@mwichary The statistician in me suggests that these statistical engines do just what they are designed to do: mirror the main features of their training data set.