DisTools Contents

DisTools User Guide

testkd

TESTKD

Test k-NN classifier for dissimilarity data

    [E,C] = TESTKD(D,K,PAR)
    [E,C] = D*TESTKD([],K,PAR)

Input
 D Dissimilarity dataset. Object labels are assumed to be the  true labels. Feature labels are assumed to be the labels of  the objects they are related to.
 K Desired number of neighbors to take into account, default K = 1.
 PAR 'LOO' - leave-one-out option. This should be used if  the objects are related to themselves. If D is not square,  it is assumed that the first sets of objects in columns and  rows match.
 'ALL' use all objects (default).

Output
 E Estimated error
 C Dataset with confidences, size M x N, if D has size M x L and  the labels are given for N classes. Note that for K < 3 these  confidences are derived from the nearest neigbor distances  and that for K >= 3 they are the Bayes estimators of the  neighborhood class probabilities. D*TESTC returns E.

Description

TESTKD is based on just counting errors and does not weight with class  class priors stored in D. Use D*(DL*KNNDC)*TESTC if this is needed.  DL is the dissimilarity matrix of the representation objects used for D.

See also

datasets, knndc, nne, nnerror1, nnerror2,

DisTools Contents

DisTools User Guide

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