ntiles is a popular aggregation over a window and is commonly used to divide an input dataset into n parts.

For example, if we want to partition the statesPopulationDF by State (window specification is shown previously), order by population, and then divide into two portions, we can use ntile over the windowspec:

import org.apache.spark.sql.functions._
scala> statesPopulationDF.select(col("State"), col("Year"),
 ntile(2).over(windowSpec), rank().over(windowSpec)).sort("State",
 "Year").show(10)
+-------+----+-------------------------------------------------------------
 ----------------------------------------------------------+----------------
 ---------------------------------------------------------------------------
 --------------------------+
| State|Year|ntile(2) OVER (PARTITION BY State ORDER BY Population DESC
 NULLS LAST ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW)|RANK() OVER
 (PARTITION BY State ORDER BY Population DESC NULLS LAST ROWS BETWEEN
 UNBOUNDED PRECEDING AND CURRENT ROW)|
 +-------+----+-------------------------------------------------------------
 ----------------------------------------------------------+----------------
 ---------------------------------------------------------------------------
 --------------------------+
 |Alabama|2010| 2| 6|
 |Alabama|2011| 2| 7|
 |Alabama|2012| 2| 5|
 |Alabama|2013| 1| 4|
 |Alabama|2014| 1| 3|
 |Alabama|2015| 1| 2|
 |Alabama|2016| 1| 1|
 | Alaska|2010| 2| 7|
 | Alaska|2011| 2| 6|
 | Alaska|2012| 2| 5|
 +-------+----+-------------------------------------------------------------
 ----------------------------------------------------------+----------------
 --------------------------------------------------------------

As shown previously, we have used the Window function and ntile() together to divide the rows of each State into two equal portions.