Function _make_seed_index

tests/test_signal_embedding.py:14–31

(tmp_path: Path)

def _make_seed_index(tmp_path: Path):
    dim = 384
    rng = np.random.RandomState(42)
    # Create cluster of similar vectors for tier 0 so KNN has enough
    # same-label neighbors to exceed MIN_CONFIDENCE_TO_VOTE (0.30).
    base = rng.randn(dim).astype(np.float32)
    vecs = [base + rng.randn(dim).astype(np.float32) * 0.2 for _ in range(5)]
    # Add a distant vector for tier 3
    far = rng.randn(dim).astype(np.float32)
    vecs.append(far)
    embeddings = np.array(vecs, dtype=np.float32)
    norms = np.linalg.norm(embeddings, axis=1, keepdims=True)
    embeddings = embeddings / norms
    labels = [0, 0, 0, 0, 0, 3]
    np.save(tmp_path / "seed_embeddings.npy", embeddings)
    with open(tmp_path / "seed_labels.json", "w") as f:
        json.dump(labels, f)
    return embeddings, labels


def test_embedding_signal_with_seed_index(tmp_path):