Deployment Templates and Scoring in H2O MLOps
- Observability
- H2O MLOps
This lesson covers the deployment templates available in H2O MLOps once a model is finished. It walks through creating a deployment, choosing a single model or a competing AB test or Champion Challenger setup, and a development or production environment. It covers predictions only versus predictions plus reason codes, and shows downloading either a Python scoring pipeline or a Mojo, a combined object with feature engineering, predictions, and optional Shapley values, that runs locally or in the cloud in Java, Python, or R. It closes by pointing to a public repository for using a Mojo inside a database like Snowflake, Teradata, or Hive.
Key takeaways
- How to choose between a single model deployment and a competing setup like an AB test or a Champion Challenger in H2O MLOps
- What the choice between predictions only and predictions plus reason codes changes about a deployment's output
- What a Mojo bundles together: feature engineering, predictions, and optionally Shapley values, in one portable scoring object
- How a Mojo can run predictions locally or in the cloud, in Java, Python, or R, using either the UI or the API
- How to find examples for running a Mojo inside a database like Snowflake, Teradata, or Hive
Start owning your AI