Gating Methods

Abstract

A gating method, as defined in this text, refers to an overseer that uses one or more gate variables to weight the opinions of two or more models in order to obtain an opinion superior to what could be obtained from a single model. These models may be numeric predictors or classifiers. What distinguishes the techniques of this chapter from the model combination methods described in prior chapters is that a gated combiner requires the use of a dedicated variable or variables to control the combination process, while the previously discussed techniques operate without the benefit of such outside help.