Scala-Spark(version1.5.2) Dataframes split error

Question

I have an input file foo.txt with the following content:

c1|c2|c3|c4|c5|c6|c7|c8|
00| |1.0|1.0|9|27.0|0||
01|2|3.0|4.0|1|10.0|1|1|

I want to transform it to a Dataframe to perform some Sql queries:

var text = sc.textFile("foo.txt")
var header = text.first()
var rdd = text.filter(row => row != header)
case class Data(c1: String, c2: String, c3: String, c4: String, c5: String, c6: String, c7: String, c8: String)

Until this point everything is ok, the problem comes in the next sentence:

var df = rdd.map(_.split("\\|")).map(p => Data(p(0), p(1), p(2), p(3), p(4), p(5), p(6), p(7))).toDF()

If I try to print df with df.show, I get an error message:

scala> df.show()
java.lang.ArrayIndexOutOfBoundsException: 7

I know that the error might be due to the split sentence. I also tried to split foo.txt using the following syntax:

var df = rdd.map(_.split("""|""")).map(p => Data(p(0), p(1), p(2), p(3), p(4), p(5), p(6), p(7))).toDF()

And then I get something like this:

scala> df.show()
+------+---------+----------+-----------+-----+-----------+----------------+----------------+
|  c1  |     c2  |    c3    |     c4    |  c5 |     c6    |        c7      |       c8       |
+------+---------+----------+-----------+-----+-----------+----------------+----------------+
|     0|        0|         ||           |    ||          1|               .|               0|
|     0|        1|         ||          2|    ||          3|               .|               0|
+------+---------+----------+-----------+-----+-----------+----------------+----------------+

Therefore, my question is how can I correctly pass this file to a Dataframe.

EDIT: The error is in the first row due to || field without an intermediate space. This type of field definition depending on the examples works fine or crashes.

Answer 1

This is because one of your lines is shorter than the others:

scala> var df = rdd.map(_.split("\\|")).map(_.length).collect()
df: Array[Int] = Array(7, 8)

You can fill in the rows manually (but you need to handle each case manually):

val df = rdd.map(_.split("\\|")).map{row =>
  row match {
    case Array(a,b,c,d,e,f,g,h) => Data(a,b,c,d,e,f,g,h)
    case Array(a,b,c,d,e,f,g) => Data(a,b,c,d,e,f,g," ")
  }
}

scala> df.show()
+---+---+---+---+---+----+---+---+
| c1| c2| c3| c4| c5|  c6| c7| c8|
+---+---+---+---+---+----+---+---+
| 00|   |1.0|1.0|  9|27.0|  0|   |
| 01|  2|3.0|4.0|  1|10.0|  1|  1|
+---+---+---+---+---+----+---+---+

EDIT:

A more generic solution would be something like this:

val df = rdd.map(_.split("\\|", -1)).map(_.slice(0,8)).map(p => Data(p(0), p(1), p(2), p(3), p(4), p(5), p(6), p(7))).toDF()

If you assume that you always have the right number of delimiters, it is safe to use this syntax an truncate the last value.

Answer 2

My suggestion would be to use databrick's csv parser.

Link : https://github.com/databricks/spark-csv

To load your example :

I loaded a sample file similar to yours:

c1|c2|c3|c4|c5|c6|c7|c8|
00| |1.0|1.0|9|27.0|0||
01|2|3.0|4.0|1|10.0|1|1|

To create the dataframe use the below code:

  val df = sqlContext.read
    .format("com.databricks.spark.csv")
    .option("header", "true") // Use first line of all files as header
    .option("inferSchema", "true") // Automatically infer data types
    .option("delimiter", "|") // default is ","
    .load("foo.txt")
    .show

I got the below output

+---+---+---+---+---+----+---+----+---+
| c1| c2| c3| c4| c5|  c6| c7|  c8|   |
+---+---+---+---+---+----+---+----+---+
|  0|   |1.0|1.0|  9|27.0|  0|null|   |
|  1|  2|3.0|4.0|  1|10.0|  1|   1|   |
+---+---+---+---+---+----+---+----+---+

This way you do not have to bother about parsing the file yourself. You get a dataframe directly