Skip to contents

Changelog

Source: .github/py_changelog.md

All notable changes to the Python pyshapr library will be documented in this file. Changes to the underlying shapr R package is documented in shapr CHANGELOG

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

[0.5.1] - 2026-06-24

Fixed

  • Constrained numpy<2.5 to keep pip install pyshapr resolvable on Python 3.12 and 3.13. No numba release yet supports numpy 2.5, so an unconstrained install (via the transitive shap -> numba dependency) backtracked to an ancient, incompatible numba that fails to build. The cap may be lifted once numba supports numpy 2.5.

[0.5.0] - 2026-06-23

Changed

  • Renamed the package from shaprpy to pyshapr (both the PyPI distribution name and the import name). Update installs to pip install pyshapr and imports to import pyshapr. This is the first release published under the pyshapr name. The previous shaprpy package remains available on PyPI for a transition period as a thin compatibility shim (released as shaprpy 0.4.4) that depends on and forwards to pyshapr.

[0.4.4] - 2026-06-23

Added

  • Added basic snapshot tests for approach="arf" on both numerical and mixed categorical feature sets.
  • Added an arf/vaeac end-to-end example script for numerical and mixed categorical feature sets.

Changed

  • Updated documentation to explicitly list arf and vaeac as supported approaches.
  • Updated local vaeac snapshot tests to pass vaeac_* arguments directly to explain().
  • Made the python/ package pass ruff check and ruff format.

Fixed

  • Fixed minor ruff lint issues in the source code.

Removed

  • Dropped support for Python 3.10. The minimum supported version is now Python 3.11.

[0.4.3] - 2026-01-24

Fixed

  • Fixed compatibility with pandas 3.0 for categorical data handling in tests.
  • Fixed XGBoost model support when feature_names_in_ attribute is not available.

[0.4.2] - 2026-01-23

Fixed

  • Adding predictor names to DMatrix when using XGBoost models to avoid issues with missing feature names.

[0.4.1] - 2025-12-22

Added

  • NEW: ShaprSummary class for enhanced summary functionality
    • Dedicated class for handling summary results with improved display and access methods
    • Support for tab-completion in IPython/Jupyter environments
  • Adjusted the unit tests to cover the new ShaprSummary class functionality

Changed

  • Update class_exploration.py example to demonstrate the new ShaprSummary class

[0.4.0] - 2025-11-28

Added

  • NEW: Shapr class for exploration and analysis of explanation results
    • Object-oriented interface for working with Shapley value explanations
    • Methods for extracting results: get_results(), get_explanation_dict(), get_r_object()
    • Summary and printing functionality: summary(), print()
    • Integration with SHAP library via to_shap() method for plotting
  • Enhanced example scripts demonstrating new class-based functionality
  • Added comprehensive unit and snapshot tests covering all approaches and full functionality, including categorical features and causal ordering

Changed

  • BREAKING: explain() function now returns a Shapr object instead of a plain dictionary
  • Restructured internal module organization with private _explain.py and _rutils.py modules
  • Updated all example scripts, tests and README to demonstrate new class-based API
  • Improved lazy loading for R package dependencies
  • Cleaned up unused library components

Fixed

  • Enhanced R package import reliability by removing empty paths from library locations, effectively resolving confusing warnings from R on package load
  • Fixed a bug in handling categorical features (R factors) during conversion to Python
  • Fixed an issue with passing boolean vectors for the confounding argument

[0.3.0] - 2025-08-27

Added

  • Publish shaprpy on PyPI
  • Restructured the shaprpy package for improved modularity and maintainability
  • Test suite with snapshot testing for output validation using syrupy
  • GitHub Actions CI/CD pipeline for automated testing across Python versions
  • Input validation tests for error handling

Changed

  • BREAKING: Minimum Python version requirement increased from 3.8+ to 3.10+
  • Updated package classifiers to reflect Python 3.10+ requirement

Dependencies

  • Added tabulate>=0.8.10 dependency for markdown table formatting in DataFrame comparisons

Documentation

  • Added Python 3.10+ requirement notice in README
  • Improved installation instructions
  • Added this changelog for better version tracking

[0.2.1] - Previous Release

Known Issues

  • Limited testing coverage
  • Some numerical precision issues in certain environments
  • Python 3.8-3.9 compatibility issues with newer type annotations