From: ted.dunning@gmail.com   
      
   On Feb 22, 7:01 pm, Symbolware  wrote:   
   >   
   > I know some product(such as weka, RapidMiner) is written in Java.   
   > However, I don't know if it could process high volume data.   
      
   Java works great for high volumes.   
      
   I built the framework used by ID Analytics in pure Java and that has   
   been extended more recently to process milliions of cases in a few   
   hours (in pure Java).  FICO uses java extensively for their fraud   
   models as well, at least in the training phase (and I think in the   
   real-time evaluation of the models as well in some cases).   
      
   Java is very fast for numerical computations and for matrix based   
   code, the cost of garbage collection can be very low.  In any case,   
   for most real-world applications of models, getting the input to the   
   model and acting on the output vastly dominates the computational   
   cost.  You could probably implement many models in QBasic on an Apple   
   II without actually having any problems (if you could just present the   
   data).  There are important and substantial exceptions, of course.   
   Any system that has a really large model (such as a Bayesian model   
   that is actually running thousands of models) or which requires   
   retrieval of comparable items such as a nearest-neighbor model may   
   actually cost a bit of time to run, but those cases aren't real common   
   in deployed systems.   
      
   [ 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)