What data from MATLAB's svmstruct are needed for classification in a different language?

Question

As the title already states Iam wondering what data exactly are needed from the MATLAB svmstruct to be able to classify a new instance outside of MATLAB, e.g. in python or c++.

I certainly need:

• the SupportVectors themselves (for calculating the distance using the kernel function).
• The kernel function (including parameter[s]) ported / rewritten
• The bias

Now I am uncertain about ScaleData, Alpha and SupportVectorindices... are they needed? Do I lack some base understanding here?

Answer 1

ScaleData is probably needed as this supplies normalizing info for unseen data.

Support vector indices is needed to reference the data inside the SupportVectors variable.

Alpha is also needed as it supplies the weights for the support vectors.

Shawe-Taylor and Cristianini's book provides some background on the equations for an SVM.