PytorchWrapper class

Wrapper class for PyTorch Lite operations.

This class provides a testable interface around the static methods of PytorchLite, allowing for dependency injection and easier testing. It wraps the core functionality for loading classification models while maintaining the same API surface.

Purpose

The wrapper serves several important purposes:

• Testability: Enables mocking of PyTorch operations in unit tests
• Dependency Injection: Allows services to depend on an interface rather than static methods
• Future Extensibility: Provides a place to add logging, error handling, or caching
• API Consistency: Maintains a consistent interface even if underlying implementation changes

Usage Example

final wrapper = PytorchWrapper();
final model = await wrapper.loadClassificationModel(
  'assets/models/mosquito_classifier.pt',
  224,
  224,
  labelPath: 'assets/labels/mosquito_species.txt',
);

Testing

For unit tests, this class can be easily mocked:

class MockPytorchWrapper extends Mock implements PytorchWrapper {}

test('classification service loads model', () async {
  final mockWrapper = MockPytorchWrapper();
  final mockModel = MockClassificationModel();
  
  when(mockWrapper.loadClassificationModel(any, any, any, labelPath: any))
      .thenAnswer((_) async => mockModel);
  
  final service = ClassificationService(pytorchWrapper: mockWrapper);
  await service.loadModel();
  
  expect(service.isModelLoaded, isTrue);
});

See also:

• PytorchLite for the underlying PyTorch Lite implementation
• ClassificationService which uses this wrapper
• ClassificationModel for the model interface