LABELD

Fixed mapping finding labels of classification dataset

(perform crisp classification)

    LABELS = LABELD(A,W)
    LABELS = A*W*LABELD
    LABELS = LABELD(A*W,THRESH)
    LABELS = A*W*LABELD(THRESH)
    LABELS = LABELD(A,W,THRESH)

Input
A	Dataset
W	Trained classifier
THRESH	Rejection threshold

Output
LABELS	List of labels

Description

Returns the labels of the classification dataset Z=A*W. For each object  in Z (i.e. each row) the feature label or class label (i.e. the column  label) of the maximum column value is returned.

Effectively, this performs the classification. It can also be considered  as a conversion from soft labels (posteriors) stored in Z to crisp labels.

Be aware that in case A is not a dataset but an array of doubles, A*W is  also an array of doubles and the output of A*W*LABELD is a column vector  pointing to the column with maximum classification confidence. If the  real labels strored in W are needed, convert A to a dataset first.

When the parameter THRESH is supplied, then all objects which  classifier output falls below this value are rejected. The returned  label is then NaN or a string with spaces (depending if the labels are  numeric or string). Because the output of the classifier is used, it  is recommended to convert the output to a posterior prob. output using

CLASSC.

(David Tax, 27-12-2004)

See also

mappings, datasets, testc, plotc,