An external PySpark module that works like R's read.csv or Panda's read_csv, with automatic type inference and null value handling. Parses csv data into SparkSQL DataFrames. No installation required, simply include pyspark_csv.py via SparkContext.

Supports type inference by evaluating data within each column. In the case of column having multiple data types, pyspark-csv will assign the lowest common denominator type for that column. For example,

  Name,   Model,  Size, Width,  Dt
  Jag,    63,     4,    4,      '2014-12-23'
  Pog,    7.0,    5,    5,      '2014-12-23'
  Peek,   68xp,   5,    5.5,    ''

generates DataFrame with the following schema:

  csv_file 
  |--Name: string  
  |--Model: string
  |--Size: int
  |--Width: double
  |--Dt: timestamp

Required Python packages: pyspark, csv, dateutil

Assume we have the following context

  sc = SparkContext
  sqlCtx = SQLContext or HiveContext

First, distribute pyspark-csv.py to executors using SparkContext

import pyspark_csv as pycsv
sc.addPyFile('pyspark_csv.py')

Read csv data via SparkContext and convert it to DataFrame

plaintext_rdd = sc.textFile('hdfs://x.x.x.x/blah.csv')
dataframe = pycsv.csvToDataFrame(sqlCtx, plaintext_rdd)

By default, pyspark-csv parses the first line as column names. To supply your own column names

plaintext_rdd = sc.textFile('hdfs://x.x.x.x/blah.csv')
dataframe = pycsv.csvToDataFrame(sqlCtx, plaintext_rdd, columns=['Name','Model','Size','Width','Dt'])

To convert DataFrames to RDDs, call the .rdd method.

To change separator

plaintext_rdd = sc.textFile('hdfs://x.x.x.x/blah.csv')
datarame = pycsv.csvToDataFrame(sqlCtx, plaintext_rdd, sep=",")

Skipping date and time parsing can lead to significant performance gain on large datasets

plaintext_rdd = sc.textFile('hdfs://x.x.x.x/blah.csv')
dataframe = pycsv.csvToDataFrame(sqlCtx, plaintext_rdd, parseDate=False)

Currently, the following data types are supported:

• int
• double
• string
• date
• time
• datetime

It also recognises None, ?, NULL, and '' as null values

Contributors welcomed! Contact [email protected]