Changed LDA Feature class to handle non-continuous sets of class labels

[vonclites]

Before, class labels were assumed to be {0,1,2,3,..., n}. Now, labels may be {1, 3, 7, ..., n}. I found myself needing to train and test on subsets of my subjects, as in just subjects 1, 3, 7, etc., and wanted to retain their original class labeling. Furthermore, my first class label was not 0, it was 1, as subject 1 is the first subject in the AT&T dataset. I wanted my class labels to directly reflect the subject's number.