Real Time Drift Monitoring in H2O MLOps
- Observability
- H2O MLOps
This lesson shows how to set up real time drift monitoring for a deployed model in H2O MLOps. It covers choosing which columns to track and setting baselines when a deployment starts scoring data, so basic drift information is visible directly from the MLOps UI. For more detailed tracking, it moves into the built in Apache Superset integration, selecting a database schema and deployment, and shows how the platform stores aggregated statistics, including bin edges, bin counts, and sums, for both numeric columns like age and categorical columns like region or education, so a user can build a custom dashboard from that data.
Key takeaways
- How choosing columns and setting baselines when a deployment goes live enables basic drift tracking directly in the MLOps UI
- What data H2O MLOps stores for deeper drift analysis: aggregated bin edges, bin counts, and sums per column
- How to use the built in Apache Superset integration to select a database schema and deployment for tracking
- How a user can build a custom dashboard from the stored aggregates for both numeric columns and categorical columns
Start owning your AI