striders 0.0.4

Efficient surrogate-based model explanations (XAI) using landmark-based kernel approximations for scalable SHAP values.

Striders is a lightning-fast, surrogate-based model explanations (XAI). It provides an efficient alternative to traditional SHAP by leveraging landmark-based kernel approximations. Striders implements a landmark-based approximation of the Shapley Kernel. By selecting representative landmarks, it reduces the complexity of the explanation process while maintaining high correlation with the true Shapley values.

Installation

pip install striders

Performance Benchmarking

Dataset (Task)	Samples / Features	Metric	TreeSHAP	Striders	Speed-up
CA Housing (Reg.)	20,640 / 8	Execution Time	6.4243s	0.0784s	82.0x 🚀
		Fidelity ($R^2$)	-	0.9093
		Correlation	-	0.9506
Credit Default (Clf.)	30,000 / 23	Execution Time	13.7760s	0.4027s	34.2x 🚀
		Fidelity ($R^2$)	-	0.9776
		Correlation	-	0.9428

Reproducibility: You can reproduce these results by running the script.

Acknowledgments & Citations

This is an unofficial implementation based on the principles described in:

@article{ko2025stride,
  title={STRIDE: Subset-Free Functional Decomposition for XAI in Tabular Settings},
  author={Ko, Chaeyun},
  journal={arXiv preprint arXiv:2509.09070},
  year={2025}
}

If you find this implementation useful in your work, please consider citing this repository:

@software{striders2026,
  author={RektPunk},
  title={Striders: A High-Performance Rust-based Implementation of STRIDE},
  year={2026},
  url={https://github.com/RektPunk/striders},
}