I think you have to give dummy features to the kernel and mkl in order for this to work. If it doesnt, please paste a gist with a full minimal example to reproduce the problem. H
2016-02-10 17:11 GMT+00:00 Tobias Sterbak <[email protected]>: > Hi everyone, > > > > I have some precomputed kernel matrices and want to use them in a > mkl-setup. That means there are no features (or I don’t want to use them). > > I create a combined kernel and append these matrices. Then set the kernels > for the mklClassifier. > > Then I want to use cross-validation > > splitting_strategy=StratifiedCrossValidationSplitting(labels, 5) > > evaluation_criterium=ContingencyTableEvaluation(ACCURACY) > > cross_validation=CrossValidation(mkl, labels, splitting_strategy, > evaluation_criterium) > > > > This works so far. But when I now want to get the result > > result=cross_validation.evaluate() > > Then I get this error > > SystemError: [ERROR] In file > /shogun-4.0.0/src/shogun/evaluation/CrossValidation.cpp line 76: > CrossValidation::evaluate() is only possible if features are attached > > > > So cross validation is not possible without features? > > > > I hope you can help me. > > > > Best regards, > > Tobias >
