convert each row of dataframe to a map

Question

I have a dataframe with columns A & B of type String. Let's assume the below dataframe

+--------+
|A  | B  |
|1a | 1b |
|2a | 2b |

I want to add a third column that creates a map of A & B column

+-------------------------+
|A  | B  |  C             |
|1a | 1b | {A->1a, B->1b} |
|2a | 2b | {A->2a, B->2b} |

I'm attempting to do it the following way. I have udf which takes in a dataframe and returns a map

val test = udf((dataFrame: DataFrame) => {
val result = new mutable.HashMap[String, String]
dataFrame.columns.foreach(col => {
  result.put(col, dataFrame(col).asInstanceOf[String])
})
result
})

I'm calling this udf in following way which is throwing a RunTimeException as I'm trying to pass a DataSet as a literal

df.withColumn("C", Helper.test(lit(df.select(df.columns.head, df.columns.tail: _*)))

I don't want to pass df('a') df('b') to my helper udf as I want them to be generic list of columns that I could select. any pointers?

Answer 1

map way

You can just use map inbuilt function as

import org.apache.spark.sql.functions._
val columns = df.columns
df.withColumn("C", map(columns.flatMap(x => Array(lit(x), col(x))): _*)).show(false)

which should give you

+---+---+---------------------+
|A  |B  |C                    |
+---+---+---------------------+
|1a |1b |Map(A -> 1a, B -> 1b)|
|2a |2b |Map(A -> 2a, B -> 2b)|
+---+---+---------------------+

Udf way

Or you can use define your udf as

//collecting column names to be used in the udf
val columns = df.columns
//definining udf function
import org.apache.spark.sql.functions._
def createMapUdf = udf((names: Seq[String], values: Seq[String])=> names.zip(values).toMap)
 //calling udf function 
df.withColumn("C", createMapUdf(array(columns.map(x => lit(x)): _*), array(col("A"), col("B")))).show(false)

I hope the answer is helpful

Answer 2

@ Ramesh Maharjan - Your answers are already great, my answer is just make your UDF answer also in dynamic way using string interpolation.

Column D is giving that in dynamic way.

df.withColumn("C", createMapUdf(array(columns.map(x => lit(x)): _*), 
array(col("A"), col("B"))))
.withColumn("D", createMapUdf(array(columns.map(x => lit(x)): _*), 
array(columns.map(x => col(s"$x") ): _* ))).show()