search

Trusted-AI / AIX360

Interpretability and explainability of data and machine learning models
https://aix360.res.ibm.com/
Apache License 2.0
1.61k stars 304 forks source link

Faithfulness metric on imags explanations #99

Open dsvrsec opened 4 years ago

dsvrsec commented 4 years ago

Can the faithfulness metric can be used as a metric for explanations obtained using LIME or SHAP.

sadhamanus commented 4 years ago

Yes. It can used for any feature based local explainability method.

Nadiatriki commented 2 years ago

@sadhamanus what dose it means when faithfulness_metric gave me 'nan' while using SHAP and LIME for explaining a Tabular data ? and thanks

sadhamanus commented 2 years ago

Hard to say exactly but the metric just computes correlation so either the feature importance are all the same or the prediction of your model hasn't changed when setting each feature to base value. In either of these cases it means all features are equally unimportant. Another possibility could be model outputting nans for different inputs it tries by setting each feature individually to base values.