I have this dataframe
----- ------- ----------- ------------------- -----
|empID|Zipcode|ZipCodeType|City |State|
----- ------- ----------- ------------------- -----
|1000 |704 |STANDARD |PARC PARQUE |PR |
|1000 |704 |STANDARD |PASEO COSTA DEL SUR|PR |
|1001 |709 |STANDARD |BDA SAN LUIS |PR |
|1001 |76166 |UNIQUE |CINGULAR WIRELESS |TX |
|1002 |76177 |STANDARD |FORT WORTH |TX |
|1002 |76177 |STANDARD |FT WORTH |TX |
|1003 |704 |STANDARD |URB EUGENE RICE |PR |
|1003 |85209 |STANDARD |MESA |AZ |
|1004 |85210 |STANDARD |MESA |AZ |
|1004 |32046 |STANDARD |HILLIARD |FL |
----- ------- ----------- ------------------- -----
For each empID need to print the column names for which values that are different.
----- ---------------------------------
|empID|nonMatchingColumnNames |
----- ---------------------------------
|1002 |City |
|1000 |City |
|1001 |State, City, ZipCodeType, Zipcode|
|1003 |State, City, Zipcode |
|1004 |State, City, Zipcode |
----- ---------------------------------
The strategy I have taken is, build a struct and collect set all the values. Check if the count of each set is > 1, then print the column name. Here is my code
val schema = new StructType()
.add("empID", IntegerType, true)
.add("Zipcode", StringType, true)
.add("ZipCodeType", StringType, true)
.add("City", StringType, true)
.add("State", StringType, true)
val idColumn = "empID"
val dfJSON = dfFromText.withColumn("jsonData",from_json(col("value"),schema))
.select("jsonData.*")
dfJSON.printSchema()
dfJSON.show(false)
val aggMap = dfJSON.columns
.filterNot(x => x == idColumn)
.map(colName => (collect_set(colName).alias(s"${colName}_asList"), s"${colName}_asList"))
aggMap.foreach(println)
val aggMapColumns = aggMap.map(x => x._1)
val columnsAsList = dfJSON.groupBy(col(idColumn)).agg(aggMapColumns.head, aggMapColumns.tail : _ *)
columnsAsList.show(false)
val combinedDF = columnsAsList.select(col(idColumn), struct(
aggMap.map(x => col(x._2)) : _ * ).alias("combined_struct")
)
combinedDF.printSchema()
combinedDF.show(false)
val columnsToCompare = dfJSON.columns.filterNot(x => x == idColumn).zipWithIndex.map({ case (x,y) => (y,x)})
val output = combinedDF.rdd.map({row => {
val empNo = row.getAs[Int](0)
val conbinedStruct: Row = row.getAs[AnyRef]("combined_struct").asInstanceOf[Row]
val nonMatchingColumns = columnsToCompare.foldLeft(List[String]())((acc, item) => {
val counts = conbinedStruct.getAs[Seq[String]](item._1).length
if (counts == 1) acc else item._2 :: acc
})
(empNo, nonMatchingColumns.mkString(", "))
}}).toDF(idColumn, "nonMatchingColumnNames")
output.show(false)
It works perfectly fine in my local machine, when I port it to spark-shell (it is an adhoc query), I am getting null pointer exception when I am trying to convert the dataframe into RDD and iterate through each item in the struct.
CodePudding user response:
You can use only spark's builtin functions to get a string containing the list of columns whose value is not unique:
- use
countDistinct
to determine whether there are several values in a specific column for a specificempID
- save name of the column if count distinct is greater than 2 using
when
- iterate over columns and save this iteration into an array using
array
- build a string from this array using
concat_ws
The complete code is as below:
import org.apache.spark.sql.functions.{array, concat_ws, countDistinct, lit, when}
val output = dfJSON.groupBy("empID").agg(
concat_ws(
", ",
array(dfJSON.columns.filter(_ != "empID").map(c => when(countDistinct(c) > 1, lit(c))): _*)
).as("nonMatchingColumnNames")
)
And with your input dataframe, you get the following output:
----- ---------------------------------
|empID|nonMatchingColumnNames |
----- ---------------------------------
|1002 |City |
|1000 |City |
|1001 |Zipcode, ZipCodeType, City, State|
|1003 |Zipcode, City, State |
|1004 |Zipcode, City, State |
----- ---------------------------------