H2O-3 Open Source, Distributed Machine Learning for Everyone at H2O World Sydney 2022
- Predictive AI
- H2O-3
- H2O Driverless AI
Overview
Recorded at H2O World Sydney 2022, this session gives a tour of H2O-3, H2O's open source distributed machine learning platform, presented by a quadruple Kaggle Grandmaster on the H2O team. It covers how H2O-3 stores and processes data as H2O Frames, the range of supervised and unsupervised algorithms it supports, and how H2O AutoML automates model selection, tuning, and ensembling. The talk closes with a live code walkthrough and audience Q&A on deployment, explainability, and federated data.
Key takeaways
- How H2O-3 works as a distributed, multi-node, multi-cluster platform, with its core written in Java and exposed through Python, R, Scala, and a web GUI called Flow
- How data is structured as H2O Frames, distributed column vectors that support pandas- and data.table-style row and column operations
- The range of supported algorithms, from GLM, GAM, SVM, and Naive Bayes to tree-based models, deep learning MLPs, stacked ensembles, and unsupervised methods like k-means, PCA, GLRM, and isolation forest
- How H2O AutoML automates preprocessing, trains many models, and builds a stacked ensemble, producing a leaderboard to compare results
- How to run a full AutoML experiment in a few lines of code and generate model explainability outputs, such as variable importance, partial dependence plots, and Shapley values, with a single command
- How trained models export as POJO or MOJO artifacts for deployment across environments like Spark, Snowflake, and H2O MLOps
Start owning your AI