Download scripts/verify.py from DosimeterAudit/ai-disinformation-benchmark: direct link, hf CLI and curl.
- Browser
- Download file 2.27 kB
-
https://huggingface.co/datasets/DosimeterAudit/ai-disinformation-benchmark/resolve/main/scripts/verify.py
- Command line
-
hf download hf://datasets/DosimeterAudit/ai-disinformation-benchmark/scripts/verify.py
-
curl -L -o verify.py https://huggingface.co/datasets/DosimeterAudit/ai-disinformation-benchmark/resolve/main/scripts/verify.py
2.27 kB
| """Checks this repository: the checksums of the data files, and the figures of datasets.csv counted again from the responses. | |
| python3 scripts/verify.py | |
| Python 3.9 or later, nothing to install. The last line is VERIFY: OK when everything matches. | |
| """ | |
| import csv, hashlib, json, pathlib, sys | |
| ROOT = pathlib.Path(__file__).resolve().parent.parent | |
| bad = 0 | |
| for line in (ROOT / "CHECKSUMS.txt").read_text(encoding="utf-8").splitlines(): | |
| want, name = line.split(" ", 1) | |
| got = hashlib.sha256((ROOT / name).read_bytes()).hexdigest() | |
| if got != want: | |
| bad += 1 | |
| print("CHECKSUM DIFFERS:", name) | |
| print("checksums checked:", len((ROOT / "CHECKSUMS.txt").read_text(encoding="utf-8").splitlines())) | |
| board = {} | |
| cited = consulted = 0 | |
| for d in csv.DictReader(open(ROOT / "datasets.csv", encoding="utf-8")): | |
| folder = ROOT / d["folder"] | |
| rows = [json.loads(l) for l in open(folder / "responses.jsonl", encoding="utf-8")] | |
| table = list(csv.DictReader(open(folder / "responses.csv", encoding="utf-8"))) | |
| counted = sum(r["counted"] for r in rows) | |
| ok = len(rows) == int(d["responses"]) == len(table) and counted == int(d["counted"]) | |
| bad += not ok | |
| print(f"{d['dataset']}: {len(rows)} responses, {counted} counted", "" if ok else " <- DIFFERS from datasets.csv") | |
| for r in rows: | |
| cited += sum(1 for s in r["sources"] if s["shown"] != "in the list of consulted pages") | |
| consulted += sum(1 for s in r["sources"] if s["shown"] == "in the list of consulted pages") | |
| if r["on_board"]: | |
| x = board.setdefault(r["assistant"], [0, 0]) | |
| x[0] += 1 | |
| x[1] += r["counted"] | |
| print() | |
| total = sum(int(d["responses"]) for d in csv.DictReader(open(ROOT / "datasets.csv", encoding="utf-8"))) | |
| print(f"All datasets together: {total:,} responses, {sum(n for n, _ in board.values()):,} of them on the board") | |
| print(f"Cited sources in the responses: {cited:,}, and {consulted:,} consulted pages") | |
| print() | |
| print("Our own measurements on the board, by assistant:") | |
| for name, (n, c) in sorted(board.items(), key=lambda kv: -kv[1][1] / kv[1][0]): | |
| print(f" {name:22} {c:4} of {n:4} {100 * c / n:5.1f}%") | |
| print() | |
| print("VERIFY: OK" if not bad else f"VERIFY: {bad} PROBLEMS") | |
| sys.exit(1 if bad else 0) | |