From Joseph Bradley:
Supporting Pipelines and advanced use cases: There really needs to be more design discussion around SparkR. Felix Cheung would you be interested in leading some discussion? I'm envisioning something similar to what was done a while back for Pipelines in Scala/Java/Python, where we consider several use cases of MLlib: fitting a single model, creating and tuning a complex Pipeline, and working with multiple languages. That should help inform what APIs should look like in Spark R.
Certain ML model, such as OneVsRest, is harder to represent in a single call R API. Having advanced API or Pipeline API like this could help to expose that to our users.