Add unit tests for pipeline.py + fix broken export.py import

[Copilot]

• Fix pre-existing import error in src/animl/export.py
• Create tests/test_pipeline.py (29 tests)
• Create tests/test_split.py (46 tests) covering get_animals, get_empty, train_val_test

Original prompt

Task

Create a comprehensive unittest file at tests/test_pipeline.py on the dev branch for src/animl/pipeline.py.

Source file to test

src/animl/pipeline.py contains two public functions:

• from_paths(image_dir, detector_file, classifier_file, classlist_file, ...)
• from_config(config)

Both functions are orchestration pipelines that call many submodules. All submodule calls must be mocked — do not load real models or real images.

Key dependencies to mock

All of the following must be patched in tests (patch target is the animl.pipeline namespace):

• animl.pipeline.file_management.WorkingDirectory
• animl.pipeline.file_management.build_file_manifest
• animl.pipeline.file_management.check_file
• animl.pipeline.file_management.load_data
• animl.pipeline.file_management.save_data
• animl.pipeline.video_processing.extract_frames
• animl.pipeline.detection.load_detector
• animl.pipeline.detection.detect
• animl.pipeline.detection.parse_detections
• animl.pipeline.split.get_animals
• animl.pipeline.split.get_empty
• animl.pipeline.classification.load_classifier
• animl.pipeline.classification.classify
• animl.pipeline.classification.single_classification
• animl.pipeline.classification.sequence_classification
• animl.pipeline.export.export_folders
• animl.pipeline.visualization.plot_all_bounding_boxes

WorkingDirectory mock setup

The WorkingDirectory mock must have these attributes set as MagicMock() or Path-like objects:

• filemanifest
• imageframes
• detections
• mdraw
• predictions
• results
• linkdir
• visdir

Test data setup (setUpClass)

Create shared test fixtures:

cls.image_dir = tempfile.mkdtemp()
cls.detector_file = str(Path(cls.image_dir) / 'detector.pt')
cls.classifier_file = str(Path(cls.image_dir) / 'classifier.pt')
cls.classlist_file = str(Path(cls.image_dir) / 'classes.csv')
Path(cls.detector_file).touch()
Path(cls.classifier_file).touch()

cls.mock_manifest = pd.DataFrame({
    'filepath': ['a.jpg', 'b.jpg'],
    'frame': [0, 0],
    'station': ['cam1', 'cam1'],
    'datetime': ['2023-01-01 10:00:00', '2023-01-01 10:01:00'],
    'extension': ['.jpg', '.jpg'],
})
cls.mock_animals = cls.mock_manifest.copy()
cls.mock_animals['category'] = 1
cls.mock_animals['conf'] = 0.9
cls.mock_animals['bbox_x'] = 0.1
cls.mock_animals['bbox_y'] = 0.1
cls.mock_animals['bbox_w'] = 0.2
cls.mock_animals['bbox_h'] = 0.3

cls.mock_empty = pd.DataFrame(columns=cls.mock_manifest.columns)
cls.mock_predictions = np.array([[0.9, 0.05, 0.05], [0.1, 0.8, 0.1]])
cls.mock_results = cls.mock_animals.copy()
cls.mock_results['prediction'] = 'cat'
cls.mock_results['confidence'] = 0.9
cls.mock_class_list = pd.DataFrame({'class': ['cat', 'dog', 'bird']})

from_paths tests — class TestFromPaths

Write tests using unittest.mock.patch as context managers or decorators. For each test, mock all dependencies listed above. The check_file mock should return False by default (simulate no cached files).

Tests to write:

1. test_returns_dataframe — returns a pd.DataFrame
2. test_calls_build_file_manifest — build_file_manifest is called once
3. test_calls_extract_frames — video_processing.extract_frames is called once
4. test_calls_load_detector_for_pt_file — load_detector called with model_type="mdv5" for .pt file
5. test_calls_load_detector_for_onnx_file — load_detector called with "onnx" for .onnx detector file. Create a separate detector_onnx path for this.
6. test_calls_detect — detection.detect is called once
7. test_calls_parse_detections — detection.parse_detections is called once
8. test_calls_load_classifier — classification.load_classifier is called once
9. test_calls_classify — classification.classify is called once
10. test_single_classification_called_by_default — single_classification called when sequence=False
11. test_sequence_classification_called_when_flag_set — sequence_classification called when sequence=True
12. test_detect_only_returns_early — when detect_only=True, returns detections df without calling classifier
13. test_sort_calls_export_folders — when sort=True, export.export_folders is called
14. test_sort_false_does_not_call_export — when sort=False, export.export_folders not called
15. test_visualize_calls_plot_bounding_boxes — when visualize=True, visualization.plot_all_bounding_boxes called
16. test_visualize_false_does_not_plot — when visualize=False, plot function not called
17. test_save_data_called — file_management.save_data is called once with the final manifest
18. test_uses_cached_detections_when_check_file_true — when check_file returns True for detections, load_data is called and detect is NOT called
19. test_missing_image_dir_raises — patch WorkingDirectory to raise FileNotFoundError, assert it propagates

from_config tests — class TestFromConfig

Create a helper `_write_config(tmp...

This pull request was created from Copilot chat.