Clarification required regarding type of learner (model)

[albalansi]

Hi everyone. i find this section in the book not very informative. Page 85/86 talks about using the SVM learner but introduces something new: type="C-classification" . No further elaboration is presented as to what this changes, what other options exist and what's the default.
the code in question is the following:

learner = lrn("classif.svm",
type = "C-classification",
kernel = "radial",
cost = to_tune(1e-1, 1e5),
gamma = to_tune(1e-1, 1)
)
learner

can anyone please clarify what this is about and provide the other options. THANK YOU ALL.

[bblodfon]

Hi @mr1lazycoder ,

These are parameters from the e1071::svm() wrapped train function for this learner. You can see the parameter defaults and ranges here => https://mlr3learners.mlr-org.com/reference/mlr_learners_classif.svm.html#parameters, while for specific details of each one (like the type you asked), you have to look at the original documentation of the wrapped function, i.e. here.

Now you would be correct to say that even if you look at the original documentation, they authors don't exactly specify what does this C-classification or nu-classification types actually do! I guess when these packages were written they thought that the users have already the theory of SVMs in mind, so please take a look here for a similar question and answers to what these types actually mean!

[albalansi]

well this makes sense now. I still think it would be nice if the authors mention nu-classification. Thanks @bblodfon for the clarification.

[be-marc]

I remember that there was a sentence about nu-classification in the early drafts of this chapter. Seems we decided against it.

[bblodfon]

Maybe worth adding a sentence now? :)