| import numpy as np | |
| from scripts.run_pairwise_unary_stack import score_metrics | |
| def test_stacked_metrics_compute_per_class_accuracy_and_pair_pass(): | |
| y = np.array([0, 1, 0, 1], dtype=np.int8) | |
| pred = np.array([0, 1, 1, 0], dtype=np.int8) | |
| result = score_metrics(y, pred, pair_count=2) | |
| assert result["accuracy"] == 0.5 | |
| assert result["safe_accuracy"] == 0.5 | |
| assert result["vulnerable_accuracy"] == 0.5 | |
| assert result["pair_pass_at_1"] == 0.5 | |
| assert result["pair_pass_wilson_95"] is not None | |
| def test_stacked_metrics_track_single_class_collapse(): | |
| y = np.array([0, 1, 0, 1], dtype=np.int8) | |
| pred = np.zeros(4, dtype=np.int8) | |
| result = score_metrics(y, pred, pair_count=2) | |
| assert result["safe_accuracy"] == 1.0 | |
| assert result["vulnerable_accuracy"] == 0.0 | |
| assert result["predicted_vulnerable_fraction"] == 0.0 | |
| assert result["pair_pass_at_1"] == 0.0 | |