Download scripts/study_evidence.py from Gaurav711/HyperFlow: direct link, hf CLI and curl.
- Browser
- Download file 2.7 kB
-
https://huggingface.co/spaces/Gaurav711/HyperFlow/resolve/main/scripts/study_evidence.py
- Command line
-
hf download hf://spaces/Gaurav711/HyperFlow/scripts/study_evidence.py
-
curl -L -o study_evidence.py https://huggingface.co/spaces/Gaurav711/HyperFlow/resolve/main/scripts/study_evidence.py
2.7 kB
| """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 | |