Aggregate over column arrays in DataFrame in PySpark?

Question

Let's say I have the following DataFrame:

[Row(user='bob', values=[0.5, 0.3, 0.2]),
Row(user='bob', values=[0.1, 0.3, 0.6]),
Row(user='bob', values=[0.8, 0.1, 0.1])]

I would like to groupBy user and do something like avg(values) where the average is taken over each index of the array values like this:

[Row(user='bob', avgerages=[0.466667, 0.233333, 0.3])]

How can I do this in PySpark?

Answer 1

You can expand array and compute average for each index.

Python

from pyspark.sql.functions import array, avg, col

n = len(df.select("values").first()[0])

df.groupBy("user").agg(
    array(*[avg(col("values")[i]) for i in range(n)]).alias("averages")
)

Scala

import spark.implicits._
import org.apache.spark.functions.{avg, size}

val df = Seq(
  ("bob", Seq(0.5, 0.3, 0.2)),
  ("bob", Seq(0.1, 0.3, 0.6))
).toDF("user", "values")

val n = df.select(size($"values")).as[Int].first
val values = (0 to n).map(i => $"values"(i))

df.select($"user" +: values: _*).groupBy($"user").avg()