PRTools Contents

PRTools User Guide



Convert classifier to normalized classifier (yielding confidences)

   V = CLASSC(W)
   D = CLASSC(A*W)
   D = CLASSC(A,W)

 W Trained or untrained classifier
 A Dataset

 V Normalized classifier producing confidences instead of  densities or distances (after training if W is untrained)


The trained or untrained classifier W may yield densities or unnormalised  confidences. The latter holds for two-class discriminants like FISHERC and SVC as well as for neural networks. Such classifiers use or should  use CNORMC to convert distances to confidences. In multi-class problems  as well as in combining schemes they do not produce normalised  confidences. These outcomes, like the density outcomes of classifiers  liek QDC, LDC and PARZENC, can be converted by CLASSC into confidences
the sum of the outcomes will be one for every object.

In case W is a one-dimensional mapping, it is converted into a two-class  classifier, provided that during the construction a class label was  supplied. If not, the mapping cannot be converted and an error is  generated.

CLASSC lists the outcomes on the screen in case no output argument is  supplied. Also true and estimated labels are supplied.

See also

mappings, datasets, cnormc, labeld,

PRTools Contents

PRTools User Guide

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