Scalable Automatic Machine Learning with H2O
summary
H2O AutoML runs a distributed Java cluster accessed from R/Python to train many models in parallel using random grid search over GBMs, XGBoost, GLMs, random forests and deep nets, then builds stacked ensemble meta-learners (an all-models ensemble and a lightweight best-per-algo ensemble). User supplies data, response and a runtime or max-models limit (default 1 hour); output is a cross-validated leaderboard and exportable Java model. It scales to big data, usually improves performance, and has caveats: deep learning non‑determinism on multicore and careful memory sizing.
AI-written summary · about
chapters (23)
- 0:00Intro
- 1:02What is H2O
- 2:18Overview
- 2:47About H2O
- 6:07Machine Learning
- 7:43H2O Library
- 8:29Automated Machine Learning
- 9:32Data Preprocessing
- 11:39What is AutoML
- 13:19H2O AutoML
- 15:19Why AutoML
- 17:00Stacked Ensembles
- 19:27AutoML Overview
- 22:19AutoML Interface
- 25:11H2O Flow
- 26:24Leaderboard
- 29:26Tips
- 30:18Exclude Algos
- 34:16Memory
- 37:00Roadmap
- 37:33Code
- 39:59Customization
- 41:01Benchmarking
related talks
30mScalable Automatic Machine Learning with H2O
01h 18mIntroduction to H20 AutoML with Python
01h 10mIntroduction to Machine Learning with H2O and Python
50mScalable Data Science & Deep Learning with H2O
27mDeploying distributed AI and machine learning in financial services
20mMulti-algorithm Ensemble Learning at Scale: Software...
01h 26mMachine Learning Algorithms with H2o and PowerShell
43mMachine Learning with R