PRTools Contents

PRTools User Guide



Computation of a combined classifier according to adaboost.


 A Dataset
 CLASSF Untrained weak classifier
 N Number of classifiers to be trained, default 200
 RULE Combining rule (default: weighted voting)
 VERBOSE Suppress progress report if 0 (default)

 W Combined trained classifier
 V Cell array of all classifiers  Use VC = stacked(V) for combining
 ALF Weights


In total N weighted versions of the training set A are generated  iteratevely and used for the training of the specified classifier.  Weights, to be used for the probabilities of the objects in the training  set to be selected, are updated according to the Adaboost rule.

The generation of base classiifers may be stopped prematurely by PRTIME.

The entire set of generated classifiers is given in V.  The set of classifier weigths, according to Adaboost is returned in ALF

Various aggregating possibilities can be given in  the final parameter rule

 []: WVOTEC, weighted voting.
 VOTEC voting
 MEANC sum rule
 AVERAGEC averaging of coeffients (for linear combiners)
 PRODC product rule
 MAXC maximum rule
 MINC minimum rule
 MEDIANC median rule


Ji Zhu, Saharon Rosset, Hui Zhou and Trevor Hastie, Multiclass Adaboost. A multiclass generalisation of the Adaboost algorithm, based on a generalisation of the exponential loss.

See also

mappings, datasets, prtime,

PRTools Contents

PRTools User Guide

This file has been automatically generated. If badly readable, use the help-command in Matlab.