Tutorials

Serving a model as an endpoint is only the first step; managing many models in production is the real job. This page holds two end-to-end tutorials for that, first A/B serving on Kubernetes and then a full training-and-serving stack.

The same notes are in Tutorials.

The first tutorial is for the moment when simply serving up endpoints is not good enough any more. Kubernetes, scikit-learn, s2i, gcloud, and Seldon: random serving for A/B testing is Gus Cavanaugh's walkthrough of deploying a first A/B test with scikit-learn and Kubernetes, the start of model management.

The same notes are in A/B Testing.

The second tutorial widens model management from one deployment to the whole loop of training many models and shipping the best one. Polyaxon for training, an Argo package and deployment, and Seldon for serving is Daniel Rodriguez's version of that loop in Kubernetes, where model management means tooling and pipelines for data scientists to develop, deploy, measure, and improve models.