Spaces:
Sleeping
Sleeping
File size: 23,739 Bytes
23f9faf 9765f33 23f9faf 9765f33 23f9faf | 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 |
"""Unit tests for the multi-round balanced assignment + completion-code
feature.
Runs offline (no network / no huggingface_hub calls): assignment.py's core
functions are pure, and the round-lifecycle integration tests patch main.py's
HF-facing I/O functions with an in-memory fake store.
Run with: python -m unittest backend.test_assignment (from repo root)
or: python -m unittest test_assignment (from backend/)
"""
import json
import os
import random
import sys
import unittest
from datetime import datetime, timedelta, timezone
from unittest import mock
sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
import assignment # noqa: E402
def make_record(user, video_id, created_at, extra=None):
record = {
"id": "x",
"video_id": video_id,
"user": user,
"created_at": created_at,
"annotations": {"responses": {}},
}
if extra:
record.update(extra)
return record
def iso(dt):
return dt.isoformat().replace("+00:00", "Z")
class DedupeLatestTests(unittest.TestCase):
def test_existing_annotations_preserved_and_counted(self):
records = [
make_record("Adi", "v1", "2026-01-01T00:00:00Z"),
make_record("Adi", "v2", "2026-01-01T00:00:00Z"),
make_record("youngsun", "v1", "2026-01-01T00:00:00Z"),
]
original_len = len(records)
deduped, skipped = assignment.dedupe_latest(records)
self.assertEqual(len(records), original_len, "dedupe_latest must not mutate its input")
self.assertEqual(skipped, 0)
self.assertEqual(len(deduped), 3)
by_video = assignment.completed_by_video(deduped)
self.assertEqual(by_video["v1"], {"adi", "youngsun"})
self.assertEqual(by_video["v2"], {"adi"})
def test_same_annotator_same_video_counted_once_latest_kept(self):
records = [
make_record("Adi", "v1", "2026-01-01T00:00:00Z", {"annotations": {"marker": "old"}}),
make_record("Adi", "v1", "2026-01-02T00:00:00Z", {"annotations": {"marker": "new"}}),
make_record("adi", "v1", "2026-01-03T00:00:00Z", {"annotations": {"marker": "newest"}}),
]
deduped, skipped = assignment.dedupe_latest(records)
self.assertEqual(skipped, 0)
self.assertEqual(len(deduped), 1, "same annotator + same video, across case variants, is one record")
self.assertEqual(deduped[("adi", "v1")]["annotations"]["marker"], "newest")
def test_malformed_or_partial_records_skipped_not_counted(self):
records = [
make_record("Adi", "v1", "2026-01-01T00:00:00Z"),
{"id": "y", "video_id": "v2", "user": "Adi"}, # missing created_at (autosave/reserve-like)
{"id": "z", "user": "Adi", "created_at": "2026-01-01T00:00:00Z"}, # missing video_id
{"id": "w", "video_id": "v3", "user": "", "created_at": "2026-01-01T00:00:00Z"}, # empty user
{"id": "q", "video_id": "v4", "user": "Adi", "created_at": "not-a-date"}, # unparsable timestamp
]
deduped, skipped = assignment.dedupe_latest(records)
self.assertEqual(len(deduped), 1)
self.assertEqual(skipped, 4)
class SavedResponsesForRoundTests(unittest.TestCase):
def test_returns_responses_only_for_this_annotator_and_this_rounds_videos(self):
records = [
make_record("Adi", "v1", "2026-01-01T00:00:00Z", {"annotations": {"responses": {"marker": "adi-v1"}}}),
make_record("Adi", "v2", "2026-01-01T00:00:00Z", {"annotations": {"responses": {"marker": "adi-v2"}}}),
make_record(
"youngsun", "v1", "2026-01-01T00:00:00Z", {"annotations": {"responses": {"marker": "youngsun-v1"}}}
),
]
deduped, _skipped = assignment.dedupe_latest(records)
result = assignment.saved_responses_for_round(deduped, "adi", ["v1", "v2", "v3"])
self.assertEqual(result, {"v1": {"marker": "adi-v1"}, "v2": {"marker": "adi-v2"}})
def test_excludes_videos_outside_the_given_round(self):
records = [
make_record("Adi", "v9", "2026-01-01T00:00:00Z", {"annotations": {"responses": {"marker": "old-round"}}})
]
deduped, _skipped = assignment.dedupe_latest(records)
result = assignment.saved_responses_for_round(deduped, "adi", ["v1", "v2"])
self.assertEqual(result, {}, "a completion from a prior round must not leak into this round's resume state")
def test_empty_deduped_returns_empty_dict(self):
self.assertEqual(assignment.saved_responses_for_round({}, "adi", ["v1"]), {})
class PickBalancedTests(unittest.TestCase):
def test_prefers_lowest_coverage(self):
coverage = {"v1": 4, "v2": 0, "v3": 2}
picked = assignment.pick_balanced(list(coverage), coverage.get, 2, random.Random(0), target=5)
self.assertEqual(set(picked), {"v2", "v3"})
def test_excludes_at_or_over_target_when_enough_under_target_remain(self):
coverage = {"v1": 5, "v2": 5, "v3": 1}
picked = assignment.pick_balanced(list(coverage), coverage.get, 1, random.Random(0), target=5)
self.assertEqual(picked, ["v3"])
def test_falls_back_to_at_target_videos_when_not_enough_under_target(self):
coverage = {"v1": 5, "v2": 5, "v3": 1}
picked = assignment.pick_balanced(list(coverage), coverage.get, 3, random.Random(0), target=5)
self.assertEqual(set(picked), {"v1", "v2", "v3"})
def test_never_returns_more_than_requested_or_duplicates(self):
coverage = {f"v{i}": i % 3 for i in range(30)}
picked = assignment.pick_balanced(list(coverage), coverage.get, 7, random.Random(1), target=5)
self.assertEqual(len(picked), 7)
self.assertEqual(len(set(picked)), 7)
class CompletionCodeTests(unittest.TestCase):
def test_format(self):
code = assignment.generate_completion_code()
self.assertTrue(code.startswith("T2AV-"))
suffix = code[len("T2AV-"):]
self.assertEqual(len(suffix), 12)
self.assertTrue(all(c in assignment.COMPLETION_CODE_ALPHABET for c in suffix))
def test_codes_are_not_trivially_repeated(self):
codes = {assignment.generate_completion_code() for _ in range(200)}
self.assertEqual(len(codes), 200, "200 draws from a 36^12 space should never collide")
class BuildRoundTests(unittest.TestCase):
def test_always_excludes_already_completed_never_seeds_with_them(self):
"""Core policy change: a round is exactly N NEW videos, never padded
with history, regardless of how much history exists."""
catalog = [f"v{i}" for i in range(1, 201)] # 200 videos, like production
already_completed = set(catalog[:51]) # mirrors real production data (Adi: 51 completed)
round_record = assignment.build_round(
"adi", "Adi", catalog, already_completed, {}, {}, random.Random(0), round_number=1,
)
self.assertEqual(round_record["actual_size"], 20)
self.assertEqual(len(round_record["video_ids"]), 20)
self.assertEqual(
already_completed.intersection(round_record["video_ids"]), set(),
"none of the 51 previously completed videos may appear in the new round",
)
self.assertEqual(len(set(round_record["video_ids"])), 20, "no duplicates within the round")
self.assertEqual(round_record["round_number"], 1)
self.assertEqual(round_record["requested_size"], 20)
def test_one_historical_completion_gets_twenty_new_not_nineteen(self):
catalog = [f"v{i}" for i in range(1, 201)]
already_completed = {"v1"}
round_record = assignment.build_round(
"youngsun", "youngsun_0715", catalog, already_completed, {}, {}, random.Random(0), round_number=1,
)
self.assertEqual(round_record["actual_size"], 20)
self.assertNotIn("v1", round_record["video_ids"])
def test_brand_new_annotator_gets_exactly_twenty(self):
catalog = [f"v{i}" for i in range(1, 201)]
round_record = assignment.build_round(
"new", "New", catalog, set(), {}, {}, random.Random(0), round_number=1,
)
self.assertEqual(round_record["actual_size"], 20)
def test_reduced_size_when_not_enough_eligible_remain(self):
catalog = [f"v{i}" for i in range(1, 26)] # only 25 videos total
already_completed = set(catalog[:15]) # 15 done, only 10 eligible left
round_record = assignment.build_round(
"x", "X", catalog, already_completed, {}, {}, random.Random(0), round_number=1,
)
self.assertEqual(round_record["actual_size"], 10)
self.assertEqual(round_record["requested_size"], 20)
self.assertEqual(len(round_record["video_ids"]), 10)
self.assertEqual(already_completed.intersection(round_record["video_ids"]), set())
def test_second_round_excludes_first_rounds_completions_too(self):
"""After Adi finishes 51 (history) + 20 (round 1) = 71, round 2 must
exclude all 71, selecting from the remaining 129."""
catalog = [f"v{i}" for i in range(1, 201)]
historical = set(catalog[:51])
round1 = assignment.build_round(
"adi", "Adi", catalog, historical, {}, {}, random.Random(0), round_number=1,
)
after_round1 = historical | set(round1["video_ids"])
self.assertEqual(len(after_round1), 71)
round2 = assignment.build_round(
"adi", "Adi", catalog, after_round1, {}, {}, random.Random(1), round_number=2,
)
self.assertEqual(round2["actual_size"], 20)
self.assertEqual(after_round1.intersection(round2["video_ids"]), set())
remaining_pool_size = len(catalog) - len(after_round1)
self.assertEqual(remaining_pool_size, 129)
self.assertEqual(round2["round_number"], 2)
def test_existing_completions_affect_new_round_priority(self):
catalog = ["v1", "v2", "v3"]
completed_by_video_map = {"v1": {"p1", "p2", "p3", "p4"}} # coverage 4, still < target 5
round_record = assignment.build_round(
"new", "New", catalog, set(), completed_by_video_map, {}, random.Random(0),
round_number=1, videos_per_annotator=2,
)
self.assertNotIn("v1", round_record["video_ids"], "v2/v3 have lower coverage and must be preferred")
self.assertEqual(set(round_record["video_ids"]), {"v2", "v3"})
class ReservedByVideoTests(unittest.TestCase):
def test_counts_open_uncompleted_unexpired_reservations(self):
now = datetime(2026, 1, 10, tzinfo=timezone.utc)
assignments = [
{"annotator_id": "a", "video_ids": ["v1", "v2"], "created_at": "2026-01-09T00:00:00Z"},
{"annotator_id": "b", "video_ids": ["v1"], "created_at": "2026-01-09T00:00:00Z"},
]
completed_by_annotator = {"a": {"v2"}}
reserved = assignment.reserved_by_video(assignments, completed_by_annotator, now, ttl_seconds=7 * 24 * 3600)
self.assertEqual(reserved.get("v1"), 2)
self.assertNotIn("v2", reserved)
def test_excludes_expired_reservations_without_needing_to_delete_anything(self):
now = datetime(2026, 1, 10, tzinfo=timezone.utc)
assignments = [{"annotator_id": "a", "video_ids": ["v1"], "created_at": "2025-01-01T00:00:00Z"}]
reserved = assignment.reserved_by_video(assignments, {}, now, ttl_seconds=24 * 3600)
self.assertEqual(reserved.get("v1", 0), 0)
def test_excludes_the_requesting_annotator_from_their_own_reservations(self):
now = datetime(2026, 1, 10, tzinfo=timezone.utc)
assignments = [{"annotator_id": "a", "video_ids": ["v1"], "created_at": "2026-01-09T00:00:00Z"}]
reserved = assignment.reserved_by_video(assignments, {}, now, ttl_seconds=24 * 3600, exclude_annotator="a")
self.assertEqual(reserved.get("v1", 0), 0)
class SequentialBalanceTests(unittest.TestCase):
def test_sequential_rounds_stay_reasonably_balanced(self):
catalog = [f"v{i}" for i in range(1, 81)] # 80 videos
reserved_by_video_map = {}
rng = random.Random(42)
for i in range(10): # 10 annotators * 20 slots = 200 slots over 80 videos
round_record = assignment.build_round(
f"person{i}", f"person{i}", catalog, set(), {}, reserved_by_video_map, rng, round_number=1,
)
for video_id in round_record["video_ids"]:
reserved_by_video_map[video_id] = reserved_by_video_map.get(video_id, 0) + 1
counts = list(reserved_by_video_map.values())
self.assertEqual(len(reserved_by_video_map), 80, "every video should have been touched")
self.assertLessEqual(max(counts) - min(counts), 2, "balancing should keep coverage tight")
class FakeRepoStore:
"""Stands in for the annotations dataset repo's assignments/ and
completions/ prefixes."""
def __init__(self):
self.files = {}
def read(self, path):
return self.files.get(path)
def write_kwargs(self, **kwargs):
path = kwargs["path_in_repo"]
content = json.loads(kwargs["path_or_fileobj"].getvalue().decode("utf-8"))
self.files[path] = content
def list_active_round_records(self):
records = []
for path, pointer in self.files.items():
if not path.endswith("/current.json"):
continue
annotator_id = path.split("/")[1]
round_path = f"assignments/{annotator_id}/rounds/{pointer['assignment_id']}.json"
completion_path = f"completions/{annotator_id}/{pointer['assignment_id']}.json"
round_record = self.files.get(round_path)
if round_record is None or completion_path in self.files:
continue
records.append(round_record)
return records
class RoundLifecycleIntegrationTests(unittest.TestCase):
"""Exercises main.get_or_create_current_round() with every HF network
call replaced by an in-memory fake, so this stays fully offline."""
def setUp(self):
import main as backend_main
self.backend_main = backend_main
self.store = FakeRepoStore()
self.catalog = [f"v{i}" for i in range(1, 201)] # 200 videos, like production
self.completed_records = [] # raw annotation records fed to both list_annotation_records variants
patches = [
mock.patch.object(backend_main, "_read_json_from_repo", side_effect=self.store.read),
# Both the cached and fresh variants must be mocked - the completion
# check in get_or_create_current_round() deliberately bypasses the
# cache (fresh=True) so a real annotator sees their code promptly;
# leaving list_annotation_records() unmocked would silently fall
# through to a real network call here.
mock.patch.object(
backend_main, "list_annotation_records_cached", side_effect=lambda: self.completed_records
),
mock.patch.object(
backend_main, "list_annotation_records", side_effect=lambda: self.completed_records
),
mock.patch.object(
backend_main, "list_active_round_records", side_effect=self.store.list_active_round_records
),
mock.patch.object(backend_main, "ensure_dataset_exists", return_value=None),
mock.patch.object(backend_main, "ensure_dataset_configured", return_value=None),
mock.patch.object(backend_main, "_current_catalog_video_ids", return_value=self.catalog),
mock.patch.object(backend_main, "upload_with_retry", side_effect=self.store.write_kwargs),
]
for patcher in patches:
patcher.start()
self.addCleanup(patcher.stop)
def _complete_round(self, annotator_raw, round_record):
"""Simulates the annotator finishing every video in a round by
appending matching annotation records - exactly what save_annotation
would have produced."""
now = datetime.now(timezone.utc)
for video_id in round_record["video_ids"]:
self.completed_records.append(make_record(annotator_raw, video_id, iso(now)))
def test_brand_new_annotator_gets_round_one_of_twenty(self):
result = self.backend_main.get_or_create_current_round("Scratch Tester")
self.assertEqual(result["status"], "in_progress")
self.assertEqual(result["round_number"], 1)
self.assertEqual(len(result["video_ids"]), 20)
self.assertEqual(len(set(result["video_ids"])), 20)
def test_refresh_while_incomplete_returns_identical_round(self):
first = self.backend_main.get_or_create_current_round("Scratch Tester")
second = self.backend_main.get_or_create_current_round("Scratch Tester")
self.assertEqual(first["video_ids"], second["video_ids"])
self.assertEqual(first["assignment_id"], second["assignment_id"])
def test_no_completion_code_at_nineteen_of_twenty(self):
first = self.backend_main.get_or_create_current_round("Scratch Tester")
for video_id in first["video_ids"][:19]:
self.completed_records.append(make_record("Scratch Tester", video_id, iso(datetime.now(timezone.utc))))
result = self.backend_main.get_or_create_current_round("Scratch Tester")
self.assertEqual(result["status"], "in_progress")
self.assertNotIn("completion_code", result)
def test_brand_new_round_has_no_saved_annotations(self):
result = self.backend_main.get_or_create_current_round("Scratch Tester")
self.assertEqual(result["saved_annotations"], {})
def test_resume_returns_saved_responses_for_completed_videos_only(self):
first = self.backend_main.get_or_create_current_round("Scratch Tester")
completed_ids = first["video_ids"][:3]
for video_id in completed_ids:
self.completed_records.append(
make_record(
"Scratch Tester",
video_id,
iso(datetime.now(timezone.utc)),
{"annotations": {"responses": {"video": {"tcRel": {"label": "PASS", "rationale": ""}}}}},
)
)
result = self.backend_main.get_or_create_current_round("Scratch Tester")
self.assertEqual(result["status"], "in_progress")
self.assertEqual(set(result["saved_annotations"].keys()), set(completed_ids))
for video_id in completed_ids:
self.assertEqual(result["saved_annotations"][video_id]["video"]["tcRel"]["label"], "PASS")
def test_completion_code_persisted_once_at_twenty_of_twenty(self):
first = self.backend_main.get_or_create_current_round("Scratch Tester")
self._complete_round("Scratch Tester", first)
result_a = self.backend_main.get_or_create_current_round("Scratch Tester")
result_b = self.backend_main.get_or_create_current_round("Scratch Tester")
self.assertEqual(result_a["status"], "completed")
self.assertTrue(result_a["completion_code"].startswith("T2AV-"))
self.assertEqual(result_a["completion_code"], result_b["completion_code"], "code must be stable on refresh")
completion_files = [p for p in self.store.files if p.startswith("completions/")]
self.assertEqual(len(completion_files), 1, "completion record written exactly once")
def test_within_grace_period_shows_same_completed_round_not_a_new_one(self):
first = self.backend_main.get_or_create_current_round("Scratch Tester")
self._complete_round("Scratch Tester", first)
self.backend_main.get_or_create_current_round("Scratch Tester") # triggers completion-record creation
# Directly age the completion record to just inside the grace window.
completion_path = f"completions/scratch-tester/{first['assignment_id']}.json"
recent = datetime.now(timezone.utc) - timedelta(
seconds=assignment.COMPLETION_GRACE_PERIOD_SECONDS - 30
)
self.store.files[completion_path]["completed_at"] = iso(recent)
result = self.backend_main.get_or_create_current_round("Scratch Tester")
self.assertEqual(result["status"], "completed")
self.assertEqual(result["round_number"], 1)
def test_past_grace_period_auto_creates_next_round_with_no_extra_call(self):
first = self.backend_main.get_or_create_current_round("Scratch Tester")
self._complete_round("Scratch Tester", first)
self.backend_main.get_or_create_current_round("Scratch Tester") # triggers completion-record creation
# Directly age the completion record past the grace window - simulates "came back later".
completion_path = f"completions/scratch-tester/{first['assignment_id']}.json"
stale = datetime.now(timezone.utc) - timedelta(
seconds=assignment.COMPLETION_GRACE_PERIOD_SECONDS + 60
)
self.store.files[completion_path]["completed_at"] = iso(stale)
result = self.backend_main.get_or_create_current_round("Scratch Tester")
self.assertEqual(result["status"], "in_progress")
self.assertEqual(result["round_number"], 2)
self.assertEqual(set(result["video_ids"]).intersection(first["video_ids"]), set())
def test_never_assigns_the_same_video_twice_across_rounds(self):
first = self.backend_main.get_or_create_current_round("Scratch Tester")
self._complete_round("Scratch Tester", first)
self.backend_main.get_or_create_current_round("Scratch Tester")
completion_path = f"completions/scratch-tester/{first['assignment_id']}.json"
stale = datetime.now(timezone.utc) - timedelta(seconds=assignment.COMPLETION_GRACE_PERIOD_SECONDS + 60)
self.store.files[completion_path]["completed_at"] = iso(stale)
second = self.backend_main.get_or_create_current_round("Scratch Tester")
all_seen = first["video_ids"] + second["video_ids"]
self.assertEqual(len(all_seen), len(set(all_seen)))
def test_different_annotators_each_get_their_own_round(self):
a = self.backend_main.get_or_create_current_round("Person A")
b = self.backend_main.get_or_create_current_round("Person B")
self.assertEqual(len(a["video_ids"]), 20)
self.assertEqual(len(b["video_ids"]), 20)
round_files = [p for p in self.store.files if "/rounds/" in p]
self.assertEqual(len(round_files), 2)
def test_simultaneous_requests_do_not_create_duplicate_rounds(self):
"""The lock fully serializes get_or_create_current_round, so we
simulate "simultaneous" by calling it back-to-back and asserting only
one round/pointer combination was ever written - the real concurrency
guarantee is exercised live against a scratch HF repo separately."""
for _ in range(5):
self.backend_main.get_or_create_current_round("Scratch Tester")
round_files = [p for p in self.store.files if "/rounds/" in p]
self.assertEqual(len(round_files), 1)
def test_partial_records_never_mistakenly_counted_as_completed(self):
first = self.backend_main.get_or_create_current_round("Scratch Tester")
# An autosave/reservation-like record missing created_at for every video.
for video_id in first["video_ids"]:
self.completed_records.append({"id": "x", "video_id": video_id, "user": "Scratch Tester"})
result = self.backend_main.get_or_create_current_round("Scratch Tester")
self.assertEqual(result["status"], "in_progress")
if __name__ == "__main__":
unittest.main()
|