Reading the Interpretability Report in H2O Driverless AI
- Observability
- H2O Driverless AI
This lesson walks through the Interpretability Report, also called the MLI report, generated after a Driverless AI experiment completes. It covers running the report with default settings from the Interpret the model step, then reading the MLI Explanations page once the report status shows complete. The report is organized into three tabs, Summary, Surrogate Models, and DAI Models, with the DAI Models tab providing Feature Importance and Shapley plots along with Partial Dependence, Disparate Impact Analysis, Sensitivity Analysis, and Permutation Feature Importance plots for the trained model.
Key takeaways
- How to generate an MLI interpretability report from the Interpret the model step after an experiment completes
- What the Summary tab shows, an experiment overview and the dataset variables ranked by importance
- The difference between the Surrogate Models tab and the DAI Models tab in an MLI report
- What plots the DAI Models tab includes for classification and regression experiments, such as Shapley and partial dependence
- When the MLI report is fully ready to review, once zero explanations remain running
Start owning your AI