According to Question I optimizing the MAP estimate is the right" thing to do (right in the sense of minimizing 0-1 classification loss). Remember that log is monotonic: if f(a) 0.5 output "class 2" if o(wx) 0.5 output "class 2" if o(wx)