Skip to content

Artifacts And Results

Training, distillation, import, and evaluation runs write tracked outputs under results/. Each run stores artifacts, configs, logs, and performance files.

Result Layout

results/
  teacher/
  student/
  recdistill/

Each run follows this structure:

results/<kind>/<run>/
  artifacts/
  config/
  logs/
  perf/

Run names contain the framework, model, dataset, timestamp, and experiment id. The current layout no longer creates nested result directories for individual backbones such as bprmf/, lgcn/, or nmf/; model names are encoded in the run and artifact filenames instead.

Artifact Types

.teacher            trained or imported teacher
.student            plain student baseline
.distilled_student  distilled student model

Checkpointing

STUDENT_CHECKPOINT_FORMAT = 'recdistill.student.v1' module-attribute

config_hash(config: dict[str, Any] | None) -> str | None

Source code in recdistill/checkpointing.py 
17
18
19
20
21
def config_hash(config: dict[str, Any] | None) -> str | None:
    if config is None:
        return None
    encoded = json.dumps(config, sort_keys=True, default=str, separators=(",", ":")).encode("utf-8")
    return hashlib.sha256(encoded).hexdigest()

current_git_commit(repo_root: Path | None = None) -> str | None

Source code in recdistill/checkpointing.py 
24
25
26
27
28
29
30
31
32
33
34
35
36
37
def current_git_commit(repo_root: Path | None = None) -> str | None:
    repo_root = repo_root or Path(__file__).resolve().parents[1]
    try:
        result = subprocess.run(
            ["git", "rev-parse", "HEAD"],
            cwd=repo_root,
            check=True,
            capture_output=True,
            text=True,
        )
    except Exception:
        return None
    commit = result.stdout.strip()
    return commit or None

enrich_student_checkpoint_payload(payload: dict[str, Any]) -> dict[str, Any]

Source code in recdistill/checkpointing.py 
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
def enrich_student_checkpoint_payload(payload: dict[str, Any]) -> dict[str, Any]:
    config = payload.get("config") if isinstance(payload.get("config"), dict) else {}
    enriched = dict(payload)
    enriched.setdefault("format_version", STUDENT_CHECKPOINT_FORMAT)
    enriched.setdefault("created_at_utc", utc_now_iso())
    enriched.setdefault("config_hash", config_hash(config))
    enriched.setdefault("git_commit", current_git_commit())
    enriched.setdefault("dataset", config.get("dataset"))
    enriched.setdefault("teacher", config.get("teacher_model"))
    enriched.setdefault("student", config.get("student_backbone"))
    enriched.setdefault("distiller", _distiller_from_config(config))
    metadata = dict(enriched.get("metadata") or {})
    metadata.setdefault("format_version", enriched["format_version"])
    metadata.setdefault("created_at_utc", enriched["created_at_utc"])
    metadata.setdefault("config_hash", enriched["config_hash"])
    metadata.setdefault("git_commit", enriched["git_commit"])
    metadata.setdefault("dataset", enriched.get("dataset"))
    metadata.setdefault("teacher", enriched.get("teacher"))
    metadata.setdefault("student", enriched.get("student"))
    metadata.setdefault("distiller", enriched.get("distiller"))
    enriched["metadata"] = metadata
    return enriched

save_student_checkpoint(path: str | Path, payload: dict[str, Any]) -> dict[str, Any]

Source code in recdistill/checkpointing.py 
64
65
66
67
68
69
def save_student_checkpoint(path: str | Path, payload: dict[str, Any]) -> dict[str, Any]:
    checkpoint_path = Path(path)
    checkpoint_path.parent.mkdir(parents=True, exist_ok=True)
    enriched = enrich_student_checkpoint_payload(payload)
    torch.save(enriched, checkpoint_path)
    return enriched

load_student_checkpoint(path: str | Path, map_location: str | torch.device = 'cpu') -> dict[str, Any]

Source code in recdistill/checkpointing.py 
72
73
74
75
76
77
78
def load_student_checkpoint(path: str | Path, map_location: str | torch.device = "cpu") -> dict[str, Any]:
    payload = torch.load(Path(path), map_location=map_location, weights_only=False)
    if not isinstance(payload, dict):
        raise TypeError(f"Unsupported checkpoint payload type: {type(payload)!r}")
    if "student_state_dict" not in payload:
        raise KeyError(f"`student_state_dict` missing in checkpoint: {path}")
    return payload

Runtime Paths

recdistill.paths resolves datasets, run directories, artifacts, histories, and performance files for the flat run-based result layout.

Helper Purpose
experiment_run_dir Builds results/<kind>/<run>/.
experiment_artifact_path Builds paths under a run's artifacts/ directory.
teacher_artifact_path Resolves a native teacher .teacher artifact.
student_artifact_path Resolves a plain student .student artifact.
distilled_student_artifact_path Resolves a .distilled_student artifact.
resolve_teacher_checkpoint Finds an explicit or latest matching teacher checkpoint.
resolve_student_checkpoint Resolves plain or distilled student checkpoints.

REPO_ROOT = Path(__file__).resolve().parents[1] module-attribute

PROJECT_PATH = str(REPO_ROOT) module-attribute

DATA_ROOT = REPO_ROOT / 'data' module-attribute

RESULTS_ROOT = REPO_ROOT / 'results' module-attribute

CONFIG_ROOT = REPO_ROOT / 'config' module-attribute

EXPERIMENTS_CONFIG_ROOT = CONFIG_ROOT / 'experiments' module-attribute

PRESETS_ROOT = EXPERIMENTS_CONFIG_ROOT module-attribute

TEACHER_EXT = '.teacher' module-attribute

STUDENT_EXT = '.student' module-attribute

DISTILLED_STUDENT_EXT = '.distilled_student' module-attribute

CITEULIKE = 'citeulike' module-attribute

BOOKCROSSING = 'bookcrossing' module-attribute

AMAZONDM = 'amazon_dm' module-attribute

AMAZONCD = 'amazon_cd' module-attribute

_DATASET_FILENAME_BY_TYPE = {'raw': Path('data') / 'dataset.tsv', 'processed': Path('dataset.tsv'), 'train': Path('train.tsv'), 'val': Path('val.tsv'), 'test': Path('test.tsv')} module-attribute

PathManager

Compatibility namespace for common runtime path labels.

The current result layout no longer groups artifacts by backbone-specific nested directories. Artifact paths are resolved through run directories under results/<kind>/<run>/artifacts/, while supported model names are maintained by recdistill.registry.

Source code in recdistill/paths.py 
43
44
45
46
47
48
49
50
51
52
53
54
55
class PathManager:
    """Compatibility namespace for common runtime path labels.

    The current result layout no longer groups artifacts by backbone-specific
    nested directories. Artifact paths are resolved through run directories
    under `results/<kind>/<run>/artifacts/`, while supported model names are
    maintained by `recdistill.registry`.
    """

    DISTILLERS = ["de", "rrd", "unkd", "htd", "ftd", "de_rrd", "de_unkd", "rrd_unkd", "de_rrd_unkd"]
    BACKBONES = ["bprmf", "lgcn", "nmf"]
    SUPPORTED_MODELS = sorted(_SUPPORTED_BACKBONES)
    DATASETS = ["amazon_cd", "bookcrossing", "citeulike"]

DISTILLERS = ['de', 'rrd', 'unkd', 'htd', 'ftd', 'de_rrd', 'de_unkd', 'rrd_unkd', 'de_rrd_unkd'] class-attribute instance-attribute

SUPPORTED_MODELS = sorted(_SUPPORTED_BACKBONES) class-attribute instance-attribute

DATASETS = ['amazon_cd', 'bookcrossing', 'citeulike'] class-attribute instance-attribute

_create_directory(path: Path) -> None

Source code in recdistill/paths.py 
58
59
def _create_directory(path: Path) -> None:
    path.mkdir(parents=True, exist_ok=True)

new_experiment_id() -> str

Source code in recdistill/paths.py 
62
63
def new_experiment_id() -> str:
    return uuid.uuid4().hex[:8]

timestamp_slug() -> str

Source code in recdistill/paths.py 
66
67
def timestamp_slug() -> str:
    return datetime.now().strftime("%Y%m%d_%H%M%S")

experiment_kind_slug(kind: str) -> str

Source code in recdistill/paths.py 
70
71
72
73
74
75
76
77
78
def experiment_kind_slug(kind: str) -> str:
    kind_slug = str(kind).strip().lower()
    if kind_slug in {"teachers", "train_teacher"}:
        return "teacher"
    if kind_slug in {"students", "train_student"}:
        return "student"
    if kind_slug in {"distill_student", "distillation"}:
        return "recdistill"
    return kind_slug

experiment_run_name(*, kind: str, experiment_id: str, framework: str | None = None, model: str | None = None, dataset: str | None = None, timestamp: str | None = None) -> str

Source code in recdistill/paths.py 
81
82
83
84
85
86
87
88
89
90
91
92
93
def experiment_run_name(
    *,
    kind: str,
    experiment_id: str,
    framework: str | None = None,
    model: str | None = None,
    dataset: str | None = None,
    timestamp: str | None = None,
) -> str:
    stamp = timestamp or timestamp_slug()
    if framework and model and dataset:
        return f"{stamp}_{_path_slug(framework)}_{_path_slug(model)}_{_path_slug(dataset)}_{experiment_id}"
    return f"{stamp}_{experiment_kind_slug(kind)}_{experiment_id}"

experiment_run_dir(kind: str, experiment_id: str, *, framework: str | None = None, model: str | None = None, dataset: str | None = None, timestamp: str | None = None) -> Path

Source code in recdistill/paths.py 
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
def experiment_run_dir(
    kind: str,
    experiment_id: str,
    *,
    framework: str | None = None,
    model: str | None = None,
    dataset: str | None = None,
    timestamp: str | None = None,
) -> Path:
    kind_slug = experiment_kind_slug(kind)
    return RESULTS_ROOT / kind_slug / experiment_run_name(
        kind=kind_slug,
        experiment_id=experiment_id,
        framework=framework,
        model=model,
        dataset=dataset,
        timestamp=timestamp,
    )

experiment_artifact_path(*, kind: str, experiment_id: str, filename: str, framework: str | None = None, model: str | None = None, dataset: str | None = None, timestamp: str | None = None) -> Path

Source code in recdistill/paths.py 
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
def experiment_artifact_path(
    *,
    kind: str,
    experiment_id: str,
    filename: str,
    framework: str | None = None,
    model: str | None = None,
    dataset: str | None = None,
    timestamp: str | None = None,
) -> Path:
    return experiment_run_dir(
        kind,
        experiment_id,
        framework=framework,
        model=model,
        dataset=dataset,
        timestamp=timestamp,
    ) / "artifacts" / filename

experiment_artifact_filename(*, kind: str, experiment_id: str, framework: str, model: str, dataset: str, best: bool = False) -> str

Source code in recdistill/paths.py 
136
137
138
139
140
141
142
143
144
145
146
147
148
def experiment_artifact_filename(
    *,
    kind: str,
    experiment_id: str,
    framework: str,
    model: str,
    dataset: str,
    best: bool = False,
) -> str:
    stem = f"{_path_slug(framework)}_{_path_slug(model)}_{_path_slug(dataset)}_{experiment_id}"
    if best:
        stem = f"{stem}_best"
    return f"{stem}{experiment_kind_ext(kind)}"

experiment_kind_ext(kind: str) -> str

Source code in recdistill/paths.py 
151
152
153
154
155
156
157
158
159
def experiment_kind_ext(kind: str) -> str:
    kind_slug = experiment_kind_slug(kind)
    if kind_slug == "teacher":
        return TEACHER_EXT
    if kind_slug == "student":
        return STUDENT_EXT
    if kind_slug == "recdistill":
        return DISTILLED_STUDENT_EXT
    return ".pt"

experiment_id_from_config_path(path: str | Path) -> str

Source code in recdistill/paths.py 
162
163
164
165
166
167
168
def experiment_id_from_config_path(path: str | Path) -> str:
    stem = Path(path).stem
    tail = stem.rsplit("_", 1)[-1]
    is_short_hex = 6 <= len(tail) <= 16 and all(ch in "0123456789abcdefABCDEF" for ch in tail)
    if tail.isdigit():
        return tail.zfill(3)
    return tail if is_short_hex else stem

normalize_experiment_id(value: Any = None, *, config_path: str | Path | None = None) -> str

Source code in recdistill/paths.py 
171
172
173
174
175
176
177
178
179
def normalize_experiment_id(value: Any = None, *, config_path: str | Path | None = None) -> str:
    if value is None or value == "":
        if config_path is not None:
            return experiment_id_from_config_path(config_path)
        return new_experiment_id()
    if isinstance(value, int):
        return str(value).zfill(3)
    text = str(value).strip()
    return text.zfill(3) if text.isdigit() else text

_path_slug(value: Any) -> str

Source code in recdistill/paths.py 
182
183
def _path_slug(value: Any) -> str:
    return str(value).strip().replace(" ", "_").replace("-", "_").replace("+", "_").replace("/", "_").replace("\\", "_")

dataset_directory(dataset_name: str, create_if_not_exists: bool = True) -> str

Source code in recdistill/paths.py 
186
187
188
189
190
191
192
193
194
195
def dataset_directory(dataset_name: str, create_if_not_exists: bool = True) -> str:
    dataset_dir = DATA_ROOT / dataset_name
    if not dataset_dir.exists():
        if not create_if_not_exists:
            raise FileNotFoundError(
                f"Directory at {dataset_dir} not found. Please, check that dataset directory exists"
            )
        dataset_dir.mkdir(parents=True, exist_ok=True)
        print(f"Created directory at '{dataset_dir}'")
    return str(dataset_dir.resolve())

dataset_filepath(dataset_name: str, type: str = 'raw', exists: bool = True) -> str

Source code in recdistill/paths.py 
198
199
200
201
202
203
204
def dataset_filepath(dataset_name: str, type: str = "raw", exists: bool = True) -> str:
    if type not in _DATASET_FILENAME_BY_TYPE:
        raise AssertionError(f"Incorrect dataset type. Dataset type found {type}.")
    path = Path(dataset_directory(dataset_name)) / _DATASET_FILENAME_BY_TYPE[type]
    if exists and not path.exists():
        raise FileNotFoundError(f"File at {path} not found. Please, check your files")
    return str(path.resolve())

_relative_or_absolute(path: str | Path) -> Path

Source code in recdistill/paths.py 
207
208
209
def _relative_or_absolute(path: str | Path) -> Path:
    path = Path(path)
    return path if path.is_absolute() else REPO_ROOT / path

_framework_slug(framework: str | None, default: str = 'recbole') -> str

Source code in recdistill/paths.py 
212
213
214
def _framework_slug(framework: str | None, default: str = "recbole") -> str:
    raw = str(framework or default).strip().lower()
    return default if raw in {"", "auto"} else raw

_dataset_slug(dataset: str) -> str

Source code in recdistill/paths.py 
217
218
def _dataset_slug(dataset: str) -> str:
    return str(dataset).strip().lower()

_model_label(model: str) -> str

Source code in recdistill/paths.py 
221
222
223
224
225
226
227
228
229
def _model_label(model: str) -> str:
    raw = str(model).strip()
    try:
        return canonical_model_name(raw)
    except ValueError:
        cleaned = raw.replace(" ", "_").replace("/", "_").replace("\\", "_")
        if not cleaned:
            raise ValueError("Model name cannot be empty.")
        return cleaned

teacher_artifact_path(*, framework: str | None, model: str, dataset: str, embedding_dim: int, strategy: str = 'fixed') -> Path

Source code in recdistill/paths.py 
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
def teacher_artifact_path(
    *,
    framework: str | None,
    model: str,
    dataset: str,
    embedding_dim: int,
    strategy: str = "fixed",
) -> Path:
    framework_slug = _framework_slug(framework)
    model_name = _model_label(model)
    dataset_slug = _dataset_slug(dataset)
    experiment_id = f"{framework_slug}_{model_name.lower()}_{dataset_slug}_{int(embedding_dim)}"
    file_name = experiment_artifact_filename(
        kind="teacher",
        experiment_id=experiment_id,
        framework=framework_slug,
        model=model_name,
        dataset=dataset_slug,
    )
    return experiment_artifact_path(
        kind="teacher",
        experiment_id=experiment_id,
        framework=framework_slug,
        model=model_name,
        dataset=dataset_slug,
        filename=file_name,
    )

imported_teacher_artifact_path(*, framework: str | None, model: str, dataset: str, embedding_dim: int) -> Path

Source code in recdistill/paths.py 
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
def imported_teacher_artifact_path(
    *,
    framework: str | None,
    model: str,
    dataset: str,
    embedding_dim: int,
) -> Path:
    framework_slug = _framework_slug(framework)
    model_name = _model_label(model)
    dataset_slug = _dataset_slug(dataset)
    experiment_id = f"imported_{framework_slug}_{model_name.lower()}_{dataset_slug}_{int(embedding_dim)}"
    file_name = experiment_artifact_filename(
        kind="teacher",
        experiment_id=experiment_id,
        framework=framework_slug,
        model=model_name,
        dataset=dataset_slug,
    )
    return experiment_artifact_path(
        kind="teacher",
        experiment_id=experiment_id,
        framework=framework_slug,
        model=model_name,
        dataset=dataset_slug,
        filename=file_name,
    )

student_artifact_path(*, framework: str | None, model: str, dataset: str, embedding_dim: int, strategy: str = 'fixed') -> Path

Source code in recdistill/paths.py 
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
def student_artifact_path(
    *,
    framework: str | None,
    model: str,
    dataset: str,
    embedding_dim: int,
    strategy: str = "fixed",
) -> Path:
    framework_slug = _framework_slug(framework)
    model_name = _model_label(model)
    dataset_slug = _dataset_slug(dataset)
    experiment_id = f"{framework_slug}_{model_name.lower()}_{dataset_slug}_{int(embedding_dim)}"
    file_name = experiment_artifact_filename(
        kind="student",
        experiment_id=experiment_id,
        framework=framework_slug,
        model=model_name,
        dataset=dataset_slug,
    )
    return experiment_artifact_path(
        kind="student",
        experiment_id=experiment_id,
        framework=framework_slug,
        model=model_name,
        dataset=dataset_slug,
        filename=file_name,
    )

distilled_student_artifact_path(*, distiller: str, teacher_framework: str | None, teacher_model: str, student_framework: str | None, student_model: str, dataset: str, embedding_dim: int, strategy: str = 'fixed') -> Path

Source code in recdistill/paths.py 
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
def distilled_student_artifact_path(
    *,
    distiller: str,
    teacher_framework: str | None,
    teacher_model: str,
    student_framework: str | None,
    student_model: str,
    dataset: str,
    embedding_dim: int,
    strategy: str = "fixed",
) -> Path:
    distiller_name = distiller_slug(distiller)
    teacher_framework_slug = _framework_slug(teacher_framework)
    student_framework_slug = _framework_slug(student_framework)
    teacher_model_name = _model_label(teacher_model)
    student_model_name = _model_label(student_model)
    dataset_slug = _dataset_slug(dataset)
    experiment_id = (
        f"{distiller_name}_{teacher_framework_slug}_{teacher_model_name.lower()}_"
        f"to_{student_framework_slug}_{student_model_name.lower()}_{dataset_slug}"
    )
    model_label = f"{distiller_name}_{student_model_name}"
    file_name = experiment_artifact_filename(
        kind="recdistill",
        experiment_id=experiment_id,
        framework=student_framework_slug,
        model=model_label,
        dataset=dataset_slug,
    )
    return experiment_artifact_path(
        kind="recdistill",
        experiment_id=experiment_id,
        framework=student_framework_slug,
        model=model_label,
        dataset=dataset_slug,
        filename=file_name,
    )

resolve_teacher_checkpoint(*, dataset: str, teacher_model: str | None, teacher_embedding_dim: int | None, teacher_framework: str | None = None, teacher_path: str | Path | None = None) -> Path

Source code in recdistill/paths.py 
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
def resolve_teacher_checkpoint(
    *,
    dataset: str,
    teacher_model: str | None,
    teacher_embedding_dim: int | None,
    teacher_framework: str | None = None,
    teacher_path: str | Path | None = None,
) -> Path:
    if teacher_path is not None:
        return Path(teacher_path)

    if teacher_model is None or teacher_embedding_dim is None:
        raise ValueError(
            "When teacher_path is not set, both teacher_model and teacher_embedding_dim are required."
        )
    existing = _find_latest_teacher_artifact(
        framework=teacher_framework,
        model=teacher_model,
        dataset=dataset,
        embedding_dim=teacher_embedding_dim,
        prefer_best=True,
    )
    if existing is not None:
        return existing

    fixed_path = teacher_artifact_path(
        framework=teacher_framework,
        model=teacher_model,
        dataset=dataset,
        embedding_dim=teacher_embedding_dim,
        strategy="fixed",
    )
    best_path = teacher_artifact_path(
        framework=teacher_framework,
        model=teacher_model,
        dataset=dataset,
        embedding_dim=teacher_embedding_dim,
        strategy="best",
    )
    imported_path = imported_teacher_artifact_path(
        framework=teacher_framework,
        model=teacher_model,
        dataset=dataset,
        embedding_dim=teacher_embedding_dim,
    )
    if imported_path.exists():
        return imported_path
    if fixed_path.exists():
        return fixed_path
    if best_path.exists():
        return best_path
    raise FileNotFoundError(
        "No teacher artifact found for "
        f"framework={teacher_framework or 'auto'}, model={teacher_model}, "
        f"dataset={dataset}, embedding_dim={teacher_embedding_dim}. "
        "Set teacher.path in the distillation config or train/import the teacher first."
    )

_find_latest_teacher_artifact(*, framework: str | None, model: str, dataset: str, embedding_dim: int, prefer_best: bool = True) -> Path | None

Source code in recdistill/paths.py 
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
def _find_latest_teacher_artifact(
    *,
    framework: str | None,
    model: str,
    dataset: str,
    embedding_dim: int,
    prefer_best: bool = True,
) -> Path | None:
    root = RESULTS_ROOT / "teacher"
    if not root.exists():
        return None

    framework_slug = _framework_slug(framework)
    model_name = _model_label(model)
    dataset_slug = _dataset_slug(dataset)
    candidates: list[Path] = []
    for path in root.glob("*/artifacts/*.teacher"):
        name = path.name.lower()
        if framework_slug not in name:
            continue
        if model_name.lower() not in name:
            continue
        if dataset_slug not in name:
            continue
        candidates.append(path)

    if not candidates:
        return None
    if prefer_best:
        best_candidates = [path for path in candidates if path.stem.endswith("_best")]
        if best_candidates:
            candidates = best_candidates
    return max(candidates, key=lambda path: path.stat().st_mtime)

resolve_student_checkpoint(*, dataset: str, distiller: str, teacher_model: str | None, student_backbone: str | None, student_embedding_dim: int, teacher_framework: str | None = None, student_framework: str | None = None, output_path: str | Path | None = None, strategy: str = 'fixed') -> Path

Source code in recdistill/paths.py 
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
def resolve_student_checkpoint(
    *,
    dataset: str,
    distiller: str,
    teacher_model: str | None,
    student_backbone: str | None,
    student_embedding_dim: int,
    teacher_framework: str | None = None,
    student_framework: str | None = None,
    output_path: str | Path | None = None,
    strategy: str = "fixed",
) -> Path:
    if output_path is not None:
        return Path(output_path)

    raw_distiller = str(distiller).strip().lower()
    is_plain = raw_distiller in {"none", "plain", "no", "false", "0"}
    if is_plain:
        if student_backbone is None:
            raise ValueError("student_backbone is required for plain student path resolution.")
        return student_artifact_path(
            framework=student_framework,
            model=student_backbone,
            dataset=dataset,
            embedding_dim=student_embedding_dim,
            strategy=strategy,
        )

    if teacher_model is None or student_backbone is None:
        raise ValueError("teacher_model and student_backbone are required for distilled student path resolution.")
    return distilled_student_artifact_path(
        distiller=distiller,
        teacher_framework=teacher_framework,
        teacher_model=teacher_model,
        student_framework=student_framework,
        student_model=student_backbone,
        dataset=dataset,
        embedding_dim=student_embedding_dim,
        strategy=strategy,
    )

history_path(checkpoint_path: str | Path) -> Path

Source code in recdistill/paths.py 
493
494
def history_path(checkpoint_path: str | Path) -> Path:
    return Path(checkpoint_path).with_suffix(".history.json")

best_checkpoint_path(checkpoint_path: str | Path) -> Path

Source code in recdistill/paths.py 
497
498
499
500
501
def best_checkpoint_path(checkpoint_path: str | Path) -> Path:
    path = Path(checkpoint_path)
    if path.suffix in {STUDENT_EXT, DISTILLED_STUDENT_EXT, TEACHER_EXT}:
        return path.with_name(f"{path.stem}_best{path.suffix}")
    return path.with_name(f"{path.stem}_best{path.suffix or '.pt'}")

early_stop_checkpoint_path(checkpoint_path: str | Path) -> Path

Source code in recdistill/paths.py 
504
505
506
507
508
def early_stop_checkpoint_path(checkpoint_path: str | Path) -> Path:
    path = Path(checkpoint_path)
    if path.suffix in {STUDENT_EXT, DISTILLED_STUDENT_EXT, TEACHER_EXT}:
        return path.with_name(f"{path.stem}.earlystop_best{path.suffix}")
    return path.with_suffix(".earlystop_best.pt")

recommendation_path(*, dataset: str, distiller: str, teacher_model: str | None, student_embedding_dim: int, student_backbone: str | None = None, filename: str | None = None) -> Path

Source code in recdistill/paths.py 
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
def recommendation_path(
    *,
    dataset: str,
    distiller: str,
    teacher_model: str | None,
    student_embedding_dim: int,
    student_backbone: str | None = None,
    filename: str | None = None,
) -> Path:
    teacher_slug = model_slug(teacher_model) if teacher_model is not None else "teacher"
    student_slug = model_slug(student_backbone) if student_backbone is not None else "student"
    base = (
        RESULTS_ROOT
        / "recdistill"
        / distiller_slug(distiller)
        / teacher_slug
        / student_slug
        / str(dataset).lower()
        / "recs"
    )
    return base / (filename or f"student_{int(student_embedding_dim)}.tsv")

performance_path(*, dataset: str, distiller: str, teacher_model: str | None, student_backbone: str | None = None, phase: str = 'fixed', filename: str = 'metrics.json') -> Path

Source code in recdistill/paths.py 
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
def performance_path(
    *,
    dataset: str,
    distiller: str,
    teacher_model: str | None,
    student_backbone: str | None = None,
    phase: str = "fixed",
    filename: str = "metrics.json",
) -> Path:
    teacher_slug = model_slug(teacher_model) if teacher_model is not None else "teacher"
    student_slug = model_slug(student_backbone) if student_backbone is not None else "student"
    return (
        RESULTS_ROOT
        / "recdistill"
        / distiller_slug(distiller)
        / teacher_slug
        / student_slug
        / str(dataset).lower()
        / phase
        / "perf"
        / filename
    )

teacher_weights_path(model: str, dataset: str, embedding_dim: int, phase: str = 'best', framework: str | None = None) -> str

Compatibility helper for legacy scripts, backed by the framework path module.

Source code in recdistill/paths.py 
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
def teacher_weights_path(
    model: str,
    dataset: str,
    embedding_dim: int,
    phase: str = "best",
    framework: str | None = None,
) -> str:
    """Compatibility helper for legacy scripts, backed by the framework path module."""
    path = teacher_artifact_path(
        framework=framework,
        model=model,
        dataset=dataset,
        embedding_dim=embedding_dim,
        strategy=phase,
    )
    path.parent.mkdir(parents=True, exist_ok=True)
    return str(path)

student_weights_path(distiller: str, teacher: str, dataset: str, embedding_dim: int, phase: str = 'fixed', student: str | None = None) -> str

Compatibility helper for exported distilled student payloads.

Source code in recdistill/paths.py 
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
def student_weights_path(
    distiller: str,
    teacher: str,
    dataset: str,
    embedding_dim: int,
    phase: str = "fixed",
    student: str | None = None,
) -> str:
    """Compatibility helper for exported distilled student payloads."""
    path = distilled_student_artifact_path(
        distiller=distiller,
        teacher_framework=None,
        teacher_model=teacher,
        student_framework=None,
        student_model=student or teacher,
        dataset=dataset,
        embedding_dim=embedding_dim,
        strategy=phase,
    )
    path.parent.mkdir(parents=True, exist_ok=True)
    return str(path)

resolve_student_checkpoint_from_args(args: Any, distiller_name: str) -> Path

Source code in recdistill/paths.py 
600
601
602
603
604
605
606
607
608
609
610
611
def resolve_student_checkpoint_from_args(args: Any, distiller_name: str) -> Path:
    return resolve_student_checkpoint(
        dataset=args.dataset,
        distiller=distiller_name,
        teacher_framework=getattr(args, "teacher_framework", None),
        teacher_model=args.teacher_model,
        student_backbone=args.student_backbone,
        student_framework=getattr(args, "student_framework", None),
        student_embedding_dim=args.student_embedding_dim,
        output_path=args.output_path,
        strategy=getattr(args, "output_strategy", "fixed"),
    )

resolve_teacher_checkpoint_from_args(args: Any) -> Path

Source code in recdistill/paths.py 
614
615
616
617
618
619
620
621
def resolve_teacher_checkpoint_from_args(args: Any) -> Path:
    return resolve_teacher_checkpoint(
        dataset=args.dataset,
        teacher_model=args.teacher_model,
        teacher_embedding_dim=args.teacher_embedding_dim,
        teacher_framework=getattr(args, "teacher_framework", None),
        teacher_path=args.teacher_path,
    )

Tracking

WandBRunLogger

Source code in recdistill/tracking.py 
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
class WandBRunLogger:
    def __init__(
        self,
        project: str,
        run_name: str | None = None,
        entity: str | None = None,
        tags: list[str] | None = None,
        group: str | None = None,
        notes: str | None = None,
        config: dict | None = None,
    ):
        try:
            import wandb
        except ModuleNotFoundError as exc:
            raise ModuleNotFoundError(
                "wandb is required for W&B logging. Install it with `pip install wandb`."
            ) from exc

        self._wandb = wandb
        self.run = self._wandb.init(
            project=project,
            name=run_name,
            entity=entity,
            tags=tags,
            group=group,
            notes=notes,
            config=config or {},
        )

    def log_start(self, payload: dict) -> None:
        if self.run is None:
            return
        self.run.summary["status"] = payload.get("status", "running")
        self.run.summary["started_at_utc"] = payload.get("started_at_utc", utc_now_iso())
        for key, value in payload.items():
            if key in {"status", "started_at_utc"}:
                continue
            self.run.config[key] = value

    def log_epoch(self, epoch_payload: dict) -> None:
        if self.run is None:
            return
        self._wandb.log(epoch_payload, step=int(epoch_payload.get("epoch", 0)))

    def log_end(self, payload: dict) -> None:
        if self.run is None:
            return
        self.run.summary["status"] = payload.get("status", "completed")
        self.run.summary["ended_at_utc"] = payload.get("ended_at_utc", utc_now_iso())
        for key, value in payload.items():
            if key in {"status", "ended_at_utc"}:
                continue
            self.run.summary[key] = value
        exit_code = 1 if payload.get("status") == "failed" else 0
        self._wandb.finish(exit_code=exit_code)

run = self._wandb.init(project=project, name=run_name, entity=entity, tags=tags, group=group, notes=notes, config=(config or {})) instance-attribute

__init__(project: str, run_name: str | None = None, entity: str | None = None, tags: list[str] | None = None, group: str | None = None, notes: str | None = None, config: dict | None = None)

Source code in recdistill/tracking.py 
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
def __init__(
    self,
    project: str,
    run_name: str | None = None,
    entity: str | None = None,
    tags: list[str] | None = None,
    group: str | None = None,
    notes: str | None = None,
    config: dict | None = None,
):
    try:
        import wandb
    except ModuleNotFoundError as exc:
        raise ModuleNotFoundError(
            "wandb is required for W&B logging. Install it with `pip install wandb`."
        ) from exc

    self._wandb = wandb
    self.run = self._wandb.init(
        project=project,
        name=run_name,
        entity=entity,
        tags=tags,
        group=group,
        notes=notes,
        config=config or {},
    )

log_start(payload: dict) -> None

Source code in recdistill/tracking.py 
40
41
42
43
44
45
46
47
48
def log_start(self, payload: dict) -> None:
    if self.run is None:
        return
    self.run.summary["status"] = payload.get("status", "running")
    self.run.summary["started_at_utc"] = payload.get("started_at_utc", utc_now_iso())
    for key, value in payload.items():
        if key in {"status", "started_at_utc"}:
            continue
        self.run.config[key] = value

log_epoch(epoch_payload: dict) -> None

Source code in recdistill/tracking.py 
50
51
52
53
def log_epoch(self, epoch_payload: dict) -> None:
    if self.run is None:
        return
    self._wandb.log(epoch_payload, step=int(epoch_payload.get("epoch", 0)))

log_end(payload: dict) -> None

Source code in recdistill/tracking.py 
55
56
57
58
59
60
61
62
63
64
65
def log_end(self, payload: dict) -> None:
    if self.run is None:
        return
    self.run.summary["status"] = payload.get("status", "completed")
    self.run.summary["ended_at_utc"] = payload.get("ended_at_utc", utc_now_iso())
    for key, value in payload.items():
        if key in {"status", "ended_at_utc"}:
            continue
        self.run.summary[key] = value
    exit_code = 1 if payload.get("status") == "failed" else 0
    self._wandb.finish(exit_code=exit_code)

utc_now_iso() -> str

Source code in recdistill/tracking.py 
7
8
def utc_now_iso() -> str:
    return datetime.now(timezone.utc).isoformat()

parse_csv_list(value: str | None) -> list[str] | None

Source code in recdistill/tracking.py 
68
69
70
71
72
def parse_csv_list(value: str | None) -> list[str] | None:
    if value is None:
        return None
    parsed = [item.strip() for item in value.split(",") if item.strip()]
    return parsed or None

resolve_wandb_logger(args: Any, base_config: dict) -> WandBRunLogger | None

Source code in recdistill/tracking.py 
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
def resolve_wandb_logger(args: Any, base_config: dict) -> WandBRunLogger | None:
    if not args.wandb_log:
        return None
    if not args.wandb_project:
        raise ValueError("--wandb-project is required when --wandb-log is enabled.")

    return WandBRunLogger(
        project=args.wandb_project,
        run_name=args.wandb_run_name,
        entity=args.wandb_entity,
        tags=parse_csv_list(args.wandb_tags),
        group=args.wandb_group,
        notes=args.wandb_notes,
        config=base_config,
    )