From: mailbox@dmitry-kazakov.de   
      
   On Fri, 08 Apr 2005 19:13:53 GMT, Ted Dunning wrote:   
      
   > What   
   > normal decision trees are particularly good at is learning decision   
   > surfaces that are largely parallel to the feature axes.   
      
   [ Another important advantage is that decision trees tolerate inseparable   
   classes in which case they just cluster the feature space in *some* way. ]   
      
   > What linear   
   > classifier techniques are good at is learning linearly separable   
   > surfaces in any orientation whatsoever.   Depending on your problem,   
   > you might be able to project into an interesting higher dimensional   
   > space to use a linear classifier, or you might have something that is   
   > ideal for a decision tree.  I recommend that you always try both.   
      
   Yes indeed.   
      
   It is also sometimes useful to mix both. That is when classes are locally   
   linearly separable, but globally are not. Then the decision tree could have   
   linear classifiers in its leaves, or least distance classifiers (not to   
   forget this important branch.)   
      
   > In practice, I have found that an interesting option is to use a   
   > decision tree on a problem and then let something like a logistic   
   > regression classifier cheat by getting to see some outputs of the rules   
   > used by the decision tree system.  You can then turn the tables and let   
   > the decision tree system see the output of the logistic regression.   
   > You know you are done with this process when the decision tree system   
   > ignores everything except your logistic regression output.   
      
   That is really interesting. Have you published something on that?   
      
   --   
   Regards,   
   Dmitry A. Kazakov   
   http://www.dmitry-kazakov.de   
      
   [ 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)