CausalBO: A Python Package for Causal Bayesian Optimization

• Published in: Proceedings of IEEE Southeastcon pp. 1370 - 1375
• Main Authors: Roberts, Jeremy, Javidian, Mohammad Ali
• Format: Conference Proceeding
• Language: English
• Published: IEEE 15.03.2024
• Subjects: Bayes methods; Bayesian optimization; BoTorch; causality; Cause effect analysis; DoWhy; Gaussian processing; Interconnected systems; Libraries; Medical services; Optimization; Usability
• ISSN: 1558-058X
• DOI: 10.1109/SoutheastCon52093.2024.10500062

This paper introduces CausalBO, a Python package developed to enhance the applicability and utility of the Causal Bayesian Optimization (CBO) algorithm. The original CBO algorithm, developed by Virginia Aglietti et al.[1], integrated causality into Bayesian optimization to address its limitations in complex, interconnected systems like healthcare. However, the initial implementation had several drawbacks, such as specificity to certain datasets, non-modularity, and complex setup require-ments. CausalBO addresses these issues by offering increased readability and usability, integration with widely used libraries like BoTorch and DoWhy, and simplification of the CBO loop. It is designed to be user-friendly, catering to those with limited knowledge of causality and do-calculus. The effectiveness of CausalBO is demonstrated through case studies, including a synthetic experiment and a healthcare scenario, showcasing its applicability and versatility.