| """Write a JSON record of a train/eval split: ids and per-scenario counts per file, plus which scenarios are missing. | |
| usage: python scripts/record_split.py out.json name1=file1.jsonl name2=file2.jsonl ... | |
| """ | |
| import json, sys | |
| from collections import Counter | |
| out, specs = sys.argv[1], sys.argv[2:] | |
| res = {} | |
| for s in specs: | |
| name, path = s.split("=", 1) | |
| with open(path, encoding="utf-8") as f: | |
| rows = [json.loads(l) for l in f if l.strip()] | |
| sc = Counter((r.get("meta") or {}).get("scenario") or "?" for r in rows) | |
| res[name] = {"path": path, "n": len(rows), "n_with_box_questions": sum(any(q.get("type") == "box" for q in r["questions"].values()) for r in rows), | |
| "scenarios": dict(sorted(sc.items())), "ids": [r["id"] for r in rows]} | |
| allsc = set().union(*(set(v["scenarios"]) for v in res.values())) | |
| for v in res.values(): | |
| v["missing_scenarios"] = sorted(allsc - set(v["scenarios"])) | |
| with open(out, "w", encoding="utf-8", newline="\n") as f: | |
| json.dump(res, f, indent=1, ensure_ascii=False) | |
| print(json.dumps({k: {"n": v["n"], "box": v["n_with_box_questions"], "n_scen": len(v["scenarios"]), "missing": v["missing_scenarios"]} for k, v in res.items()})) | |