Spark DataFrame

Updated March 20, 2023

Introduction to Spark DataFrame

A spark dataframe can be said to be a distributed data collection organized into named columns and is also used to provide operations such as filtering, computation of aggregations, grouping, and can be used with Spark SQL. Dataframes can be created by using structured data files, existing RDDs, external databases, and Hive tables. It is basically termed and known as an abstraction layer which is built on top of RDD and is also followed by the dataset API, which was introduced in later versions of Spark (2.0 +). Moreover, the datasets were not introduced in Pyspark but only in Scala with Spark, but this was not the case in the case of Dataframes. Dataframes, popularly known as DFs, are logical columnar formats that make working with RDDs easier and more convenient, also making use of the same functions as RDDs in the same way. If you talk more on the conceptual level, it is equivalent to the relational tables along with good optimization features and techniques.

How to Create a DataFrame?

A DataFrame is generally created by any one of the mentioned methods. It can be created by making use of Hive tables, external databases, Structured data files or even in the case of existing RDDs. These all ways can create these named columns known as Dataframes used for the processing in Apache Spark. By making use of SQLContext or SparkSession, applications can be used to create Dataframes.

Spark DataFrame Operations

In Spark, a dataframe is the distribution and collection of an organized form of data into named columns which is equivalent to a relational database or a schema or a dataframe in a language such as R or python but along with a richer level of optimizations to be used. It is used to provide a specific domain kind of language that could be used for structured data manipulation.

The below mentioned are some basic Operations of Structured Data Processing by making use of Dataframes.

1. Reading a document which is of type: JSON: We would be making use of the command sqlContext.read.json.

Example: Let us suppose our filename is student.json, then our piece of code will look like:

val dfs = sqlContext.read.json("student.json")

Output: In this case, the output will be that the field names will be automatically taken from the file student.json.

2. Showing of Data: In order to see the data in the Spark dataframes, you will need to use the command:

dfs.show()

Example: Let us suppose our filename is student.json, then our piece of code will look like:

val dfs= sqlContext.read.json("student.json")
dfs.show()

Output: The student data will be present to you in a tabular format.

3. Using printSchema method: If you are interested to see the structure, i.e. schema of the dataframe, then make use of the following command: dfs.printSchema()

Example: Let us suppose our filename is student.json, then our piece of code will look like:

val dfs= sqlContext.read.json("student.json")
dfs.printSchema()

Output: The structure or the schema will be present to you

4. Use the select method: In order to use the select method, the following command will be used to fetch the names and columns from the list of dataframes.

dfs.select("column-name").show()

Example: Let us suppose our filename is student.json, then our piece of code will look like:

val dfs= sqlContext.read.json("student.json")
dfs.select("name").show()

Output: The values of the name column can be seen.

5. Using Age filter: The following command can be used to find the range of students whose age is more than 23 years.

dfs.filter(dfs("column-name") > value).show()

Example: Let us suppose our filename is student.json, then our piece of code will look like:

val dfs= sqlContext.read.json("student.json")
dfs.filter(dfs("age")>23).show()

Output: The filtered age for greater than 23 will appear in the results.

6. Using the groupBy method: The following method could be used to count the number of students who have the same age.

dfs.groupBy("column-name").count().show()

Example: Let us suppose our filename is student.json, then our piece of code will look like:

val dfs= sqlContext.read.json("student.json")
dfs.groupBy("age").count().show()

7. Using SQL function upon a SparkSession: It enables the application to execute SQL type queries programmatically and hence returns the result in the form of a dataframe.
spark.sql(query)

Example: Suppose we have to register the SQL dataframe as a temp view then:

df.createOrReplaceTempView("student")
sqlDF=spark.sql("select * from student")
sqlDF.show()

Output: A temporary view will be created by the name of the student, and a spark.sql will be applied on top of it to convert it into a dataframe.

8. Using SQL function upon a Spark Session for Global temporary view: This enables the application to execute SQL type queries programmatically and hence returns the result in the form of a dataframe.
spark.sql(query)

Example: Suppose we have to register the SQL dataframe as a temp view then:

df.createGlobalTempView("student")
park.sql("select * from global_temp.student").show()
spark.newSession().sql("Select * from global_temp.student").show()

Output: A temporary view will be created by the name of the student, and a spark.sql will be applied on top of it to convert it into a dataframe.

Advantages of Spark DataFrame

1.  The dataframe is the Data’s distributed collection, and therefore the data is organized in named column fashion.
2. They are more or less similar to the table in the case of relational databases and have a rich set of optimization.
3. Dataframes are used to empower the queries written in SQL and also the dataframe API
4. It can be used to process both structured as well as unstructured kinds of data.
5. The use of a catalyst optimizer makes optimization easy and effective.
6. The libraries are present in many languages such as Python, Scala, Java, and R.
7. This is used to provide strong compatibility with Hive and is used to run unmodified Hive queries on the already present hive warehouse.
8. It can scale very well, right from a few kbs on the personal system to many petabytes on the large clusters.
9. It is used to provide an easy level of integration with other big data technologies and frameworks.
10. The abstraction which they provide to RDDs is efficient and makes processing faster.

Conclusion – Spark DataFrame

In this post, you have learned a very critical feature of Apache Spark, which is the dataframes and their usage in the applications running today, along with operations and advantages. I hope you have liked our article. Stay tuned for more like these.

Recommended Articles

This has been a guide to Spark DataFrame. Here we discuss steps to create a DataFrame its advantages, and different operations of DataFrames along with the appropriate sample code. You can also go through our other suggested articles to learn more –

1. Spark Streaming
2. How to Install Spark
3. Career in Spark
4. Spark Interview Questions
5. Data Frames in R
6. 7 Different Types of Joins in Spark SQL (Examples)
7. PySpark SQL | Modules and Methods of PySpark SQL
8. Spark Components | Overview of Components of Spark
9. Complete Guide to Top Spark Tools