Katib 101: HyperParameter Tuning via Jupyter Notebook, Kale & Katib
Working with Katib allows you to perform hyperparameter tuning to improve models built with Kale in Jupyter Notebooks. In this module, we will prepare you to define Katib experiments using Kubeflow pipelines, run multiple experiments in parallel and and interpret the results to identify the ideal model.
Because it’s the simplest way to get started, we will use MiniKF as our Kubeflow environment. We will teach you how to build on this functionality to define and run multiple Katib experiments. This does not require any specialized knowledge of Kubernetes. Instead, we’ll use the open-source Kale JupyterLab extension.
We are assuming that you know how to organize and annotate cells in a Jupyter Notebook to define a Kubeflow pipeline that will run on a Kubernetes cluster. If you need a refresher on these skills, please review our Kale 101 course