Data Science with H2O.ai: An Introduction to Machine Learning and Predictive Modeling
- Predictive AI
- H2O Driverless AI
Overview
In this session, H2O senior solutions engineer and data scientist John Farland introduces core machine learning concepts, including the standard data science lifecycle and the differences between supervised, unsupervised, and reinforcement learning. He then walks through H2O's platform, from its open source H2O-3 roots to its enterprise AI Cloud, built around the make, operate, and innovate principles. The video closes with a live demonstration of Driverless AI, training a regression model to predict customer lifetime value for an auto insurance dataset.
Key takeaways
- The five-step data science project lifecycle: gathering data, feature engineering, model training, evaluation, and deployment, plus the importance of monitoring for drift after deployment
- The differences between supervised learning (regression and classification), unsupervised learning (clustering, anomaly detection, dimensionality reduction), and reinforcement learning, with examples of each
- Common algorithms for regression and classification problems, including GLM, GAM, gradient boosting, random forest, and neural networks, and the error metrics used to evaluate them, such as RMSE, MAE, AUC, and F1
- How H2O-3's open source library supports data preprocessing, cross-validation, and model explainability tools like partial dependence plots and Shapley values
- A live Driverless AI walkthrough: importing and visualizing a customer lifetime value dataset with AutoViz, splitting train and test data, and running a supervised regression experiment
- How to interpret Driverless AI's scoring pipeline, download an AutoDoc report, and review feature importance for the final stacked ensemble model
Start owning your AI