0

I'm using sklearns SVC with rbf kernel and ovr decision function. While studying the decision_function I noticed that the label with highest confidence score doesn't necessarily correspond to the prediction. Is such behavior normal? If yes, why?

Thx in advance.

Example: For the following decision function output

5.99088671, 3.96528944, 6.02144331, 1.94929957, 9.05033791, 9.04567359, 2.98166027, 1.97837266, 1.96593488, 9.07656409, 2.97453757

the SVM predicted the label with value 9.05033791

user11113738
  • 11
  • 3
  • Possible duplicate of [Confusing probabilities of the predict\_proba of scikit-learn's svm](https://stackoverflow.com/questions/30674164/confusing-probabilities-of-the-predict-proba-of-scikit-learns-svm) – Vivek Kumar Feb 25 '19 at 14:30
  • @VivekKumar I did read these questions and the documentation. The stackoverflow questions are about `predict_proba` (as noted in the documentation) which implements an internal cross-validation and another fitting process, which(if I'm correct) `decision_function` does not. – user11113738 Feb 26 '19 at 06:35
  • Aah yes. Sorry for this. I overlooked that. This is explained here: https://github.com/scikit-learn/scikit-learn/issues/8277 – Vivek Kumar Feb 26 '19 at 09:31

0 Answers0