In this article, we will discuss how to perform union on two dataframes with different amounts of columns in PySpark in Python.
Let’s consider the first dataframe
Here we are having 3 columns named id, name, and address.
Python3
# importing module import pyspark # import when and lit function from pyspark.sql.functions import when, lit # importing sparksession from pyspark.sql module from pyspark.sql import SparkSession # creating sparksession and giving an app name spark = SparkSession.builder.appName( 'sparkdf' ).getOrCreate() # list of employee data data = [[ "1" , "sravan" , "kakumanu" ], [ "2" , "ojaswi" , "hyd" ], [ "3" , "rohith" , "delhi" ], [ "4" , "sridevi" , "kakumanu" ], [ "5" , "bobby" , "guntur" ]] # specify column names columns = [ 'ID' , 'NAME' , 'Address' ] # creating a dataframe from the lists of data dataframe1 = spark.createDataFrame(data, columns) # display dataframe1.show() |
Output:
Let’s consider second dataframe
Here we are going to create dataframe with 2 columns
Python3
# importing module import pyspark # import when and lit function from pyspark.sql.functions import when, lit # importing sparksession from pyspark.sql module from pyspark.sql import SparkSession # creating sparksession and giving an app name spark = SparkSession.builder.appName( 'sparkdf' ).getOrCreate() # list of employee data data = [[ "1" , 23 ], [ "2" , 21 ], [ "3" , 32 ], ] # specify column names columns = [ 'ID' , 'Age' ] # creating a dataframe from the lists of data dataframe2 = spark.createDataFrame(data, columns) # display dataframe2.show() |
Output:
We can not perform union operations because the columns are different, so we have to add the missing columns. Here In first dataframe (dataframe1) , the columns [‘ID’, ‘NAME’, ‘Address’] and second dataframe (dataframe2 ) columns are [‘ID’,’Age’].
Now we have to add the Age column to the first dataframe and NAME and Address in the second dataframe, we can do this by using lit() function. This function is available in pyspark.sql.functions which is used to add a column with a value. Here we are going to add a value with None.
Syntax:
for column in [column for column in dataframe1.columns if column not in dataframe2.columns]:
dataframe2 = dataframe2.withColumn(column, lit(None))
where,
- dataframe1 is the firstdata frame
- dataframe2 is the second dataframe
Example: Add missing columns to both the dataframes
Python3
# importing module import pyspark # import lit function from pyspark.sql.functions import lit # importing sparksession from pyspark.sql module from pyspark.sql import SparkSession # creating sparksession and giving an app name spark = SparkSession.builder.appName( 'sparkdf' ).getOrCreate() # list of employee data data = [[ "1" , "sravan" , "kakumanu" ], [ "2" , "ojaswi" , "hyd" ], [ "3" , "rohith" , "delhi" ], [ "4" , "sridevi" , "kakumanu" ], [ "5" , "bobby" , "guntur" ]] # specify column names columns = [ 'ID' , 'NAME' , 'Address' ] # creating a dataframe from the lists of data dataframe1 = spark.createDataFrame(data, columns) # list of employee data data = [[ "1" , 23 ], [ "2" , 21 ], [ "3" , 32 ], ] # specify column names columns = [ 'ID' , 'Age' ] # creating a dataframe from the lists of data dataframe2 = spark.createDataFrame(data, columns) # add columns in dataframe1 that are missing from dataframe2 for column in [column for column in dataframe2.columns if column not in dataframe1.columns]: dataframe1 = dataframe1.withColumn(column, lit( None )) # add columns in dataframe2 that are missing from dataframe1 for column in [column for column in dataframe1.columns if column not in dataframe2.columns]: dataframe2 = dataframe2.withColumn(column, lit( None )) # now see the columns of dataframe1 print (dataframe1.columns) # now see the columns of dataframe2 print (dataframe2.columns) |
Output:
['ID', 'NAME', 'Address', 'Age'] ['ID', 'Age', 'NAME', 'Address']
Example 1: Using union()
Now we can perform union by using union() function. This function will join two dataframes.
Syntax: dataframe1.union(dataframe2)
Example:
Python3
# importing module import pyspark # import lit function from pyspark.sql.functions import lit # importing sparksession from pyspark.sql module from pyspark.sql import SparkSession # creating sparksession and giving an app name spark = SparkSession.builder.appName( 'sparkdf' ).getOrCreate() # list of employee data data = [[ "1" , "sravan" , "kakumanu" ], [ "2" , "ojaswi" , "hyd" ], [ "3" , "rohith" , "delhi" ], [ "4" , "sridevi" , "kakumanu" ], [ "5" , "bobby" , "guntur" ]] # specify column names columns = [ 'ID' , 'NAME' , 'Address' ] # creating a dataframe from the lists of data dataframe1 = spark.createDataFrame(data, columns) # list of employee data data = [[ "1" , 23 ], [ "2" , 21 ], [ "3" , 32 ], ] # specify column names columns = [ 'ID' , 'Age' ] # creating a dataframe from the lists of data dataframe2 = spark.createDataFrame(data, columns) # add columns in dataframe1 that are missing from dataframe2 for column in [column for column in dataframe2.columns if column not in dataframe1.columns]: dataframe1 = dataframe1.withColumn(column, lit( None )) # add columns in dataframe2 that are missing from dataframe1 for column in [column for column in dataframe1.columns if column not in dataframe2.columns]: dataframe2 = dataframe2.withColumn(column, lit( None )) # perform union dataframe1.union(dataframe2).show() |
Output:
Example 2: Using unionAll()
Syntax: dataframe1.unionAll(dataframe2)
Python3
# importing module import pyspark # import lit function from pyspark.sql.functions import lit # importing sparksession from pyspark.sql module from pyspark.sql import SparkSession # creating sparksession and giving an app name spark = SparkSession.builder.appName( 'sparkdf' ).getOrCreate() # list of employee data data = [[ "1" , "sravan" , "kakumanu" ], [ "2" , "ojaswi" , "hyd" ], [ "3" , "rohith" , "delhi" ], [ "4" , "sridevi" , "kakumanu" ], [ "5" , "bobby" , "guntur" ]] # specify column names columns = [ 'ID' , 'NAME' , 'Address' ] # creating a dataframe from the lists of data dataframe1 = spark.createDataFrame(data, columns) # list of employee data data = [[ "1" , 23 ], [ "2" , 21 ], [ "3" , 32 ], ] # specify column names columns = [ 'ID' , 'Age' ] # creating a dataframe from the lists of data dataframe2 = spark.createDataFrame(data, columns) # add columns in dataframe1 that are missing # from dataframe2 for column in [column for column in dataframe2.columns\ if column not in dataframe1.columns]: dataframe1 = dataframe1.withColumn(column, lit( None )) # add columns in dataframe2 that are missing # from dataframe1 for column in [column for column in dataframe1.columns \ if column not in dataframe2.columns]: dataframe2 = dataframe2.withColumn(column, lit( None )) # perform unionAll dataframe1.unionAll(dataframe2).show() |
Output:
Example 3: Using unionByName
We can also perform unionByName, This will join dataframes by name.
Syntax: dataframe1.unionByName(dataframe2)
Example:
Python3
# importing module import pyspark # import lit function from pyspark.sql.functions import lit # importing sparksession from pyspark.sql module from pyspark.sql import SparkSession # creating sparksession and giving an app name spark = SparkSession.builder.appName( 'sparkdf' ).getOrCreate() # list of employee data data = [[ "1" , "sravan" , "kakumanu" ], [ "2" , "ojaswi" , "hyd" ], [ "3" , "rohith" , "delhi" ], [ "4" , "sridevi" , "kakumanu" ], [ "5" , "bobby" , "guntur" ]] # specify column names columns = [ 'ID' , 'NAME' , 'Address' ] # creating a dataframe from the lists of data dataframe1 = spark.createDataFrame(data, columns) # list of employee data data = [[ "1" , 23 ], [ "2" , 21 ], [ "3" , 32 ], ] # specify column names columns = [ 'ID' , 'Age' ] # creating a dataframe from the lists of data dataframe2 = spark.createDataFrame(data, columns) # add columns in dataframe1 that are missing from dataframe2 for column in [column for column in dataframe2.columns \ if column not in dataframe1.columns]: dataframe1 = dataframe1.withColumn(column, lit( None )) # add columns in dataframe2 that are missing from dataframe1 for column in [column for column in dataframe1.columns \ if column not in dataframe2.columns]: dataframe2 = dataframe2.withColumn(column, lit( None )) # perform unionByName dataframe1.unionByName(dataframe2).show() |
Output: