XPost: comp.ai.neural-nets   
   From: marina@sapir.us   
      
   I do not know about generalizations of SMO. Instead, I would like to   
   ask you some questions about the existing algorithm.   
      
   We did some experiments with SMO, which bring some peculiar results.   
      
   1. We build SVR with linear kernel. We dropped one variable at a time,   
   and looked how the criterion SVR minimizes changes. We were very   
   surprised to find that the criterion value got smaller and smaller as   
   more variables were dropped.  If SMO finds the global minimum, it could   
   not have happened.   
      
   Is this a known fact that SMO does not find the global minimum?   
      
   1. We wanted to find, if the order of the instances affects the   
   results. As another surprise, we could not find a case when the order   
   makes any difference.   
   Yet, if the algorithm does not find the global minimum, than the   
   results could depend on the order.   
      
   Do you know something about it, or could you advice me, where I can   
   find some answers?   
      
   Thank you.   
      
   Marina Sapir   
      
   [ 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)