Example of forward prototype selection
This example shows the use of PROTSELFD for a greedy forward selection of prototypes from a square dissimilarity matrix in order to optimize the representation set.
The final plot shown is a 'feature curve'. This is the error as a function of the number of features used (here the number of prototypes). The error measure is the mean classification error of the 1-NN rule using the given dissimilarities as distances (KNNDC(,1))