H2O Driverless AI Demo
- Observability
- H2O Driverless AI
Overview
This hands on walkthrough demonstrates the full Driverless AI workflow using a credit card default dataset from Kaggle, from data import and an 80/20 train/test split through automated visualization, experiment setup, and model interpretation. It shows the accuracy, time, and interpretability dials that control an experiment, the real time model building metrics, and the resulting deployment options, including Python and Java scoring pipelines and one click deployment to platforms like AWS. The presenter also references a past Kaggle competition, BNP Paribas, where a similar run would have placed in the top 1 percent.
Key takeaways
- How to import a dataset and use Driverless AI's built in tool to split it into train and test sets, for example an 80/20 split
- How the Autoviz tool automatically generates visualizations like outlier plots and correlation heat maps before you build a model
- How the accuracy, time, and interpretability dials shape an experiment, and how to fine tune settings like which algorithms to include
- How to read real time experiment metrics, including precision and recall, lift and gains charts, KS charts, and resource usage
- How to interpret a finished model using feature importance, Shapley explanations, and surrogate models like decision trees
- The deployment options available after training: Python and Java (MOJO) scoring pipelines, one click deployment to platforms like AWS Lambda and EC2, and automatic documentation of the full experiment
Start owning your AI