Adding classes_ to svm makes my head hurt #1037
Description
I couldn't figure out to add support for arbitrary class labels to the SVMs and add a classes_ attribute.
I'm moving to other stuff for the moment, it would be great if someone found some time for this.
The trouble comes with class_weight and regression. I think it would be good to add a LabelEncoder somewhere to take care of the encoding of the class_weights. But apparently the implementation also requires a class-weights on the python side in the regression case. And the LabelEncoder should obviously not be used in the SVR.
This is doable but I couldn't make it pretty so I just left it :-/