Interpretability on AWS

You can use Jupyter instances that are managed by Amazon SageMaker AI to easily install Python modules through Conda and pip. For information about Python packages for SHAP and integrated gradient-based methods, see the Resources section. For smaller jobs and local testing on a SageMaker AI Jupyter instance, using the methods from these Python packages might be sufficient. If you are using a SageMaker AI managed model, SageMaker AI Clarify provides convenience methods for launching Kernel SHAP on a dedicated instance, and offloading the computation while a model developer continues to work on their Jupyter instance. For more information, see Create Feature Attribute Baselines and Explainability Reports in the SageMaker AI documentation.