I've tried all the of SVM nodes I could find in KNIME, and found that only the 'LIBSVMLearner' node accurately classifies my data (as tested with the 'LIBSVMPredictor' node). I would like to save this model to disk, so that I can use it in another workflow. Unfortunately, the 'PMML Writer' node only works with the 'SVM Learner' node, and not LIBSVM. Does anyone know how I can save a LIBSVM model as PMML or any other format?
there is a model writer node (and reader) in the IO Category. It generates a zip file with the information about the model. It works with every Knime Model Port.
Unfortunately, the file created by the model writer node is not in PMML format. Does anyone else have a solution to this? I hit the same problem for the same reason as George.
I had some code that could convert libsvm models to PMML, though those are not released anywhere. Maybe R developers have already have some package that can perform that.
I have to note that not all libsvm models can be represented -imho- in PMML and the way I have transformed, it was not compatible with the SVMPredictor node's expected PMML model.