CODEWITHSUNDEEP
pythonscalaapache-sparkpysparkapache-spark-sql
user3735871
user3735871

Reputation: 317

Spark- count the percentage of one column after groupBy another

I have a dataframe as below. Trying to figure out how to calculate the percentage of each colour per product, and generate something like the expected output. I tried to use a window w= Window.partitionBy["prod_name","colour"] and do a count df.withColumn("cnt", F.count("colour").over("w"))but that's far from correct as that only count the number of the "colour" column. Could someone please help? Many thanks.

Input:

prod_name | colour 
---------------
A      | blue
A      | blue
A      | yellow
B      | green
B      | blue
C      | red

Output:

prod_name | colour | percentage
----------------------
A         | blue.  |  0.67. ---- as blue account for 2/3 of product A
A         | yellow.|  0.33  --- yellow account for 1/3 of product A
B         | green  |. 0.5  --- green account for 1/2 of product B
B         | blue.   | 0.5  --- blue account for 1/2 of product B
C         | red.    |  1  --- red account for 100% of product C

Upvotes: 2

Views: 643

Answers (2)

Surbhi Jain
Surbhi Jain

Reputation: 150

In pyspark, you can do like this if you want to use window function:-

df=df.withColumn("cntProduct",count("product_name").over(Window.partitionBy("product_name")))
df=df.withColumn("cntProduct_colour",count("colour").over(Window.partitionBy(["product_name","colour"])))
df=df.withColumn("required",df.cntProduct_colour/df.cntProduct)
df=df.select("product_name","colour","required").distinct()
df.show()

Upvotes: 4

Sivakumar
Sivakumar

Reputation: 1751

The following solution based on Scala snippet might help you,

Generate total count for each prod_name.

val productCountDF = inputDF
                       .groupBy("prod_name").agg(count("*") as "total_products")

Join the above DF with the original Input DF

// Assuming the data footprint of productCountDF is small enough to fit for broadcast join in order to avoid shuffle.
val newDF = inputDF
              .join(broadcast(productCountDF), Seq("prod_name"))

The above join will get the total product count for each prod_name.

val finalDF = newDF
   .groupBy("product_name", "total_products", "colour")
   .agg(count("*") as "total_products_per_colour")

finalDF
.withColumn("percentage", col("total_products_per_colour")/ col("total_products"))
.drop("total_products_per_colour", "total_products")
.show(false)

Upvotes: 2

Related Questions