CODEWITHSUNDEEP
pythonmachine-learningscikit-learn
Maksim Khaitovich
Maksim Khaitovich

Reputation: 4792

scikit-learn, linearsvc - how to get support vectors from the trained SVM?

I am using LinearSVC from scikit-learn library and I wonder if it is possible to somehow pull out the vectors which my model uses after training to make predictions. Tried to google it for some time but without any luck. Anyone knows?

Upvotes: 13

Views: 16063

Answers (4)

dinatrina
dinatrina

Reputation: 163

There is actually a way: I found here how to obtain the support vectors from linearSVC - I'm reporting the relevant portion of code:

from sklearn.svm import LinearSVC

clf = LinearSVC('''whatever fits your specs''')
clf.fit(X,y)
# get the support vectors through the decision function
decision_function = clf.decision_function(X)
# we can also calculate the decision function manually as previously noted
# decision_function = np.dot(X, clf.coef_[0]) + clf.intercept_[0]
# The support vectors are the samples that lie within the margin
# boundaries, whose size is conventionally constrained to 1
support_vector_indices = np.where(np.abs(decision_function) <= 1 + 1e-15)[0]
support_vectors = X[support_vector_indices]

Upvotes: 1

Ravid Cohen
Ravid Cohen

Reputation: 126

I am not sure if it helps, but I was searching for something similar and the conclusion was that when:

clf = svm.LinearSVC()

Then this:

clf.decision_function(x)

Is equal to this:

clf.coef_.dot(x) + clf.intercept_

Upvotes: 8

Eswar
Eswar

Reputation: 1212

This could help you.

clf = svm.SVC( kernel='rbf',C=0.05)
clf.fit(traindata,y)
print clf.support_vectors_

This link can you more information if needed. http://scikit-learn.org/stable/modules/svm.html

Upvotes: 3

elyase
elyase

Reputation: 40973

Unfortunately there seems to be no way to do that. LinearSVC calls liblinear (see relevant code) but doesn't retrieve the vectors, only the coefficients and the intercept.

One alternative would be to use SVC with the 'linear' kernel (libsvm instead of liblinear based), but also poly, dbf and sigmoid kernel support this option:

from sklearn import svm

X = [[0, 0], [1, 1]]
y = [0, 1]

clf = svm.SVC(kernel='linear')
clf.fit(X, y)
print clf.support_vectors_

Output:

[[ 0.  0.]
 [ 1.  1.]]

liblinear scales better to large number of samples, but otherwise the are mostly equivalent.

Upvotes: 9

Related Questions