Site icon Spark By {Examples}

PySpark Convert DataFrame to RDD

Naveen Nelamali
pyspark convert DataFrame to RDD

PySpark dataFrameObject.rdd is used to convert PySpark DataFrame to RDD; there are several transformations that are not available in DataFrame but present in RDD hence you often required to convert PySpark DataFrame to RDD.

Since PySpark 1.3, it provides a property .rdd on DataFrame which returns the PySpark RDD class object of DataFrame (converts DataFrame to RDD). 


rddObj=df.rdd

Convert PySpark DataFrame to RDD

PySpark DataFrame is a list of Row objects, when you run df.rdd, it returns the value of type RDD<Row>, let’s see with an example. First create a simple DataFrame


data = [('James',3000),('Anna',4001),('Robert',6200)]
df = spark.createDataFrame(data,["name","salary"])
df.show()

#Displays
+------+------+
|  name|salary|
+------+------+
| James|  3000|
|  Anna|  4001|
|Robert|  6200|
+------+------+

Example

The below example converts DataFrame to RDD and displays the RDD after collect().


#converts DataFrame to rdd
rdd=df.rdd

print(rdd.collect())
#Displays
[Row(name='James', salary=3000), Row(name='Anna', salary=4001), Row(name='Robert', salary=6200)]

PySpark doesn’t have a partitionBy(), map(), mapPartitions() transformations and these are present in RDD so let’s see an example of converting DataFrame to RDD and applying map() transformation. Read PySpark map() vs mapPartitions() to learn the differences.


#Apply map() transformation
rdd2=df.rdd.map(lambda x: [x[0],x[1]*20/100])
print(rdd2.collect())

#Displays
[['James', 600.0], ['Anna', 800.2], ['Robert', 1240.0]]

In case if you wanted to Convert PySpark DataFrame to Python List

Below is an example of Convert RDD back to PySpark DataFrame by using toDF() function.


#Conver back to DataFrame
df2=rdd2.toDF(["name","bonus"])
df2.show()

#Displays
+------+------+
|  name| bonus|
+------+------+
| James| 600.0|
|  Anna| 800.2|
|Robert|1240.0|
+------+------+

Conclusion

Spark DataFrame doesn’t have methods like map(), mapPartitions() and partitionBy() instead they are available on RDD hence you often need to convert DataFrame to RDD and back to DataFrame.

Happy Learning !!

Exit mobile version