File size: 2,697 Bytes
2b86c60 | 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 | """Write portable, source-linked evidence for explicitly invoked offline studies."""
from datetime import datetime, timezone
import importlib.metadata
import platform
from pathlib import Path
import sys
sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
from backend.flagship.ml import _atomic, _hash, _json
def write_evidence(destination, *, study, result, elapsed_seconds, input_path=None):
root = Path(__file__).resolve().parents[1]
sources = ["backend/flagship/ml.py", "scripts/study_evidence.py", f"scripts/study_{study}.py"]
if study == "censoring":
sources.append("backend/ml/censored_demand.py")
packages = {}
for name in ("numpy", "pandas", "scipy", "scikit-learn", "lightgbm", "openpyxl"):
try:
packages[name] = importlib.metadata.version(name)
except importlib.metadata.PackageNotFoundError:
packages[name] = "unavailable"
report = {
"schema_version": 1,
"study": study,
"captured_at": datetime.now(timezone.utc).isoformat(),
"command": ["python3", *sys.argv],
"elapsed_wall_seconds": elapsed_seconds,
"runtime": {"python": platform.python_version(), "platform": platform.platform(), "packages": packages},
"source_sha256": {source: _hash((root / source).read_bytes()) for source in sources},
"result_sha256": _hash(_json(result)),
"result": result,
}
if input_path is not None:
# Use the bytes ingestion consumed, not a file that might have been
# replaced after the study finished.
metadata = result.get("manifest", {}).get("dataset", result.get("provenance", {}))
report["input"] = {"filename": Path(input_path).name, "sha256": metadata.get("dataset_sha256")}
report["limitations"] = [
"A UK-based online retailer's recorded positive sales; not Indian quick-commerce evidence.",
"Cohort includes transactions across countries in the official workbook, not a UK-only customer filter.",
"Top products are selected from the initial training period; this is not an all-product benchmark.",
"One-step walk-forward evaluation uses preceding observed days; not a 28-day recursive forecast.",
"No stockout labels, warehouse inventory observations, causal deployment benefit, or recovered hidden sales.",
"Serving history includes observed holdout days; model fitting and validation selection exclude their targets.",
"No model is promoted by this study command.",
]
else:
report["limitations"] = result["limitations"]
_atomic(Path(destination), _json(report))
return report
|