T = templateEnsemble (Method, NLearn, Learners) returns a template for an ensemble grown by Method from NLearn learners, each fitted as Learners says. A template names a learner and the options it is to be fitted with, without fitting anything: it is given to fitcecoc, which grows one such ensemble for every binary learner it trains.

Method is one of the methods of fitcensemble or fitrensemble, in any letter case. NLearn is the number of learning cycles and Learners a learner name, such as 'tree', or a template of one, such as templateTree returns; they are the 'NumLearningCycles' and 'Learners' options of fitcensemble.

T = templateEnsemble (…, name, value) also stores the given options. They are the name-value arguments of fitcensemble, such as 'LearnRate'.

 T = templateEnsemble ('GentleBoost', 50, templateTree ('MaxNumSplits', 1));
 Mdl = fitcecoc (X, Y, 'Learners', T);

T is a structure carrying Method, Type, LearnerTemplates, NLearn and one field per option given. Type is 'regression' for LSBoost and 'classification' for every other method; a 'Type' option may choose it for Bag, which serves both.

Deviation from MATLAB

MATLAB returns an object of a class whose name we cannot use, which has no public properties and one method this package declines package wide, so a structure carries everything a user can observe. This is what ModelParameters already does throughout the package.

Only the method is checked here. The ensemble owns the list of options it takes and checks them, NLearn and Learners included, when the template is used, so a value it refuses is refused then rather than now. MATLAB checks 'LearnRate' already here.

See also: fitcecoc, fitcensemble, templateTree

Source Code: templateEnsemble