This repository builds a Session Based Recommender. The trained ML Model will be served by KFServing. The Metadata will be tracked by MLflow.
The Dataset can be found here: https://www.kaggle.com/mkechinov/ecommerce-behavior-data-from-multi-category-store
Related Blog Post: https://www.novatec-gmbh.de/blog/ausrollen-und-betreiben-von-ml-modellen-in-produktion-mit-kfserving/
- Installation of KFServing as part of Kubeflow or Standalone
- Install Requirements from requirements
- Start a MLflow-Server. If you are new to MLflow: https://github.com/felix-exel/mlflow
- Download any Dataset from https://www.kaggle.com/mkechinov/ecommerce-behavior-data-from-multi-category-store
- Train the session based recommender model by running the notebook session_based_recommender_with_mlflow
- If your MLflow Artifact Storage does not point to a Cloud Store, you need to upload your model files manually for an easy deployment with KFServing.
- Alternatively you can mount your model by using Persistent Volumes which isn't covered by this repository
- Write your Credentials in aws-secret_serviceaccount
kubectl apply -f aws-secret_serviceaccount.yaml- Change the StorageUri in tf-deployment-recommender to your Cloud Path
kubectl apply -f tf-deployment-recommender.yaml- or deploy through the notebook deploy-kfserving-recommender-service