memory=False,
    ):
        _require_label_columns(chunk.columns)
        chunk_counts = chunk[MULTICLASS_LABEL].astype("int64").value_counts()
        for label, count in chunk_counts.items():
            counts[int(label)] = counts.get(int(label), 0) + int(count)
    return dict(sorted(counts.items()))


def label_counts_from_manifests(manifest_dir: Path) -> Dict[int, int]:
    benign_manifest = manifest_dir / "big_benign_runs_manifest.csv"
    attack_manifest = manifest_dir / "overlay_manifest.csv"

    if not benign_manifest.exists() or not attack_manifest.exists():
        missing = [
            str(path)
            for path in (benign_manifest, attack_manifest)
            if not path.exists()
        ]
        raise FileNotFoundError("Missing source manifest(s): " + ", ".join(missing))

    benign = pd.read_csv(benign_manifest)
    attacks = pd.read_csv(attack_manifest)
    required_benign_cols = {"records_mobility"}
    required_attack_cols = {"attack_type", "overlay_rows"}
    if not required_benign_cols.issubset(benign.columns):
        raise ValueError(f"{benign_manifest} is missing {required_benign_cols}")
    if not required_attack_cols.issubset(attacks.columns):
        raise ValueError(f"{attack_manifest} is missing {required_attack_cols}")

    counts = {BENIGN_LABEL: int(benign["records_mobility"].sum())}
    grouped = attacks.groupby("attack_type")["overlay_rows"].sum()
    for attack_type, rows in grouped.items():
        if attack_type not in ATTACK_TYPE_TO_LABEL:
            raise ValueError(f"Unknown attack_type in manifest: {attack_type}")
        counts[ATTACK_TYPE_TO_LABEL[attack_type]] = int(rows)
    return dict(sorted(counts.items()))


def get_source_counts(path: Path, chunksize: int, count_source: str, manifest_dir: Path) -> tuple[Dict[int, int], str]:
    if count_source in ("auto", "manifests"):
        try:
            return label_counts_from_manifests(manifest_dir), "manifests"
        except (FileNotFoundError, ValueError):
            if count_source == "manifests":
                raise

    return label_counts_from_chunks(path, chunksize), "scan"


def make_attack_quotas(
    label_counts: Mapping[int, int],
    attack_rows: int | None,
    strategy: str,
) -> Dict[int, int]:
    _validate_source_counts(label_counts)

    benign_rows_available = label_counts.get(BENIGN_LABEL, 0)
    max_attack_by_benign = benign_rows_available // BENIGN_TO_ATTACK_RATIO
    max_attack_by_attacks = sum(label_counts[label] for label in ATTACK_LABELS)
    max_attack_rows = min(max_attack_by_benign, max_attack_by_attacks)

    if attack_rows is None:
        attack_rows = max_attack_rows
    if attack_rows < len(ATTACK_LABELS):
        minimum_benign = len(ATTACK_LABELS) * BENIGN_TO_ATTACK_RATIO
        raise ValueError(
            "Cannot satisfy strict 99:1 and all 26 attack classes with "
            f"{attack_rows} attack rows. Need at least {len(ATTACK_LABELS)} "
            f"attack rows and {minimum_benign} benign rows."
        )
    if attack_rows > max_attack_rows:
        raise ValueError(
            f"Requested {attack_rows} attack rows, but this input supports at most "
            f"{max_attack_rows} with a strict 99:1 ratio. Available benign rows: "
            f"{benign_rows_available}; available attack rows: {max_attack_by_attacks}."
        )

    if strategy == "balanced":
        return _balanced_attack_quotas(label_counts, attack_rows)
    if strategy == "proportional":
        return _proportional_attack_quotas(label_counts, attack_rows)
    raise ValueError(f"Unknown attack strategy: {strategy}")


def _balanced_attack_quotas(label_counts: Mapping[int, int], attack_rows: int) -> Dict[int, int]:
    quotas = {label: 1 for label in ATTACK_LABELS}
    remaining = attack_rows - len(ATTACK_LABELS)

    while remaining:
        progressed = False
        for label in ATTACK_LABELS:
            if quotas[label] >= label_counts[label]:
                continue
            quotas[label] += 1
            remaining -= 1
            progressed = True
            if remaining == 0:
                break
        if not progressed:
            raise ValueError("Attack quotas exceed available rows for the 26 attack classes.")
    return quotas


def _proportional_attack_quotas(label_counts: Mapping[int, int], attack_rows: int) -> Dict[int, int]:
    total_available = sum(label_counts[label] for label in ATTACK_LABELS)
    raw = {
        label: (label_counts[label] / total_available) * attack_rows
        for label in ATTACK_LABELS
    }
    quotas = {
        label: min(label_counts[label], max(1, int(raw[label])))
        for label in ATTACK_LABELS
    }

    while sum(quotas.values()) > attack_rows:
        candidates = [label for label in ATTACK_LABELS if quotas[label] > 1]
        label = max(candidates, key=lambda item: (quotas[item] - raw[item], quotas[item]))
        quotas[label] -= 1

    while sum(quotas.values()) < attack_rows:
        candidates = [label for label in ATTACK_LABELS if quotas[label] < label_counts[label]]
        if not candidates:
            raise ValueError("Attack quotas exceed available rows for the 26 attack classes.")
        label = max(candidates, key=lambda item: (raw[item] - quotas[item], label_counts[item]))
        quotas[label] += 1

    return quotas


def build_split_in_memory(
    input_path: Path,
    output_path: Path,
    attack_rows: int | None,
    attack_strategy: str,
    random_state: int,
) -> pd.DataFrame:
    df = pd.read_csv(input_path, low_memory=False)
    counts = label_counts_from_memory(df)
    attack_quotas = make_attack_quotas(counts, attack_rows, attack_strategy)
    benign_rows = sum(attack_quotas.values()) * BENIGN_TO_ATTACK_RATIO

    rng = check_random_state(random_state)
    sampled_frames = [
        df.loc[df[MULTICLASS_LABEL].astype("int64") == BENIGN_LABEL].sample(
            n=benign_rows,
            random_state=int(rng.randint(0, 2**31 - 1)),
            replace=False,
        )
    ]

    for label, quota in attack_quotas.items():
        sampled_frames.append(
            df.loc[df[MULTICLASS_LABEL].astype("int64") == label].sample(
                n=quota,
                random_state=int(rng.randint(0, 2**31 - 1)),
                replace=False,
            )
        )

    final_df = _shuffle(pd.concat(sampled_frames, ignore_index=True), rng)
    write_output(final_df, output_path)
    print_evaluation(final_df, counts, attack_quotas, output_path)
    return final_df


def build_split_chunked(
    input_path: Path,
    output_path: Path,
    chunksize: int,
    attack_rows: int | None,
    attack_strategy: str,
    random_state: int,
    count_source: str,
    manifest_dir: Path,
) -> None:
    counts, resolved_count_source = get_source_counts(input_path, chunksize, count_source, manifest_dir)
    attack_quotas = make_attack_quotas(counts, attack_rows, attack_strategy)
    benign_rows = sum(attack_quotas.values()) * BENIGN_TO_ATTACK_RATIO

    rng = check_random_state(random_state)
    reservoirs: Dict[int, pd.DataFrame] = {}
    final_counts = {label: 0 for label in (BENIGN_LABEL, *ATTACK_LABELS)}
    tmp_output_path = output_path.with_name(output_path.name + ".tmp")

    output_path.parent.mkdir(parents=True, exist_ok=True)
    if tmp_output_path.exists():
        tmp_output_path.unlink()

    print(f"Source counts loaded from: {resolved_count_source}")
    print(f"Target benign rows: {benign_rows:,}")
    print(f"Target attack rows: {sum(attack_quotas.values()):,}")
    print(f"Writing temporary output: {tmp_output_path}")

    with _open_text_output(tmp_output_path) as out_f:
        wrote_header = False
        for chunk_number, chunk in enumerate(
            pd.read_csv(input_path, chunksize=chunksize, low_memory=False),
            start=1,
        ):
            _require_label_columns(chunk.columns)
            labels = chunk[MULTICLASS_LABEL].astype("int64")

            benign_remaining = benign_rows - final_counts[BENIGN_LABEL]
            if benign_remaining > 0:
                benign_chunk = chunk.loc[labels == BENIGN_LABEL]
                if len(benign_chunk) > benign_remaining:
                    benign_chunk = benign_chunk.iloc[:benign_remaining]
                if not benign_chunk.empty:
                    benign_chunk.to_csv(out_f, index=False, header=not wrote_header)
                    wrote_header = True
                    final_counts[BENIGN_LABEL] += len(benign_chunk)

            attack_labels_in_chunk = sorted(
                label for label in labels.unique() if int(label) in ATTACK_LABELS
            )
            for label in attack_labels_in_chunk:
                label = int(label)
                quota = attack_quotas[label]
                subset = chunk.loc[labels == label].copy()
                if subset.empty:
                    continue
                subset["_sample_key"] = rng.random_sample(len(subset))

                if label in reservoirs:
                    subset = pd.concat([reservoirs[label], subset], ignore_index=True)
                reservoirs[label] = subset.nsmallest(quota, "_sample_key")

            if chunk_number % 100 == 0:
                sampled_attack_rows = sum(len(frame) for frame in reservoirs.values())
                print(
                    f"chunks={chunk_number:,} benign_written={final_counts[BENIGN_LABEL]:,} "
                    f"attack_reservoir={sampled_attack_rows:,}"
                )

        attack_frames = []
    for label, quota in attack_quotas.items():
        sampled = reservoirs.get(label)
        if sampled is None or len(sampled) != quota:
            found = 0 if sampled is None else len(sampled)
            raise RuntimeError(f"Sampled {found} rows for label {label}; expected {quota}.")
        clean_sample = sampled.drop(columns="_sample_key")
        final_counts[label] = len(clean_sample)
        attack_frames.append(clean_sample)

    attack_df = _shuffle(pd.concat(attack_frames, ignore_index=True), rng)
    with _open_text_output(tmp_output_path, append=True) as out_f:
        attack_df.to_csv(out_f, index=False, header=False)

    if output_path.exists():
        output_path.unlink()
    tmp_output_path.replace(output_path)

    print_evaluation_from_counts(final_counts, counts, attack_quotas, output_path)
    write_split_manifest(
        output_path=output_path,
        input_path=input_path,
        source_counts=counts,
        final_counts=final_counts,
        attack_quotas=attack_quotas,
        attack_strategy=attack_strategy,
        random_state=random_state,
        chunksize=chunksize,
        count_source=resolved_count_source,
    )


def write_output(df: pd.DataFrame, output_path: Path) -> None:
    output_path.parent.mkdir(parents=True, exist_ok=True)
    df.to_csv(output_path, index=False)


def print_evaluation(
    df: pd.DataFrame,
    source_counts: Mapping[int, int],
    expected_attack_quotas: Mapping[int, int],
    output_path: Path,
) -> None:
    final_binary = df[BINARY_LABEL].astype("int64").value_counts().sort_index()
    final_multi = df[MULTICLASS_LABEL].astype("int64").value_counts().sort_index()

    benign_rows = int(final_binary.get(BENIGN_LABEL, 0))
    attack_rows = int(final_binary.get(1, 0))
    total_rows = len(df)
    benign_pct = benign_rows / total_rows * 100
    attack_pct = attack_rows / total_rows * 100

    print()
    print("WROTE REALISTIC VANET-IDS26 TEST SPLIT")
    print(f"Output: {output_path}")
    print()
    print("Source multiclass counts:")
    print(pd.Series(source_counts, name="source_rows").sort_index().to_string())
    print()
    print("Final binary distribution:")
    print(f"benign rows: {benign_rows:,} ({benign_pct:.2f}%)")
    print(f"attack rows: {attack_rows:,} ({attack_pct:.2f}%)")
    print(f"total rows:  {total_rows:,}")
    print(f"benign:attack ratio: {benign_rows}:{attack_rows} = {benign_rows / attack_rows:.0f}:1")
    print()
    print("Final attack-family distribution:")
    attack_distribution = pd.DataFrame(
        {
            "expected_rows": pd.Series(expected_attack_quotas),
            "actual_rows": final_multi.reindex(ATTACK_LABELS, fill_value=0),
        }
    )
    attack_distribution["attack_pct"] = (
        attack_distribution["actual_rows"] / attack_rows * 100
    ).round(4)
    print(attack_distribution.to_string())

    assert benign_rows == attack_rows * BENIGN_TO_ATTACK_RATIO
    assert attack_rows == sum(expected_attack_quotas.values())
    assert set(final_multi.loc[final_multi.index > 0].index) == set(ATTACK_LABELS)
    assert all(
        int(final_multi.get(label, 0)) == quota
        for label, quota in expected_attack_quotas.items()
    )

    print()
    print("Verification: PASS - strict 99% benign / 1% attack with all 26 attacks represented.")


def print_evaluation_from_counts(
    final_counts: Mapping[int, int],
    source_counts: Mapping[int, int],
    expected_attack_quotas: Mapping[int, int],
    output_path: Path,
) -> None:
    benign_rows = int(final_counts.get(BENIGN_LABEL, 0))
    attack_rows = sum(int(final_counts.get(label, 0)) for label in ATTACK_LABELS)
    total_rows = benign_rows + attack_rows
    benign_pct = benign_rows / total_rows * 100
    attack_pct = attack_rows / total_rows * 100

    print()
    print("WROTE REALISTIC VANET-IDS26 TEST SPLIT")
    print(f"Output: {output_path}")
    print()
    print("Source multiclass counts:")
    print(pd.Series(source_counts, name="source_rows").sort_index().to_string())
    print()
    print("Final binary distribution:")
    print(f"benign rows: {benign_rows:,} ({benign_pct:.2f}%)")
    print(f"attack rows: {attack_rows:,} ({attack_pct:.2f}%)")
    print(f"total rows:  {total_rows:,}")
    print(f"benign:attack ratio: {benign_rows}:{attack_rows} = {benign_rows / attack_rows:.0f}:1")
    print()
    print("Final attack-family distribution:")
    attack_distribution = pd.DataFrame(
        {
            "attack_type": pd.Series(ATTACK_TYPES),
            "expected_rows": pd.Series(expected_attack_quotas),
            "actual_rows": pd.Series(
                {label: int(final_counts.get(label, 0)) for label in ATTACK_LABELS}
            ),
        }
    )
    attack_distribution["attack_pct"] = (
        attack_distribution["actual_rows"] / attack_rows * 100
    ).round(4)
    print(attack_distribution.to_string())

    assert benign_rows == attack_rows * BENIGN_TO_ATTACK_RATIO
    assert attack_rows == sum(expected_attack_quotas.values())
    assert set(label for label in ATTACK_LABELS if final_counts.get(label, 0) > 0) == set(ATTACK_LABELS)
    assert all(
        int(final_counts.get(label, 0)) == quota
        for label, quota in expected_attack_quotas.items()
    )

    print()
    print("Verification: PASS - strict 99% benign / 1% attack with all 26 attacks represented.")


def write_split_manifest(
    output_path: Path,
    input_path: Path,
    source_counts: Mapping[int, int],
    final_counts: Mapping[int, int],
    attack_quotas: Mapping[int, int],
    attack_strategy: str,
    random_state: int,
    chunksize: int,
    count_source: str,
) -> None:
    manifest_path = output_path.with_name(output_path.name + ".manifest.json")
    attack_distribution_path = output_path.with_name(
        output_path.name + ".attack_distribution.csv"
    )

    benign_rows = int(final_counts[BENIGN_LABEL])
    attack_rows = sum(int(final_counts[label]) for label in ATTACK_LABELS)
    total_rows = benign_rows + attack_rows

    attack_distribution = pd.DataFrame(
        [
            {
                "multiclass_label": label,
                "attack_type": ATTACK_TYPES[label],
                "source_rows": int(source_counts[label]),
                "sampled_rows": int(final_counts[label]),
                "attack_pct": int(final_counts[label]) / attack_rows * 100,
            }
            for label in ATTACK_LABELS
        ]
    )
    attack_distribution.to_csv(attack_distribution_path, index=False)

    manifest = {
        "dataset_name": "VANET-IDS26 realistic 99:1 operational test split",
        "created_at_utc": datetime.now(timezone.utc).isoformat(),
        "input_path": str(input_path),
        "output_path": str(output_path),
        "attack_distribution_path": str(attack_distribution_path),
        "output_format": "csv.gz" if output_path.name.endswith(".csv.gz") else output_path.suffix.lstrip("."),
        "output_file_bytes": output_path.stat().st_size if output_path.exists() else None,
        "count_source": count_source,
        "random_state": random_state,
        "chunksize": chunksize,
        "benign_to_attack_ratio": BENIGN_TO_ATTACK_RATIO,
        "attack_strategy": attack_strategy,
        "source_counts": {str(label): int(count) for label, count in source_counts.items()},
        "attack_quotas": {str(label): int(count) for label, count in attack_quotas.items()},
        "final_counts": {str(label): int(count) for label, count in final_counts.items()},
        "final_binary_counts": {
            "benign": benign_rows,
            "attack": attack_rows,
            "total": total_rows,
        },
        "final_percentages": {
            "benign": benign_rows / total_rows * 100,
            "attack": attack_rows / total_rows * 100,
        },
        "notes": [
            "This split is designed for deployed NIDS evaluation where benign traffic dominates.",
            "The benign portion uses the maximum strict 99:1 size and drops only surplus benign rows that do not fit the exact ratio.",
            "Attack rows are under-sampled with stratification across multiclass_label 1..26.",
            "Rows are written as benign stream followed by sampled attacks; shuffle downstream before order-sensitive evaluation.",
        ],
    }
    manifest_path.write_text(json.dumps(manifest, indent=2), encoding="utf-8")
    print(f"Manifest: {manifest_path}")
    print(f"Attack distribution CSV: {attack_distribution_path}")


def _open_text_output(path: Path, append: bool = False):
    mode = "at" if append else "wt"
    if ".gz" in path.suffixes:
        return gzip.open(path, mode, newline="", encoding="utf-8", compresslevel=1)
    return path.open(mode, newline="", encoding="utf-8")


def _shuffle(df: pd.DataFrame, rng) -> pd.DataFrame:
    return df.sample(
        frac=1.0,
        random_state=int(rng.randint(0, 2**31 - 1)),
    ).reset_index(drop=True)


def _validate_source_counts(label_counts: Mapping[int, int]) -> None:
    missing = [label for label in (BENIGN_LABEL, *ATTACK_LABELS) if label_counts.get(label, 0) <= 0]
    if missing:
        raise ValueError(
            "Input does not contain all required labels. Missing/empty multiclass labels: "
            + ", ".join(str(label) for label in missing)
        )


def _require_label_columns(columns: Iterable[str]) -> None:
    columns = set(columns)
    missing = [col for col in (BINARY_LABEL, MULTICLASS_LABEL) if col not in columns]
    if missing:
        raise ValueError(f"Input is missing required label column(s): {missing}")


def main() -> None:
    args = parse_args()
    input_path = DEFAULT_SAMPLE if args.sample else args.input
    mode = "memory" if args.sample else args.mode

    if not input_path.exists():
        raise FileNotFoundError(f"Input file not found: {input_path}")

    attack_rows = None if args.use_largest_feasible else args.attack_rows

    if mode == "memory":
        build_split_in_memory(
            input_path=input_path,
            output_path=args.output,
            attack_rows=attack_rows,
            attack_strategy=args.attack_strategy,
            random_state=args.random_state,
        )
    else:
        build_split_chunked(
            input_path=input_path,
            output_path=args.output,
            chunksize=args.chunksize,
            attack_rows=attack_rows,
            attack_strategy=args.attack_strategy,
            random_state=args.random_state,
            count_source=args.count_source,
            manifest_dir=args.manifest_dir,
        )


if __name__ == "__main__":
    main()