From: ted.dunning@gmail.com   
      
   Dr. Colombes wrote:   
      
   > It would seem that if the two goal classes have very unequal prior   
   > probabilities (say P(+) = 0.0001, P(-) = 0.9999), then more training   
   > samples would be required to "learn" the "+" class.   
      
   The heuristic formula you were using should probably be amended to have   
   something like min(N+, N-) instead of using the total number of   
   training examples.   
      
   And I would have serious doubts about the sqrt.  Log is probably closer   
   to reality since the volume of hyper-cube is exponential with   
   dimension.   
      
   [ comp.ai is moderated ... your article may take a while to appear. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)