There are several examples that show margins for SVMs.
We use many different techniques, and some of them are buggy. I'd like to replace all of them by the one @jakevdp uses here: https://github.com/jakevdp/PythonDataScienceHandbook/blob/master/notebooks/05.07-Support-Vector-Machines.ipynb

that is using contour levels on the decision function.

I mentioned this in #8364 but I think it deserves a separate issue to allow easier tracking.