Spark - load CSV file as DataFrame?

Question

spark-csv is part of core Spark functionality and doesn't require a separate library. So you could just do for example

df = spark.read.format("csv").option("header", "true").load("csvfile.csv")

In scala,(this works for any format-in delimiter mention "," for csv, "\t" for tsv etc)

val df = sqlContext.read.format("com.databricks.spark.csv") .option("delimiter", ",") .load("csvfile.csv")

Parse CSV and load as DataFrame/DataSet with Spark 2.x

First, initialize SparkSession object by default it will available in shells as spark

val spark = org.apache.spark.sql.SparkSession.builder
        .master("local") # Change it as per your cluster
        .appName("Spark CSV Reader")
        .getOrCreate;

Use any one of the following ways to load CSV as DataFrame/DataSet

1. Do it in a programmatic way

 val df = spark.read
         .format("csv")
         .option("header", "true") //first line in file has headers
         .option("mode", "DROPMALFORMED")
         .load("hdfs:///csv/file/dir/file.csv")

Update: Adding all options from here in case the link will be broken in future

• path: location of files. Similar to Spark can accept standard Hadoop globbing expressions.
• header: when set to true the first line of files will be used to name columns and will not be included in data. All types will be assumed string. The default value is false.
• delimiter: by default columns are delimited using, but delimiter can be set to any character
• quote: by default the quote character is ", but can be set to any character. Delimiters inside quotes are ignored
• escape: by default, the escape character is , but can be set to any character. Escaped quote characters are ignored
• parserLib: by default, it is "commons" that can be set to "univocity" to use that library for CSV parsing.
• mode: determines the parsing mode. By default it is PERMISSIVE. Possible values are:
  • PERMISSIVE: tries to parse all lines: nulls are inserted for missing tokens and extra tokens are ignored.
  • DROPMALFORMED: drops lines that have fewer or more tokens than expected or tokens which do not match the schema
  • FAILFAST: aborts with a RuntimeException if encounters any malformed line charset: defaults to 'UTF-8' but can be set to other valid charset names
• inferSchema: automatically infers column types. It requires one extra pass over the data and is false by default comment: skip lines beginning with this character. Default is "#". Disable comments by setting this to null.
• nullValue: specifies a string that indicates a null value, any fields matching this string will be set as nulls in the DataFrame
• dateFormat: specifies a string that indicates the date format to use when reading dates or timestamps. Custom date formats follow the formats at java.text.SimpleDateFormat. This applies to both DateType and TimestampType. By default, it is null which means trying to parse times and date by java.sql.Timestamp.valueOf() and java.sql.Date.valueOf().

2. You can do this SQL way as well

 val df = spark.sql("SELECT * FROM csv.`hdfs:///csv/file/dir/file.csv`")

Dependencies:

 "org.apache.spark" % "spark-core_2.11" % 2.0.0,
 "org.apache.spark" % "spark-sql_2.11" % 2.0.0,

---

Spark version < 2.0

val df = sqlContext.read
    .format("com.databricks.spark.csv")
    .option("header", "true") 
    .option("mode", "DROPMALFORMED")
    .load("csv/file/path"); 

Dependencies:

"org.apache.spark" % "spark-sql_2.10" % 1.6.0,
"com.databricks" % "spark-csv_2.10" % 1.6.0,
"com.univocity" % "univocity-parsers" % LATEST,

It's for whose Hadoop is 2.6 and Spark is 1.6 and without "databricks" package.

import org.apache.spark.sql.types.{StructType,StructField,StringType,IntegerType};
import org.apache.spark.sql.Row;

val csv = sc.textFile("/path/to/file.csv")
val rows = csv.map(line => line.split(",").map(_.trim))
val header = rows.first
val data = rows.filter(_(0) != header(0))
val rdd = data.map(row => Row(row(0),row(1).toInt))

val schema = new StructType()
    .add(StructField("id", StringType, true))
    .add(StructField("val", IntegerType, true))

val df = sqlContext.createDataFrame(rdd, schema)

With Spark 2.0, following is how you can read CSV

val conf = new SparkConf().setMaster("local[2]").setAppName("my app")
val sc = new SparkContext(conf)
val sparkSession = SparkSession.builder
  .config(conf = conf)
  .appName("spark session example")
  .getOrCreate()

val path = "/Users/xxx/Downloads/usermsg.csv"
val base_df = sparkSession.read.option("header","true").
  csv(path)

In Java 1.8 This code snippet perfectly working to read CSV files

POM.xml

<dependency>
    <groupId>org.apache.spark</groupId>
    <artifactId>spark-core_2.11</artifactId>
    <version>2.0.0</version>
</dependency>
<!-- https://mvnrepository.com/artifact/org.apache.spark/spark-sql_2.10 -->
<dependency>
    <groupId>org.apache.spark</groupId>
    <artifactId>spark-sql_2.10</artifactId>
    <version>2.0.0</version>
</dependency>

<!-- https://mvnrepository.com/artifact/org.scala-lang/scala-library -->
<dependency>
    <groupId>org.scala-lang</groupId>
    <artifactId>scala-library</artifactId>
    <version>2.11.8</version>
</dependency>
<dependency>
    <groupId>com.databricks</groupId>
    <artifactId>spark-csv_2.10</artifactId>
    <version>1.4.0</version>
</dependency>

Java

SparkConf conf = new SparkConf().setAppName("JavaWordCount").setMaster("local");
// create Spark Context
SparkContext context = new SparkContext(conf);
// create spark Session
SparkSession sparkSession = new SparkSession(context);

Dataset<Row> df = sparkSession.read().format("com.databricks.spark.csv").option("header", true).option("inferSchema", true).load("hdfs://localhost:9000/usr/local/hadoop_data/loan_100.csv");

        //("hdfs://localhost:9000/usr/local/hadoop_data/loan_100.csv");
System.out.println("========== Print Schema ============");
df.printSchema();
System.out.println("========== Print Data ==============");
df.show();
System.out.println("========== Print title ==============");
df.select("title").show();

There are a lot of challenges to parsing a CSV file, it keeps adding up if the file size is bigger, if there are non-english/escape/separator/other characters in the column values, that could cause parsing errors.

The magic then is in the options that are used. The ones that worked for me and hope should cover most of the edge cases are in code below:

### Create a Spark Session
spark = SparkSession.builder.master("local").appName("Classify Urls").getOrCreate()

### Note the options that are used. You may have to tweak these in case of error
html_df = spark.read.csv(html_csv_file_path, 
                         header=True, 
                         multiLine=True, 
                         ignoreLeadingWhiteSpace=True, 
                         ignoreTrailingWhiteSpace=True, 
                         encoding="UTF-8",
                         sep=',',
                         quote='"', 
                         escape='"',
                         maxColumns=2,
                         inferSchema=True)

Hope that helps. For more refer: Using PySpark 2 to read CSV having HTML source code

Note: The code above is from Spark 2 API, where the CSV file reading API comes bundled with built-in packages of Spark installable.

Note: PySpark is a Python wrapper for Spark and shares the same API as Scala/Java.