Class DirectFilteringCascade

ml.Predictor --+            
               |            
  cla.Classifier --+        
                   |        
cla.BinaryClassifier --+    
                       |    
           ScoredCascade --+
                           |
                          DirectFilteringCascade

A cascaded rejector of M*N weak rejectors f_1(x), ..., f_{M*N}(x).

At the m-th stage, let:
    F_M(x) = \sum_{m=1}^M c_m f_m(x),
    where
        c_m: voting weight
        f_m(x): a binary classifier outputing {-1,1}

At any stage m \in {N, 2*N, ..., M*N}, if F_m(x) < 0 then classify as 0.
Else classify as 1.