Instantiates the FeatureSet, by
computing all the relevant conditionalFreq of all features on the training set; note that a Laplace prior can be introduced, by setting
parameter priorCounts different from zero (i.e., the quantity priorCounts will be then added to each computed count).