Tutorial: From Notebook to Kubeflow Pipelines to KFServing: the Data Science Odyssey

Karl Weinmeister, Stefano Fioravanzo at KubeCon + CloudNativeCon North America 2020

A hands-on lab driven tutorial to show Data Scientists and ML Engineers alike how to turbocharge your Kubeflow efforts. In this session you will learn how to quickly build, tune, and execute complex Kubeflow workflows - as well as how to work faster using Kale to automate much of your work. Learn how to rapidly automate Kubeflow: - Deploy a Jupyter Notebook as a Kubeflow pipeline using Kale - Optimize your model training using Katib for hyperparameter tuning - Serve your model with KFServing - Run thousands of runs with caching and garbage collection - Track and reproduce pipeline steps along with their state and artifacts Data Scientists benefit from an intuitive GUI that automates and hides all of the underlying infrastructure and SDK requirements. ML Engineers can use the reproducible, automated workflows as a scaffold to quickly move to even more advanced tuning and model building.