XPost: comp.ai.edu, comp.programming   
   From: rif@mit.edu   
      
   anja.hauth@gmx.de (Anja) writes:   
      
   > Hi,   
   >   
   > I would like to learn more about support vector machines. Therefor I   
   > decide to implement a support vector machine algorithm on my one. I   
   > have looked on many papers and some books for an algorithm that I can   
   > implement. But the only statement I could found, was:   
   >   
   > maximize W(a) = ...   
   > subject to 0 <= a <= C and ...   
   >   
   > And this is easy to solve with a quadratic programming problem. But   
   > nobody discribes how I can solve the quadratic programming problem.   
   > And I have no idea to sove it. Can someone help me to find an   
   > algorithm that I can implement?   
   >   
   > Thanks   
   > Anja   
   >   
      
   My opinion is that there are much better ways to learn about SVMs then   
   implementing one on your own.  I think there's a lot more value-added   
   in playing with SVMs, getting a feel for what they can and can't do,   
   then there is in making a toy implementation.  Nevertheless...   
      
   If you must do your own implementation, I would suggest implementing   
   John Platt's Sequential Minimal Optimization algorithm.  If you Google   
   for John Platt, it'll turn up.  This algorithm is fairly easily to   
   implement and understand.  As described, it will not be as fast as   
   state-of-the-art solvers; the "action" in making SVMs fast is all   
   about caching kernel products.  However, this is not necessary to make   
   an algorithm that will work.   
      
   Cheers,   
      
   rif   
      
   [ 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)