Bagging

Online Bagging

Nikunj C. Oza, Stuart J. Russell: Online Bagging and Boosting. AISTATS 2001

General Description

Bagging is an ensemble method that improves the accuracy of a single classifier. Non-streaming bagging builds a set of base models, training each model with a bootstrap sample created by drawing random samples with replacement from the original training set. Each base model's training set contains each of the original training example a number of times that follows a binomial distribution. This binomial distribution tends towards a Poisson(1) for large values. Thus, the online version of the Bagging classifier, instead of using sampling with replacement, gives each example a weight according to a Poisson(1) distribution.

Implementation

In StreamDM, the main algorithm is implemented in Bagging, which is controlled by the following options:

• Number of classifiers (-s), which sets the number of classifiers used to build the ensemble (10 by default); and
• Base classifier (-l), which sets the base classifier to be used to build the members of the ensemble (SGDLearner by default).