Thinking what this would look like when having DataFrames with type information from clojure.spec: http://spark.apache.org/docs/latest/ml-collaborative-filtering.html#examples
Well what is difficult with .map is that we don't have a place for the return spec.