H2O Driverless AI Demo
- Observability
- H2O Driverless AI
Overview
This demo walks through H2O Driverless AI, an automated machine learning platform for supervised learning problems that runs in the cloud or on premise. It covers the full workflow: connecting to a data source, exploring and visualizing a dataset, splitting data for training and holdout testing, configuring and launching an AutoML experiment, and reviewing the resulting models. The demo also covers the machine learning interpretability module and options for deploying a finished model into production.
Key takeaways
- How to connect to data sources, including a JDBC connector for SQL data warehouses, and review column-level details and distributions before modeling
- How the visualize step surfaces interesting correlations and outliers across your dataset automatically
- How to split data for training and holdout testing, including stratified splitting for rare events and time-based splitting
- How the accuracy, time, and interpretability settings control experiment complexity, feature engineering, and cross-validation
- How the machine learning interpretability module uses SHAP values, partial dependence plots, and disparate impact analysis to explain and check a model for bias
- How to productionize a finished model using the Python scoring pipeline, the MOJO for Java, C++, or other environments, or one-click deployment to Amazon Lambda or a local REST server
Start owning your AI