CODEWITHSUNDEEP
scalaapache-sparkapache-spark-sql
Anju
Anju

Reputation: 641

How to create a Row from a List or Array in Spark using Scala

I'm trying to create a Row (org.apache.spark.sql.catalyst.expressions.Row) based on the user input. I'm not able to create a Row randomly.

Is there any functionality to create a Row from List or Array.

For eg., If I have a .csv file with the following format,

"91xxxxxxxxxx,21.31,15,0,0"

If the user input [1, 2] then I need to take only 2nd column and 3rd column along with the customer_id which is the first column

I try to parse it with the code:

val l3 = sc.textFile("/SparkTest/abc.csv").map(_.split(" ")).map(r => (foo(input,r(0)))) `

where foo is defined as

def f(n: List[Int], s: String) : Row = {
    val n = input.length
    var out = new Array[Any](n+1)
    var r = s.split(",")
    out(0) = r(0)
    for (i <- 1 to n)
        out(i) = r(input(i-1)).toDouble
    Row(out)
}

and input is a List say 

val input = List(1,2)

Executing this code I get l3 as:

Array[org.apache.spark.sql.Row] = Array([[Ljava.lang.Object;@234d2916])

But what I want is:

Array[org.apache.spark.sql.catalyst.expressions.Row] = Array([9xxxxxxxxxx,21.31,15])`

This has to be passed to create a schema in Spark SQL

Upvotes: 11

Views: 59587

Answers (3)

Joel Mata
Joel Mata

Reputation: 514

You can also try:

    Row.fromSeq(line(0).toString ++ line(1).toDouble ++ line(2).toDouble ++ line.slice(2, line.size).map(value => value.toString))

Upvotes: 1

0x0FFF
0x0FFF

Reputation: 5018

You are missing creation of the StructField and StructType. Refer to the official guide http://spark.apache.org/docs/latest/sql-programming-guide.html, part Programmatically Specifying the Schema

I'm not a Scala specialist, but in Python it would look like this:

from pyspark.sql import *
sqlContext = SQLContext(sc)

input = [1,2]

def parse(line):
    global input
    l = line.split(',')
    res = [l[0]]
    for ind in input:
        res.append(l[ind])
    return res

csv  = sc.textFile("file:///tmp/inputfile.csv")
rows = csv.map(lambda x: parse(x))

fieldnum = len(input) + 1
fields = [StructField("col"+str(i), StringType(), True) for i in range(fieldnum)]
schema = StructType(fields)

csvWithSchema = sqlContext.applySchema(rows, schema)
csvWithSchema.registerTempTable("test")
sqlContext.sql("SELECT * FROM test").collect()

In short, you should not directly convert them to Row objects, just leave as RDD and apply schema to it with applySchema

Upvotes: 2

gruggie
gruggie

Reputation: 436

Something like the following should work:

import org.apache.spark.sql._

def f(n: List[Int], s: String) : Row =
  Row.fromSeq(s.split(",").zipWithIndex.collect{case (a,b) if n.contains(b) => a}.toSeq)

Upvotes: 18

Related Questions