Changelog

All notable changes to FADVI will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Released]

Interpretability Analysis: Feature attribution support using Captum library
- Integrated Gradients (IG) method for feature importance analysis
- GradientShap (GS) method as alternative attribution approach
- Automatic feature ranking with statistical summaries
- Support for both batch and label prediction interpretability
- Integration with predict() method via interpretability parameter
- Customizable attribution method parameters
Enhanced Prediction Interface:
- get_ranked_features() method for automatic feature ranking
- Automatic integration of interpretability with prediction workflow
- Flexible return formats (dict/tuple) via return_dict parameter
- Comprehensive attribution result DataFrames with statistical summaries
Documentation Improvements:
- Advanced usage tutorial with interpretability examples
- Complete code examples for attribution methods
- Performance optimization guidelines
- Visualization and analysis workflows
- Error handling best practices
Test Suite Enhancements:
- Dedicated interpretability test module (test_interpretability.py)
- Comprehensive testing for both IG and GS methods
- Test coverage for error handling and edge cases
- Parameter validation and consistency testing
- Integration tests for prediction + interpretability workflows

Code Organization:
- Reorganized test suite with clear separation of concerns
- test_fadvi_predict.py now focuses solely on basic prediction testing
- All interpretability functionality consolidated in test_interpretability.py
- Improved test discoverability with pytest-compatible naming
API Consistency:
- Standardized return formats across prediction methods
- Enhanced parameter validation and error messages
- Backward-compatible API with new optional parameters
Performance:
- Memory-efficient batch processing for interpretability analysis
- Optimized attribution computation for large datasets
- Configurable batch sizes for different hardware configurations