answered Apr 10 '19
it's not the probability, but the distance (chi_sqr is used here) to the closest neighbour in the train db, so the "inverse" of a probability.
