I'm trying to do pattern matching inside a Dataframe map function - matching a Row with a Row pattern having a nested Case Class. This dataframe is a result of a join, and has the schema shown below. It has some columns of primitive types, and 2 compound columns:
case class MyList(values: Seq[Integer])
case class MyItem(key1: String, key2: String, field1: Integer, group1: MyList, group2: MyList, field2: Integer)
val myLine1 = new MyItem ("MyKey01", "MyKey02", 1, new MyList(Seq(1)), new MyList(Seq(2)), 2)
val myLine2 = new MyItem ("YourKey01", "YourKey02", 2, new MyList(Seq(2,3)), new MyList(Seq(4,5)), 20)
val dfRaw = Seq(myLine1, myLine2).toDF
dfRaw.printSchema
dfRaw.show
val df2 = dfRaw.map(r => r match {
case Row(key1: String, key2: String, field1: Integer, group1: MyList, group2: MyList, field2: Integer) => "Matched"
case _ => "Un matched"
})
df2.show
My problem is, after that map function, all I got was "Un matched":
root
|-- key1: string (nullable = true)
|-- key2: string (nullable = true)
|-- field1: integer (nullable = true)
|-- group1: struct (nullable = true)
| |-- values: array (nullable = true)
| | |-- element: integer (containsNull = true)
|-- group2: struct (nullable = true)
| |-- values: array (nullable = true)
| | |-- element: integer (containsNull = true)
|-- field2: integer (nullable = true)
+---------+---------+------+--------------------+--------------------+------+
| key1| key2|field1| group1| group2|field2|
+---------+---------+------+--------------------+--------------------+------+
| MyKey01| MyKey02| 1| [WrappedArray(1)]| [WrappedArray(2)]| 2|
|YourKey01|YourKey02| 2|[WrappedArray(2, 3)]|[WrappedArray(4, 5)]| 20|
+---------+---------+------+--------------------+--------------------+------+
df2: org.apache.spark.sql.Dataset[String] = [value: string]
+----------+
| value|
+----------+
|Un matched|
|Un matched|
+----------+
If I ignore those two struct columns in the case branch (replacing group1: MyList, group2: MyList with _, _, then it works
case Row(key1: String, key2: String, field1: Integer, group1: MyList, group2: MyList, field2: Integer) => "Matched"
Could you please help on how to do pattern matching on that Case class? Thanks!
struct
columns are treated as org.apache.spark.sql.catalyst.expressions.GenericRowWithSchema
in spark
so you will have to define match case as
import org.apache.spark.sql.catalyst.expressions._
val df2 = dfRaw.map(r => r match {
case Row(key1: String, key2: String, field1: Integer, group1: GenericRowWithSchema, group2: GenericRowWithSchema, field2: Integer) => "Matched"
case _ => "Un matched"
})
And defining match case with wild-card (_) works because Scala compiler implicitly evaluates org.apache.spark.sql.catalyst.expressions.GenericRowWithSchema
as datatype.
Defining case as below should work too as with the wild-card due to implicit evaluation
case Row(key1: String, key2: String, field1: Integer, group1, group2, field2: Integer) => "Matched"