| import pytest |
| from geoflip.core import score, overall_ranking, inside |
|
|
|
|
| def row(sid, task, label, pair=None): |
| return {'id': sid, 'task': task, 'reference_label': label, 'pair_id': pair} |
|
|
|
|
| def test_task_macro_class_balance_and_missing_denominator(): |
| truth = [row('a', 'x', 'still', 'p'), row('b', 'x', 'fall', 'p'), |
| row('c', 'x', 'still'), row('d', 'y', 'left')] |
| pred = [{'id': 'a', 'label': 'still', 'violation': 'clear'}, |
| {'id': 'b', 'label': 'other', 'violation': 'flagged'}, |
| {'id': 'd', 'label': 'left', 'violation': 'uncertain'}] |
| result = score(truth, pred) |
| assert result['event_percent'] == pytest.approx(62.5) |
| assert result['structure_percent'] == pytest.approx(100 / 6) |
| assert result['missing'] == 1 |
| assert result['complete_pairs'] == 1 |
| assert result['pair_complete_accuracy'] == 0 |
|
|
|
|
| def test_error_and_uncertainty_cannot_improve_score(): |
| truth = [row('a', 'x', 'still'), row('b', 'x', 'fall')] |
| result = score(truth, [{'id': 'a', 'label': 'still', 'violation': 'clear', 'error_type': 'DecodeError'}, |
| {'id': 'b', 'label': 'uncertain', 'violation': 'uncertain'}]) |
| assert result['event_percent'] == result['structure_percent'] == 0 |
| assert result['failed'] == 1 |
| assert result['prediction_coverage'] == .5 |
|
|
|
|
| def test_reject_duplicate_and_wrong_scope(): |
| truth = [row('a', 'x', 'still')] |
| with pytest.raises(ValueError, match='duplicate'): |
| score(truth, [{'id': 'a'}, {'id': 'a'}]) |
| with pytest.raises(ValueError, match='outside'): |
| score(truth, [{'id': 'b'}]) |
| with pytest.raises(ValueError, match='Empty'): |
| score([], []) |
|
|
|
|
| def test_rank_similarity_is_one_of_three_equal_dimensions(): |
| rows = [{'model': 'a', 'event': 10, 'structure': 20, 'psnr': 30, 'ssim': .8}, |
| {'model': 'b', 'event': 20, 'structure': 10, 'psnr': 10, 'ssim': .9}, |
| {'model': 'c', 'event': 10, 'structure': 20, 'psnr': 30, 'ssim': .8}] |
| result = overall_ranking(rows) |
| a = next(r for r in result if r['model'] == 'a') |
| c = next(r for r in result if r['model'] == 'c') |
| assert a['mean_rank'] == pytest.approx((2.5 + 1.5 + (1.5 + 2.5) / 2) / 3) |
| assert a['rank'] == c['rank'] |
|
|
|
|
| def test_dataset_cannot_escape_root(tmp_path): |
| with pytest.raises(ValueError, match='leaves'): |
| inside(tmp_path, '../private-file') |
| with pytest.raises(ValueError, match='relative'): |
| inside(tmp_path, str(tmp_path / 'absolute-file')) |
|
|
