SHIFTOP

Shift operating point of classifier

    V = SHIFTOP(D,E,C)
    V = D*SHIFTOP(E,C);
    V = D*SHIFTOP(TYPE,C);
    W = A*(U*SHIFTOP(E,C));

Input
D	Dataset, classification matrix (two classes only). D = A*U.
E	Desired error class C for D*TESTC
TYPE	Minimum error TYPE
C	Index of desired class (default: C = 1)
A	Dataset used to train a classifier U and the shift of the  operaing point.
U	Untrained classifier.

Output
V	Mapping, such that E = TESTC(D*S,[],LABEL)
W	Tained classifier with shifted operating point.

Description

If D = A*W, with A a test dataset and W a trained classifier, then an ROC curve can be computed and plotted by ER = PRROC(D,C), PLOTE(ER). C is the  desired class number for which the error is plotted along the horizontal  axis.  The classifier W can be given any operating point along this curve by  W = W*SHIFTOP(D,E,C).  Alternatively, an overall error TYPE can be optimised. Currently we  have
TYPE = 'minerr' minimum classification error  TYPE = 'eer' equal error rate

The class index C refers to its position in the label list of the dataset  A used for training the classifier that yielded D. The relation to LABEL is LABEL = CLASSNAME(A,C); C = GETCLASSI(A,LABEL).

In case the same dataset A is used to train an untrained classifier U as  well as to determine D to

See also

datasets, mappings, testc, prroc, plote, classnames, getclassi,