"""Multi-label metrics, archived independently of dataset/runtime imports."""
import numpy as np

def classification_metrics(targets, predicted, names):
    """Compute multi-label metrics directly; no optional compiled metrics library."""
    actual = targets.astype(bool)
    tp = (actual & predicted).sum(axis=0)
    fp = (~actual & predicted).sum(axis=0)
    fn = (actual & ~predicted).sum(axis=0)
    def divide(a, b):
        return np.divide(a, b, out=np.zeros_like(a, dtype=float), where=b != 0)
    precision = divide(tp, tp + fp)
    recall = divide(tp, tp + fn)
    f1 = divide(2 * tp, 2 * tp + fp + fn)
    denom = int(2 * tp.sum() + fp.sum() + fn.sum())
    return {"micro_f1": float(2 * tp.sum() / denom) if denom else 0.0,
            "macro_f1": float(f1.mean()),
            "per_label": {name: {"precision": float(precision[i]), "recall": float(recall[i]),
                                  "f1": float(f1[i]), "support": int(actual[:, i].sum())}
                          for i, name in enumerate(names)}}