by: Oege Dijk
This repository is a demonstration of how to deploy explainable machine learning models (using SHAP) to AWS cloud infrastructure.
Three approaches are shown:
- Using Sagemaker. This involves setting up custom training containers on ECR,
and defining the proper inference functions.
- All notebooks, code, configurations, Dockerfiles and READMEs can be found in the sagemaker folder.
- Using straight AWS Lambda functions and zappa. This is easier, but still a
number of tricky things to get to work (such as deploying from inside a lambda
compatible docker container)
- All code, Makefiles and READMEs can be found in the lambda folder.
- For completeness an example of a local on premise deployment can be in the local folder.
An example dashboard that sends requests to both a sagemaker deployment and a lambda deployment is running at http://creditexplainer.herokuapp.com