MATLAB: Do the SVMTRAIN and SVMCLASSIFY functions in the Bioinformatics Toolbox 2.5 (R2007a) return different answers than previous versions

Question

I am using the Support Vector Machine (SVM) functions in the Bioinformatics Toolbox. I notice that the SVMTRAIN and SVMCLASSIFY functions return different results compared to their counterparts in previous versions of the Bioinformatics Toolbox.

Answer 1

The difference is caused by two new parameters that were added in the Bioinformatics Toolbox 2.3.1 (R2006a+), 'boxconstraint' and 'autoscale'. The default values of these parameters are different to the settings used in R14SP3. To get the same behavior as the old version use:

svmStruct = svmtrain(TrainData,groups, 'Kernel_Function', 'rbf','BoxConstraint',1./sqrt(eps),'autoscale',false);

These changes are documented in the release notes for R2006a+ and R2006b.