Is it right to say that when I increment the C parameter in a SVM, I am incrementing the number of support vectors needed? Where does this relation come from ? Premise: I have understood the influence of C about the bias and variance tradeoff, but I can't imagine how this parameter influence the number of support vectors needed
Asked
Active
Viewed 107 times
0
-
Could you define parameter C? Possibly link to a source or state the formula here. – Nikolas Rieble Feb 20 '17 at 14:53
-
further, http://scikit-learn.org/stable/auto_examples/svm/plot_rbf_parameters.html – Nikolas Rieble Feb 20 '17 at 14:57
-
@NikolasRieble I have understood the influence of C about the bias and variance tradeoff, but I can't imagine how this parameter influence the number of support vector needed . – Simone Feb 20 '17 at 16:48
-
The support vector are defined as those datapoints which define the hyperplane. If C influences the shape of the hyperplane, it must influence the number of support vectors. But I doubt that you can find a general formula such as "high C - high number of support vectors" because it does depend on the data. In some datasets the value chosen for C might not influence the number of support vectors, whereas in other datasets it does. – Nikolas Rieble Feb 20 '17 at 16:53