How do we find out which of the support vectors for a nu-svm are its bounded or unbounded support vectors? For c-SVMs the test is easy: a support vector with $\alpha_i = C$ denotes a bounded support vector, while the others are unbounded. Whats an analogous test for nu-SVMs?

Also it would be of great help if you could direct me to how to do this test for libSVM output.



The same approach applies - depending on your scaling you will have $0 \leq \alpha_i \leq \frac{1}{n}$ or $0 \leq \alpha_i \leq 1$ usually for $\nu$-SVM. libSVM in particular scales the output so as to be equivalent to $C$-SVM. Thus, you should check the maximum absolute values for your $\alpha_i$ (since they are multiplied through by $y_i$ before being output). Note that in general, it is still possible in $C$-SVM to have a point on the margin which has $\alpha_i = C$ (but all points which have some error will have $\alpha_i = C$).

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.