this is smart, happy to learn that hierarchical groupings affect the overall performance based on sample size at each hierarchy. If OOM: resort to Spark to split across nodes for you.If small enough at a hierarchy level, prefer to use single-node in-...
