Class GeneralizedCascade

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

A cascade of M weak binary classifiers. Each classifier f_m(x) \in
{-1,+1} is associated with an score c_m, and a boolean value z_m which
tells whether the weak classifier is 'an improver' (z_m = 1) or 'a filter'
(z_m = 0) (to be explained below). Define the total scores of point x
up to stage m:
    F_m(x) = \sum{i=1}^m z_i c_i f_i(x)
An improver at stage m does not classify x, but improves F_m(x)
to reach a certain goal (e.g. to minimize the total classification error up
to stage m). A filter, on the other hand, classifies x as negative
immediately iff:
    G_m(x) = F_{m-1}(x) + c_i f_i(x) < 0
Or pass x to the next stage. Finally, the cascade classifies x as
positive iff F_M(x) >= 0, and negative otherwise.