search

oegedijk / explainerdashboard

Quickly build Explainable AI dashboards that show the inner workings of so-called "blackbox" machine learning models.
http://explainerdashboard.readthedocs.io
MIT License
2.3k stars 331 forks source link

skorch models raising The SHAP explanations do not sum up to the model's output #289

Open oegedijk opened 9 months ago

oegedijk commented 9 months ago

Seems related to this bug: https://github.com/shap/shap/issues/3363

Can be avoided with passing shap_kwargs=dict(check_additivity=False) to the explainer, but then you might get inaccurate shap values.

Added check_additivity=False param to the skorch tests for now.