Driverless AI: Diagnostics and the Scoring Pipeline
- Predictive AI
- H2O Driverless AI
This lesson walks through diagnosing a completed Driverless AI experiment against a new test dataset, opening the Diagnose Model button to generate scores, prediction downloads, and metric plots such as the ROC curve, lift chart, and confusion matrix. It then covers the Visualize Scoring Pipeline view, which shows how original features feed into two XGBoost models trained with four fold cross validation before their outputs are combined through stacking into a final prediction. The lesson also shows how drilling into a single model reveals the transformers applied to each feature and the decision criteria the model uses.
Key takeaways
- How the Diagnose Model button scores a completed experiment against a brand new test dataset
- What metric plots such as the ROC curve, lift chart, and confusion matrix reveal about model performance
- How stacking combines two cross validated XGBoost models into one final prediction pipeline
- Why drilling into a single model shows the transformers applied to each original feature
Start owning your AI