Changing numWordsInNgram breaks classification if already trained

[thomwiggers]

Due to the hash not including all of the parameters which are used in calculating the probabilities, changing the n of n-grams will not cause a re-evaluation of the probabilities with this new setting. Because the incoming instance (prompt) is being split by the new numWordsInNgrams it breaks classification.

[DanielRapp]

Fantastic, thanks!