There was a tip >The LinearSVC class regularizes the bias term, so you should center the training set first by subtracting its mean I couldn't find any resource discussing why this is so? What happens if not centered? What are the implications of (not) regularizing the bias term?