How to use DataFrame.explode with a custom UDF to

问题:

I use Spark 1.5.

I have a DataFrame A_DF as follows:

+--------------------+--------------------+
|                  id|        interactions|
+--------------------+--------------------+
|        id1         |30439831,30447866...|
|        id2         |37597858,34499875...|
|        id3         |30447866,32896718...|
|        id4         |33029476,31988037...|
|        id5         |37663606,37627579...|
|        id6         |37663606,37627579...|
|        id7         |36922232,37675077...|
|        id8         |37359529,37668820...|
|        id9         |37675077,37707778...|
+--------------------+--------------------+

where interactions is a String. I want to explode this by first splitting the interactions string into a set of substrings split by a comma which I try to do as follows:

val splitArr = udf { (s: String) => s.split(",").map(_.trim) }

val B_DF = A_DF.explode(splitArr($"interactions"))

but I am getting the following error:

error: missing arguments for method explode in class DataFrame;
follow this method with `_' if you want to treat it as a partially applied function A_DF.explode(splitArr($"interactions"))

which I don't understand. So I tried something even more complicated:

val B_DF = A_DF.explode($"interactions") { case (Row(interactions: String) =>
        interactions.split(",").map(_.trim))
     }

to which I am getting an inspection warning, that reads:

Expression of Type Array[String] does not conform to expected type TraversableOnce[A_]

Any ideas?

回答1:

Dataset.explode is deprecated as of Spark 2.0.0. Unless you have a reason, stay away from it. You've been warned.

If you do have a reason to use DataFrame.explode, see the signatures:

explode[A, B](inputColumn: String, outputColumn: String)(f: (A) ⇒ TraversableOnce[B])(implicit arg0: scala.reflect.api.JavaUniverse.TypeTag[B]): DataFrame

explode[A <: Product](input: Column*)(f: (Row) ⇒ TraversableOnce[A])(implicit arg0: scala.reflect.api.JavaUniverse.TypeTag[A]): DataFrame

In either case, explode uses two parameter groups and hence the first error.

(This is Spark 2.1.0-SNAPSHOT)

scala> spark.version
res1: String = 2.1.0-SNAPSHOT

scala> val A_DF = Seq(("id1", "30439831,30447866")).toDF("id", "interactions")
A_DF: org.apache.spark.sql.DataFrame = [id: string, interactions: string]

scala> A_DF.explode(split($"interactions", ","))
<console>:26: error: missing argument list for method explode in class Dataset
Unapplied methods are only converted to functions when a function type is expected.
You can make this conversion explicit by writing `explode _` or `explode(_)(_)(_)` instead of `explode`.
       A_DF.explode(split($"interactions", ","))
                   ^

You could do it as follows (note the warning which is about deprecation of explode as I use 2.1.0-SNAPSHOT):

scala> A_DF.explode[String, String]("interactions", "parts")(_.split(",")).show
warning: there was one deprecation warning; re-run with -deprecation for details
+---+-----------------+--------+
| id|     interactions|   parts|
+---+-----------------+--------+
|id1|30439831,30447866|30439831|
|id1|30439831,30447866|30447866|
+---+-----------------+--------+

You could use the other explode as follows:

scala> import org.apache.spark.sql.Row
import org.apache.spark.sql.Row

scala> case class Interaction(id: String, part: String)
defined class Interaction

scala> A_DF.explode[Interaction]($"id", $"interactions") { case Row(id: String, ins: String) => ins.split(",").map { it => Interaction(id, it) } }.show
warning: there was one deprecation warning; re-run with -deprecation for details
+---+-----------------+---+--------+
| id|     interactions| id|    part|
+---+-----------------+---+--------+
|id1|30439831,30447866|id1|30439831|
|id1|30439831,30447866|id1|30447866|
+---+-----------------+---+--------+

Use explode function instead and you should be fine as described in the scaladoc (quoted below):

---

Given that this is deprecated, as an alternative, you can explode columns either using functions.explode():

ds.select(explode(split('words, " ")).as("word"))

or flatMap():

ds.flatMap(_.words.split(" "))

---

You could then use explode function as follows:

A_DF.select($"id", explode(split('interactions, ",") as "part"))