Spark's 2.0 documentation for OneVsRest() can be found at:
Another way to visualize this is, to assess for a given binary classifier can we break down an n-class input into N number of logistic regressions and then pick the one that describes the data the best. There are numerous examples of this classifier using Scikit Learn library in Python as follows:
But we recommend you do a quick scan of the actual Scala source code (less than 400 lines only) at GitHub: