XPost: comp.ai.neural-nets, sci.image.processing, sci.math.num-analysis   
   From: ted.dunning@gmail.com   
      
   Trying any classifier without exploratory analysis is a major mistake.   
   Aside from the silliest and simplest problems, you generally need to   
   transform your input variables somehow to make the problem easier for   
   the classifier.   
      
   My own laundry list of exploratory methods includes scatter plots,   
   distributional analysis and clustering just as Greg suggested.  You   
   also have to decide what good performance means and how you will   
   withold data for final evaluation.   
      
   The very next step is to try a number of simple classifiers and   
   evaluate any differences in performance.  Occasionally, you will just   
   be done because performance for one of the simple options will be   
   sufficient.  The majority of the time, however, you will have to do   
   something additional.  This may include rescaling of an input, or   
   restating the entire input space by using distance to cluster   
   centroids.  Input preparation is often where you will spend most of   
   your time.  Building models is generally so fast that you can try   
   virtually everything all the time.   
      
   [ comp.ai is moderated.  To submit, just post and be patient, or if ]   
   [ that fails mail your article to , and ]   
   [ ask your news administrator to fix the problems with your system. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)