Simple example of handling soft labels in PRTools
Soft labels are implemented next to the 'crisp' and 'targets' labels. Like 'targets' labels they are stored in the target field of a dataset. Their values should be between 0 and 1. For every class a soft label values should be given. The density based classifiers can handle soft labels, interpreting them as class weights for every objects in the density estimation.
The posterior probabilities found by classifying objects can be interpreted as soft labels. They, however, sum to one (over the classes), while this is not necessary for training and test objects.
Note that the routine CLASSSIZES returns the sum of the soft labels over the dataset for every class separately. In contrast to crisp labels the sum over the classes of the output of CLASSSIZES is not necessarily equal to number of objects in the dataset.
The routine SELDATA(A,N) returns the entire dataset in case of a soft labeled dataset A for every value of N and not just class N, as all objects may participate in all classes.