Apache Spark使用管道分隔的CSV文件 [英] Apache Spark working with pipe delimited CSV files

问题描述

我是Apache Spark的新手，正尝试将SchemaRDD与管道分隔的文本文件一起使用.我在Mac上使用Scala 10独立安装了Spark 1.5.2.我有一个包含以下代表性数据的CSV文件，并且我试图根据记录的第一个值(列)将以下内容分为4个不同的文件.我将不胜感激，对此我将给予任何帮助.

I am very new to Apache Spark and am trying to use SchemaRDD with my pipe delimited text file. I have a standalone installation of Spark 1.5.2 on my Mac using Scala 10. I have a CSV file with the following representative data and I am trying to split the following into 4 different files based on the first value (column) of the record. I would very much appreciate any help I can get with this.

1|1.8|20140801T081137|115810740
2|20140714T060000|335|22159892|3657|0.00|||181
2|20140714T061500|335|22159892|3657|0.00|||157
2|20140714T063000|335|22159892|3657|0.00|||156
2|20140714T064500|335|22159892|3657|0.00|||66
2|20140714T070000|335|22159892|3657|0.01|||633
2|20140714T071500|335|22159892|3657|0.01|||1087
3|34|Starz
3|35|VH1
3|36|CSPAN: Cable Satellite Public Affairs Network
3|37|Encore
3|278|CMT: Country Music Television
3|281|Telehit
4|625363|1852400|Matlock|9212|The Divorce
4|625719|1852400|Matlock|16|The Rat Pack
4|625849|1846952|Smallville|43|Calling

推荐答案

注意:您的csv文件在每行中没有相同数量的字段-无法按原样解析到DataFrame中.(SchemaRDD已重命名为DataFrame.)如果您的csv文件格式正确，则可以执行以下操作:

Note: Your csv file does not have the same number of fields in each row - this cannot be parsed as is into a DataFrame. (SchemaRDD has been renamed to DataFrame.) Here is something you can do if your csv file were well-formed:

使用--packages com.databricks:spark-csv_2.10:1.3.0启动spark-shell或spark-submit，以便轻松解析csv文件(

launch spark-shell or spark-submit with --packages com.databricks:spark-csv_2.10:1.3.0 in order to parse csv files easily (see here). In Scala, your code would be, assuming your csv file has a header - if yes, it is easier to refer to columns:

val df = sqlContext.read.format("com.databricks.spark.csv").option("header", "true").option("inferSchema", "true").option("delimiter", '|').load("/path/to/file.csv")
// assume 1st column has name col1
val df1 = df.filter( df("col1") === 1)  // 1st DataFrame
val df2 = df.filter( df("col1") === 2)  // 2nd DataFrame  etc... 

由于文件格式不正确，因此您必须以不同的方式解析每个不同的行，例如，执行以下操作:

Since your file is not well formed, you would have to parse each of the different lines differently, so for example, do the following:

val lines = sc.textFile("/path/to/file.csv")

case class RowRecord1( col1:Int, col2:Double, col3:String, col4:Int)
def parseRowRecord1( arr:Array[String]) = RowRecord1( arr(0).toInt, arr(1).toDouble, arr(2), arr(3).toInt)

case class RowRecord2( col1:Int, col2:String, col3:Int, col4:Int, col5:Int, col6:Double, col7:Int)
def parseRowRecord2( arr:Array[String]) = RowRecord2( arr(0).toInt, arr(1), arr(2).toInt, arr(3).toInt, arr(4).toInt, arr(5).toDouble, arr(8).toInt)

val df1 = lines.filter(_.startsWith("1")).map( _.split('|')).map( arr => parseRowRecord1( arr )).toDF
val df2 = lines.filter(_.startsWith("2")).map( _.split('|')).map( arr => parseRowRecord2( arr )).toDF

这篇关于Apache Spark使用管道分隔的CSV文件的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！