Again, if the training data was labeled well enough, confusion about this sort of thing shouldn't happen.
Reply on news.ycombinator.com