Effector: A Python package for regional explanations

2024-04-03

arXiv.org

AI Summary

Existing global feature effect methods—such as Partial Dependence Plots (PDP) and Accumulated Local Effects (ALE)—often yield misleading interpretations under strong feature interactions. To address this, we propose a novel interpretability paradigm for tabular data that adaptively partitions the feature space into low-interaction subspaces and computes high-fidelity local feature effects within each region, balancing accuracy and human interpretability. Our key contributions are threefold: (1) a unified, efficient API for computing region-specific feature effects; (2) a modular, open-source Python package compatible with scikit-learn and PyTorch, integrating PDP, ALE, and our novel regional algorithm; and (3) substantially improved fidelity and practical utility in explaining complex black-box models. The framework has gained broad adoption in the machine learning community and is accompanied by comprehensive documentation and tutorials.

Technology Category

• Machine Learning: Transparent, Interpretable, Explainable ML
• Humans and AI: Explainable AI (XAI) for Human Understanding
• Computer Vision: Interpretability, Explainability, and Transparency

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

Effector is a Python package for interpreting machine learning (ML) models that are trained on tabular data through global and regional feature effects. Global effects, like Partial Dependence Plot (PDP) and Accumulated Local Effects (ALE), are widely used for explaining tabular ML models due to their simplicity -- each feature's average influence on the prediction is summarized by a single 1D plot. However, when features are interacting, global effects can be misleading. Regional effects address this by partitioning the input space into disjoint subregions with minimal interactions within each and computing a separate regional effect per subspace. Regional effects are then visualized by a set of 1D plots per feature. Effector provides efficient implementations of state-of-the-art global and regional feature effects methods under a unified API. The package integrates seamlessly with major ML libraries like scikit-learn and PyTorch. It is designed to be modular and extensible, and comes with comprehensive documentation and tutorials. Effector is an open-source project publicly available on Github at https://github.com/givasile/effector.

Problem

• Interpreting machine learning models for tabular data
• Addressing feature interaction misleading global effects
• Providing regional explanations through input space partitioning

Innovation

• Regional effects partition input space into subregions
• Effector provides unified API for feature effects
• Package integrates with scikit-learn and PyTorch