SMOreg在Java代码中花费了太多时间

Question

I am using SMOreg(SVM regression) from weka in my java code. It is taking more than 30 min at buildClassifier() method. When I ran same inputs through weka gui toolbox it is hardly taking 2 mins. Here is part of my code.

Instances trainD = DataSource.read(trin_file);
trainD.setClassIndex(trainD.numAttributes()-1);
SMOreg predict = new SMOreg();
predict.buildClassifier(trainD);

Is there any specific reason to take longer time in java-code for same task?

Answer 1

You're probably not passing all of the same options. Also, make sure you're allocating the same amount of memory for running the GUI and running your native java code.

In general, regression takes a while. As one example, I had an experiment that took about 2 hours to run using SMO/SVM (classification). When I changed the class attribute to be continuous, and ran SMOreg, it took over a day to run.