Adding Custom Scorer Recipes in Driverless AI
- Predictive AI
- H2O Driverless AI
This lesson in the Driverless AI Starter Course shows how to add a custom scoring recipe when the built in metrics do not cover what is needed. It walks through the Add Custom Recipes option in Expert Settings, browsing the official h2oai driverlessai-recipes GitHub repository for a binary classification scorer, and pasting the raw script URL for the average_mcc scorer into Driverless AI. The lesson also covers enabling the new scorer under the Validation tab, and closes by explaining the difference between Launch Experiment, which runs one experiment, and Create Leaderboard, which runs several for comparison.
Key takeaways
- How to upload a custom scorer recipe into Driverless AI using the Add Custom Recipes option in Expert Settings
- The location of official scorer recipes in the h2oai driverlessai-recipes GitHub repository, organized by problem type
- How to enable a newly uploaded scorer, such as average_mcc, under the Validation tab so it appears in the Scorer list
- The difference between Launch Experiment, which runs one experiment, and Create Leaderboard, which runs several for comparison
Start owning your AI