Pyreal: A Framework for Interpretable ML Explanations (2312.13084v1)

Abstract: Users in many domains use ML predictions to help them make decisions. Effective ML-based decision-making often requires explanations of ML models and their predictions. While there are many algorithms that explain models, generating explanations in a format that is comprehensible and useful to decision-makers is a nontrivial task that can require extensive development overhead. We developed Pyreal, a highly extensible system with a corresponding Python implementation for generating a variety of interpretable ML explanations. Pyreal converts data and explanations between the feature spaces expected by the model, relevant explanation algorithms, and human users, allowing users to generate interpretable explanations in a low-code manner. Our studies demonstrate that Pyreal generates more useful explanations than existing systems while remaining both easy-to-use and efficient.

Introduction

Machine learning (ML) is increasingly being utilized across various domains to assist in decision-making. While a multitude of algorithms exist to explain ML models, the challenge remains to present these explanations in a form that is accessible and meaningful to decision-makers who may not possess a technical background. To address this, Pyreal has been introduced, offering a system to generate interpretable ML explanations in a user-friendly manner.

Approach to Interpretable Explanations

Pyreal converts data and explanations to align with what is expected by the model, relevant explanation algorithms, and ultimately in a format that is comprehensible to humans. This Python-based system enables the production of varied interpretable explanations with minimal coding, thereby bypassing extensive development work typically required to transform explanations into intelligible states.

System Flexibility and Extensibility

The Pyreal framework is designed for flexibility and extensibility. It supports various models and data transformations and can easily integrate new explanation methods. Its modular design allows for the straightforward addition of new components, supporting an extensive array of scenarios and ensuring its long-term utility. The framework operates with tabular data which is commonly found in real-world scenarios, thereby emphasizing its practicality.

Evaluation and Real-World Application

User studies indicate that Pyreal provides more useful explanations than existing systems, and it achieves this with relative efficiency and ease of use. The added time for generating interpretable explanations is deemed minimal, and the system’s high-level API proves advantageous in real-world case studies like child welfare screening and wind turbine monitoring. Pyreal's focus on developing a human-centered approach, through interpretable explanations, demonstrates a significant advantage over conventional systems.

By employing a system of explanation transforms and a structured explanation pipeline, Pyreal ushers in a methodical revolution for ML explanations. Its contributions to the field are underscored by the potential it presents for non-technical users to engage with and benefit from complex ML predictions. The Pyreal system stands as a noteworthy stride towards making the sophisticated algorithms of ML more accessible and actionable in setting groundwork for decision support across various industries.