Two billion variables and possibly millions of labels. I'd be interested in knowing what machine-learning algorithm they're using.