What is OpenXAI?
OpenXAI is a general-purpose lightweight library that provides a comprehensive list of functions to systematically evaluate the reliability of post hoc explanation methods. The library provides implementations and easy-to-use APIs for various state-of-the-art explanation methods and evaluation metrics. It is also flexible enough to accommodate new datasets (both synthetic and real-world), explanation methods, and evaluation metrics.
Easy to Code
OpenXAI library is minimally dependent on external packages and can benchmark explanation methods with just 10 lines of code.
Easy to Evaluate
OpenXAI integrates a wide range of evaluation metrics, including faithfulness, stability, and fairness metrics.
Easy to Benchmark
OpenXAI provides an intuitive abstract template with dataloaders, trained models, and XAI-ready datasets to easily and reliably benchmark explaination methods.