xai-benchmark 0.1.0

Benchmark for Explainable AI methods

xai-benchmark

Public and extensible benchmark for XAI methods

Description

XAIB is an open benchmark that provides a way to compare different XAI methods using broad set of metrics that aimed to measure different aspects of interpretability

Installation

git clone https://github.com/Oxid15/xai-benchmark.git
cd xai-benchmark
pip install .

Then you can verify your installation by importing:

import xaib
print(xaib.__version__)

Usage

To test your own XAI method, you need to wrap it into Explainer interface (for examples of doing that see explainers folder) and pass it into explainers dict of evaluation notebook. For example for feature importance methods go to evaluation/feature_importance/feature_importance.ipynb.

Results

Metric values of all tested algorithms and baselines can be found in evaluation folder in the folder corresponding to the type of method for example for feature importance methods you can go to evaluation/feature_importance/feature_importance.ipynb and found metric values there.