Multi-algorithm Ensemble Learning at Scale: Software...
summary
Scalable stacking implementation called H2O Ensemble combines deep nets, GBMs, random forests and others into a metalearner and runs distributed across multi-node clusters (demonstrated on 96‑core and 320‑core setups) to cut wall time. Two algorithmic variants further reduce cost: Subsemble partitions data into J subsets to train on small chunks and reconstruct full predictions, and Online Super Learner uses Vowpal Wabbit to update base models and the meta‑learner incrementally. Benchmarks show large speedups with minimal accuracy loss.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:21Agenda
- 2:09Common Types of Ensemble Methods
- 4:20Why Ensembles
- 6:26History of Stacking
- 9:16The Super Learner Algorithm
- 11:50Scalable Stacking Solutions
- 14:16Modified K-fold CV in Subsemble
- 16:35The Subsemble Algorithm
- 17:31Online Super Learner
- 18:41H20 Overview
- 19:10H2O Ensemble Overview
- 20:04H2O Ensemble Resources
related talks
44mScalable Automatic Machine Learning with H2O
44mEnsemble Techniques for High Performance Machine Learning
30mScalable Automatic Machine Learning with H2O
01h 10mIntroduction to Machine Learning with H2O and Python
42mDeep Learning is Like Water (it's Everywhere!)
50mScalable Data Science & Deep Learning with H2O
31mEnsemble X Your personal strataGEM to build Ensemble
27mDeploying distributed AI and machine learning in financial services