A Practical Introduction to Ensemble Machine Learning

Ensemble machine learning methods combine the power of several different algorithms into a single algorithm with the aim of achieving higher predictive power than any of the constituent algorithms.  Practitioners may prefer ensemble algorithms when model performance is valued above other factors such as model complexity andtraining time.  In this talk, an overview of ensemble learning and ensemble software will be followed by a laptop-based demo and an Amazon EC2 demo, the latter of which performs the training in a distributed fashion on a cluster.  We will discuss three R packages for ensemble learning: SuperLearner, subsemble and h2oEnsemble.  TheH2O Ensemble project, although currently only available in R, will soon be available in Java, Scala and Python as part of the open source machine learning platform, H2O.  http://h2o.ai