PRTools contents

PRTools manual

mclassm

MCLASSM

Computation of a combined, multi-class based mapping

   W = MCLASSM(A,MAPPING,MODE,PAR)

Input
 A Dataset
 MAPPING Untrained mapping
 MODE Combining mode (optional; default: 'weight')
 PAR Parameter needed for the combining

Output
 W Combined mapping

Description

If A is a unlabeled dataset or double matrix, it is converted to a  one-class dataset. For one-class datasets A, the mapping is computed,  calling the untrained MAPPING using the labeled samples of A only.  For multi-class datasets separate mappings are determined for each class  in A. They are combined as defined by MODE and PAR. The following  combining rules are supported
'weight': weight the mapping outcome for class j by PAR(j) and sum  over the classes. This is useful for densities in which case  PAR is typically the set of class priors (these are in fact  the defaults if MODE = 'weight').

'mean' combine by averaging.
'min' combine by the minimum rule.
'max' combine by the maximum rule.

This routine is only defined for datasets with crisp labels.

See also

datasets, mappings,

PRTools contents

PRTools manual