SVM custom RBF kernel IndexError

Question

I want to implement SVM RBF kernel for my non-linearly separable data. I have defined my kernel as:

def rbf(va, vb):
    gamma = 0.7
    return exp(-gamma * linalg.norm(va - vb) ** 2)
clf = svm.SVC(kernel=rbf)
clf.fit(va, vb)

It is showing error:

if X.shape[0] != X.shape[1]: IndexError: tuple index out of range

How to resolve it?

Gambit1614 Gambit1614 · Accepted Answer · 2017-09-26T17:35:13

Your shapes for va and vb are mismatching. They need to be of same shape. The syntax for rbf kernel should be of the following form:

def rbf(va,vb) , where va are the feature of sample,say X and vb are the features of sample say X`.

va and vb are not features and labels.

You can check this link on Github for similar implementation. You can see that he is passing all pairs of samples to the rbf kernel and not the features and labels of a sample. Go to line no. 95.

SVM custom RBF kernel IndexError

1 Answers