Spark SQL: apply aggregate functions to a list of columns

Question

Is there a way to apply an aggregate function to all (or a list of) columns of a dataframe, when doing a groupBy? In other words, is there a way to avoid doing this for every column:

df.groupBy("col1")
  .agg(sum("col2").alias("col2"), sum("col3").alias("col3"), ...)

Answer 1

There are multiple ways of applying aggregate functions to multiple columns.

GroupedData class provides a number of methods for the most common functions, including count, max, min, mean and sum, which can be used directly as follows:

• Python:

  df = sqlContext.createDataFrame(
      [(1.0, 0.3, 1.0), (1.0, 0.5, 0.0), (-1.0, 0.6, 0.5), (-1.0, 5.6, 0.2)],
      ("col1", "col2", "col3"))
  
  df.groupBy("col1").sum()
  
  ## +----+---------+-----------------+---------+
  ## |col1|sum(col1)|        sum(col2)|sum(col3)|
  ## +----+---------+-----------------+---------+
  ## | 1.0|      2.0|              0.8|      1.0|
  ## |-1.0|     -2.0|6.199999999999999|      0.7|
  ## +----+---------+-----------------+---------+
  

• Scala

  val df = sc.parallelize(Seq(
    (1.0, 0.3, 1.0), (1.0, 0.5, 0.0),
    (-1.0, 0.6, 0.5), (-1.0, 5.6, 0.2))
  ).toDF("col1", "col2", "col3")
  
  df.groupBy($"col1").min().show
  
  // +----+---------+---------+---------+
  // |col1|min(col1)|min(col2)|min(col3)|
  // +----+---------+---------+---------+
  // | 1.0|      1.0|      0.3|      0.0|
  // |-1.0|     -1.0|      0.6|      0.2|
  // +----+---------+---------+---------+
  

Optionally you can pass a list of columns which should be aggregated

df.groupBy("col1").sum("col2", "col3")

You can also pass dictionary / map with columns a the keys and functions as the values:

• Python

  exprs = {x: "sum" for x in df.columns}
  df.groupBy("col1").agg(exprs).show()
  
  ## +----+---------+
  ## |col1|avg(col3)|
  ## +----+---------+
  ## | 1.0|      0.5|
  ## |-1.0|     0.35|
  ## +----+---------+
  

• Scala

  val exprs = df.columns.map((_ -> "mean")).toMap
  df.groupBy($"col1").agg(exprs).show()
  
  // +----+---------+------------------+---------+
  // |col1|avg(col1)|         avg(col2)|avg(col3)|
  // +----+---------+------------------+---------+
  // | 1.0|      1.0|               0.4|      0.5|
  // |-1.0|     -1.0|3.0999999999999996|     0.35|
  // +----+---------+------------------+---------+
  

Finally you can use varargs:

• Python

  from pyspark.sql.functions import min
  
  exprs = [min(x) for x in df.columns]
  df.groupBy("col1").agg(*exprs).show()
  

• Scala

  import org.apache.spark.sql.functions.sum
  
  val exprs = df.columns.map(sum(_))
  df.groupBy($"col1").agg(exprs.head, exprs.tail: _*)
  

There are some other way to achieve a similar effect but these should more than enough most of the time.

See also:

• Multiple Aggregate operations on the same column of a spark dataframe

Answer 2

Another example of the same concept - but say - you have 2 different columns - and you want to apply different agg functions to each of them i.e

f.groupBy("col1").agg(sum("col2").alias("col2"), avg("col3").alias("col3"), ...)

Here is the way to achieve it - though I do not yet know how to add the alias in this case

See the example below - Using Maps

val Claim1 = StructType(Seq(StructField("pid", StringType, true),StructField("diag1", StringType, true),StructField("diag2", StringType, true), StructField("allowed", IntegerType, true), StructField("allowed1", IntegerType, true)))
val claimsData1 = Seq(("PID1", "diag1", "diag2", 100, 200), ("PID1", "diag2", "diag3", 300, 600), ("PID1", "diag1", "diag5", 340, 680), ("PID2", "diag3", "diag4", 245, 490), ("PID2", "diag2", "diag1", 124, 248))

val claimRDD1 = sc.parallelize(claimsData1)
val claimRDDRow1 = claimRDD1.map(p => Row(p._1, p._2, p._3, p._4, p._5))
val claimRDD2DF1 = sqlContext.createDataFrame(claimRDDRow1, Claim1)

val l = List("allowed", "allowed1")
val exprs = l.map((_ -> "sum")).toMap
claimRDD2DF1.groupBy("pid").agg(exprs) show false
val exprs = Map("allowed" -> "sum", "allowed1" -> "avg")

claimRDD2DF1.groupBy("pid").agg(exprs) show false

Answer 3

Current answers are perfectly correct on how to create the aggregations, but none actually address the column alias/renaming that is also requested in the question.

Typically, this is how I handle this case:

val dimensionFields = List("col1")
val metrics = List("col2", "col3", "col4")
val columnOfInterests = dimensions ++ metrics

val df = spark.read.table("some_table") 
    .select(columnOfInterests.map(c => col(c)):_*)
    .groupBy(dimensions.map(d => col(d)): _*)
    .agg(metrics.map( m => m -> "sum").toMap)
    .toDF(columnOfInterests:_*)    // that's the interesting part

The last line essentially renames every columns of the aggregated dataframe to the original fields, essentially changing sum(col2) and sum(col3) to simply col2 and col3.