Chinese_Debate_Documents / scripts /validate_card.py
DBWBD's picture
updated
6342c6d
"""Validate Chinese_Debate_Documents/README.md as a HuggingFace dataset card.
Checks YAML frontmatter against HF's expected keys and value shapes, and that
declared splits match `data/train-*.parquet` reality.
Exit codes: 0 = clean, 1 = warnings, 2 = errors.
Usage:
python validate_card.py [path/to/README.md]
"""
from __future__ import annotations
import json
import re
import sys
from pathlib import Path
VALID_LICENSES = {
"mit", "apache-2.0", "bsd-3-clause", "bsd-2-clause",
"cc-by-4.0", "cc-by-sa-4.0", "cc-by-nc-4.0", "cc-by-nc-sa-4.0",
"cc0-1.0", "gpl-3.0", "lgpl-3.0", "unlicense", "other",
}
VALID_SIZE = {
"n<1K", "1K<n<10K", "10K<n<100K", "100K<n<1M", "1M<n<10M", "10M<n<100M", "100M<n<1B",
}
# A non-exhaustive whitelist of HF task category strings that fit ASR + NLP.
VALID_TASKS = {
"automatic-speech-recognition", "text-classification", "text-generation",
"token-classification", "summarization", "translation", "question-answering",
"feature-extraction", "sentence-similarity", "zero-shot-classification",
"fill-mask", "conversational",
}
def parse_frontmatter(text: str) -> tuple[dict, str]:
"""Very small YAML parser sufficient for HF dataset-card frontmatter.
Supports: top-level scalars, lists (- item), and nested mappings via indent.
Returns (parsed_dict, body_after_frontmatter)."""
if not text.startswith("---\n") and not text.startswith("---\r\n"):
return {}, text
end = text.find("\n---", 4)
if end == -1:
return {}, text
raw = text[4:end].lstrip("\n")
body = text[end + 4 :]
# Try real YAML first if available
try:
import yaml # type: ignore
return yaml.safe_load(raw) or {}, body
except ImportError:
pass
# Fallback: minimal hand-rolled parser
result: dict = {}
stack: list[tuple[int, object]] = [(0, result)]
pending_key = None
for line in raw.splitlines():
if not line.strip() or line.lstrip().startswith("#"):
continue
indent = len(line) - len(line.lstrip(" "))
while stack and indent < stack[-1][0]:
stack.pop()
stripped = line.strip()
container = stack[-1][1]
if stripped.startswith("- "):
if isinstance(container, list):
container.append(stripped[2:].strip())
elif pending_key and isinstance(container, dict):
container.setdefault(pending_key, []).append(stripped[2:].strip())
continue
if ":" in stripped:
k, _, v = stripped.partition(":")
k = k.strip()
v = v.strip()
if not v:
# nested block follows
new_container: object
# Peek next non-empty line — if it starts with `-`, list; else dict
new_container = {}
container[k] = new_container
stack.append((indent + 2, new_container))
pending_key = k
else:
if v.startswith("[") and v.endswith("]"):
items = [s.strip().strip("'\"") for s in v[1:-1].split(",") if s.strip()]
container[k] = items
else:
container[k] = v.strip().strip("'\"")
pending_key = None
return result, body
def validate(card_path: Path) -> int:
text = card_path.read_text(encoding="utf-8")
meta, body = parse_frontmatter(text)
errors: list[str] = []
warnings: list[str] = []
if not meta:
errors.append("frontmatter not found or unparseable")
_report(errors, warnings)
return 2
# Required scalars
for key in ("license", "pretty_name"):
if not meta.get(key):
errors.append(f"missing required key: {key}")
lic = meta.get("license")
if lic and lic not in VALID_LICENSES:
warnings.append(f"license '{lic}' not in SPDX whitelist (may be flagged by Hub)")
# Language
langs = meta.get("language") or meta.get("languages")
if not langs:
errors.append("missing 'language' (expected list, e.g. [zh])")
elif isinstance(langs, list) and "zh" not in [l.lower() for l in langs]:
warnings.append("language list doesn't include 'zh'")
# Size categories
sizes = meta.get("size_categories")
if sizes:
bad = [s for s in (sizes if isinstance(sizes, list) else [sizes]) if s not in VALID_SIZE]
if bad:
errors.append(f"invalid size_categories values: {bad}")
# Task categories
tasks = meta.get("task_categories")
if tasks:
bad = [t for t in (tasks if isinstance(tasks, list) else [tasks]) if t not in VALID_TASKS]
if bad:
warnings.append(f"task_categories not in known taxonomy: {bad}")
# Configs + dataset_info consistency
configs = meta.get("configs") or []
dataset_info = meta.get("dataset_info")
if not configs:
warnings.append("no 'configs' declared — Hub will default to autodetect")
if dataset_info:
if isinstance(dataset_info, dict):
splits = dataset_info.get("splits") or []
for sp in splits if isinstance(splits, list) else []:
if isinstance(sp, dict) and not sp.get("name"):
errors.append("a dataset_info.splits entry is missing 'name'")
else:
warnings.append("dataset_info is not a mapping")
else:
warnings.append("no 'dataset_info' — Hub will infer features but won't show split sizes")
# Cross-check splits against parquet on disk
repo = card_path.parent
parquets = sorted((repo / "data").glob("*.parquet")) if (repo / "data").is_dir() else []
if not parquets:
warnings.append("no data/*.parquet found — Hub will fall back to scanning raw files")
# Body section presence
required_sections = [
"Dataset Summary", "Languages", "Data Fields",
"Source Data", "Licensing", "Citation",
]
for sect in required_sections:
if sect.lower() not in body.lower():
warnings.append(f"missing prose section: {sect}")
return _report(errors, warnings)
def _report(errors: list[str], warnings: list[str]) -> int:
if errors:
print("ERRORS:")
for e in errors:
print(f" - {e}")
if warnings:
print("WARNINGS:")
for w in warnings:
print(f" - {w}")
if not errors and not warnings:
print("card OK")
return 0
return 2 if errors else 1
if __name__ == "__main__":
path = Path(sys.argv[1] if len(sys.argv) > 1 else "README.md")
sys.exit(validate(path))