HOME Multi-labeling system Dataset details Feature labelsFeature domain testing

Feature domain testing

By default features can have any real value. In case of missing data an NaN has to be supplied. In some applications however the domain of feature values is restricted and it might be needed to test incoming values before they are used in some routines. To facilitate this feature domains are introduced, stored in the featdom-field of the dataset definition. The commands setfeatdom and getfeatdom can be used to set and retrieve this field.

Setting the featdom-field is sufficient to start automatic inspection of the feature values. Whenever the data-field of a dataset changes the feature values are tested against the domains stored in the featdom-field.

The featdom-field itself has to be a cell-array with a cell for every feature containing a domain description. The following rules apply:

> Rules for feature domain description
[] empty, all numbers are allowed.
real array, size (N,2) Defines N domains on the real axis, lowerbounds are R(:,1), upperbounds R(:,2). N should be at least 2. E.g: R = [0 1; 0 1] defines a single region between 0 and 1. R = [-1 1; 5 10; 11 11] defines two regions and a single value (11).
integer array, size (1,N) Set of N integer values.
string array size (N,F) String array of N strings. The feature values in the data-field of A will be coded by integers from 1 to N. A feature like this may be assigned to the dataset by A(:,j) = S, in which S is a string array, containing a string for each object.

The last option shows how to create nominal features.

R.P.W. Duin, January 28, 2013

HOME Multi-labeling system Dataset details Feature labelsFeature domain testing