REJECTC

Construction of a rejecting classifier

    WR = REJECTC(A,W,FRAC,TYPE)

Input
A	Dataset
W	Trained or untrained classifier
FRAC	Fraction to be rejected. Default: 0.05
TYPE	String with reject type: 'ambiguity' or 'outlier'.  'a' and 'o' are supported as well. Default is 'a'.

Output
WR	Rejecting classifier

Example(s)

 a = gendatb
 w = ldc(a);
 v = rejectc(a,w,0.2);
 scatterd(a);
 plotc(w);
 plotc(v,'r')

Description

This command extends an arbitrary classifier with a reject option. If WR is used for classifying a dataset B, then D = B*WR has C+1 columns  ('features'), one for every class in A and an additional one that takes  care of the rejection: a NaN for numeric labels (classnames in A) or en  empty string for string labels.  NOTE: Objects that are rejected are not counted as an error in TESTC. The  classification error estimated by TESTC just considers the total number  of objects for wich B*WR*LABELD has a correct classname and neglects all  others. So by rejection the error estimate by TESTC may increase,  decrease or stay equal.

See also

datasets, mappings, labeld, testc, rejectm,