PRTools contents

PRTools manual

maxc

MAXC

Maximum combining classifier

    W = MAXC(V)
    W = V*MAXC

Input
 V Stacked set of classifiers

Output
 W Combined classifier using max-rule

Description

If V = [V1,V2,V3, ... ] is a set of classifiers trained on the same  classes, then W is the maximum combiner: it selects the class that gives  the maximal output of the input classifiers. This might also be used as  A*[V1,V2,V3]*MAXC in which A is a dataset to be classified. Consequently,  if S is a similarity matrix with class feature labels (e.g. S = A*PROXM(A,'r')), then S*MAXC*LABELD is the nearest neighbor classifier.

If it is desired to operate on posterior probabilities then the input  classifiers should be extended to output these, using V = V*CLASSC.

The base classifiers may be combined in a stacked way (operating in the  same feature space by V = [V1,V2,V3, ... ] or in a parallel way (operating  in different feature spaces) by V = [V1;V2;V3; ... ].

Example(s)

prex_combining,

See also

mappings, datasets, votec, minc, meanc, medianc, prodc, averagec, stacked, parallel,

PRTools contents

PRTools manual