|
Hello everybody, does anyone know an approach to perform model selection in Weka through cross validation for regression problems? afaik, the cross validation is implemented in Weka just to assess the performance of the classifier. I guess that calling Weka API from Java might solve the problem, but is there a GUI-based approach? Thanks in advance, Lucian |
|
Hi, I suppose you could use Stacking in the following way: weka.classifiers.meta.Stacking -t wisconsin.arff -M weka.classifiers.functions.SimpleLinearRegression -B weka.classifiers.functions.LinearRegression -B weka.classifiers.functions.SMOreg the specific meta classifier here would select exactly one of all the base classifiers. And if you need to optimise the base classifiers, you could wrap these into meta.CVParameterSelection or GridSearch as well. hth, Bernhard |
