amazon-sagemakeramazon-machine-learning

How to Enable SageMaker Debugger in the SageMaker AutoPilot

I'd like to (a) plot SHAP values out of the SageMaker (b) AutoML pipeline. To achieve (a), debugger shall be used according to: https://aws.amazon.com/blogs/machine-learning/ml-explainability-with-amazon-sagemaker-debugger/.

But how to enable the debug model in the AutoPilot without hacking into the background?

Solution

  • SageMaker Autopilot doesn't support SageMaker Debugger out of the box currently (as of Dec 2020). You can hack the Hyperparameter Tuning job to pass in a debug parameter.

    However, there is a way to use SHAP with Autopilot models. Take a look at this blog post explaining how to use SHAP with SageMaker Autopilot: https://aws.amazon.com/blogs/machine-learning/explaining-amazon-sagemaker-autopilot-models-with-shap/.