Inconsistent documentation for C parameter in SVM estimators

Description

The current description of the parameter C for sklearn.svm.LinearSVR is given as:

C : float, optional (default=1.0)

• Penalty parameter C of the error term. The penalty is a squared l2 penalty. The bigger this parameter, the less regularization is used.

which is more verbose than the description given for sklearn.svm.{SVR, SVC, LinearSVC}:

C : float, optional (default=1.0)

• Penalty parameter C of the error term.

Would it be time to update the other estimators so that they match this more verbose description?

Links

Link to LinearSVR Link to SVR

Versions

scikit-learn 0.19.2 (latest stable version)

[neginkrahbar]

Picking up on this issue with @nahsin

[jnothman]

That’s one reason we refer to it as “hinge” and “squared hinge” rather than “l1” and “l2” loss