Degrees in vector support regression - RBF Core

Question

I would like to ask about the RBF core on SVM.

The sklearn documentation here: http://scikit-learn.org/stable/modules/generated/sklearn.svm.SVR.html#sklearn.svm.SVR states that "the degree of kernel function is significant only in poly , rbf , sigmoid .

I can understand the meaning of degrees in the polynomial core, but what about the Gaussian kernel (rbf)? As I can see, the default value is 3 in the sklearn library. I also ran GridSearch with some numbers I came up with, and a rating of 3 was also the best. Is it really important or is it just the wrong style? If so, can someone please explain the meaning and value of this?

thanks in advance

Answer 1

The kernel is just the basic function with which you implement your model. A polynomial function of degree 3 is equal to ax^3+bx^2+cx+d . You can use polynomials with higher degrees, however you can rework, which means that your model does not generalize very well what exactly you want. There are several methods to prevent retraining.

The RBF core is based on gaussin functions, something like aexp (-bx). If you do not know anything about mechanical training, I recommend using them. As a rule, they adapt for the better.

If you need more information about machine training, the Ng course on the course is very good for beginners.