#!/usr/bin/env python3
import argparse
import json
from collections import Counter, defaultdict
from pathlib import Path
from typing import Dict


def load_jsonl(path: Path):
	with open(path, "r", encoding="utf-8") as f:
		for line in f:
			if line.strip():
				yield json.loads(line)


def parse_targets(targets: str) -> Dict[str, float]:
	result: Dict[str, float] = {}
	for part in targets.split():
		if "=" not in part:
			continue
		k, v = part.split("=", 1)
		result[k.strip()] = float(v.strip())
	return result


def main() -> None:
	parser = argparse.ArgumentParser(description="Check category balance vs targets.")
	parser.add_argument("--jsonl", required=True, help="Input JSONL (train recommended)")
	parser.add_argument("--targets", nargs="+", required=False, default=[], help='Targets like "cours=0.2 conferences=0.2 podcasts=0.2 meetings=0.2 interviews=0.2"')
	args = parser.parse_args()

	items = list(load_jsonl(Path(args.jsonl)))
	n = len(items)
	counts = Counter([it.get("category", "unknown") for it in items])
	target_map: Dict[str, float] = {}
	for tok in args.targets:
		if "=" in tok:
			k, v = tok.split("=", 1)
			target_map[k] = float(v)
	print(f"Total items: {n}")
	print("Observed proportions:")
	for k, c in counts.items():
		print(f"  {k:15s} {c:6d}  ({c / max(n,1):.3f})")
	if target_map:
		print("Target proportions:")
		for k, v in target_map.items():
			print(f"  {k:15s} {v:.3f}")
		print("Delta (observed - target):")
		for k, v in target_map.items():
			obs = counts.get(k, 0) / max(n, 1)
			print(f"  {k:15s} {obs - v:+.3f}")
		print("Recommendation: upsample categories with negative deltas; downsample positives.")


if __name__ == "__main__":
	main()