how to concat the same column value to a new column with comma delimiters in spark

Question

The format of input data as follows:

+--------------------+-------------+--------------------+
|           date     |       user  |           product  |
+--------------------+-------------+--------------------+
|        2016-10-01  |        Tom  |           computer |
+--------------------+-------------+--------------------+
|        2016-10-01  |        Tom  |           iphone   |
+--------------------+-------------+--------------------+
|        2016-10-01  |       Jhon  |             book   |
+--------------------+-------------+--------------------+
|        2016-10-02  |        Tom  |             pen    |
+--------------------+-------------+--------------------+
|        2016-10-02  |       Jhon  |             milk   |
+--------------------+-------------+--------------------+

And the format of output as follows:

+-----------+-----------------------+
|     user  |        products       |
+-----------------------------------+
|     Tom   |   computer,iphone,pen |
+-----------------------------------+
|     Jhon  |          book,milk    |  
+-----------------------------------+

The output shows all products every user bought order by date.

I want to process these data using Spark, who Can you help me, please? Thank you.

Answer 1

Better to use map-reduceBykey() combination rather than groupBy.. Also assuming the data doesn't have the

#Read the data using val ordersRDD = sc.textFile("/file/path")
val ordersRDD = sc.parallelize( List(("2016-10-01","Tom","computer"), 
    ("2016-10-01","Tom","iphone"), 
    ("2016-10-01","Jhon","book"), 
    ("2016-10-02","Tom","pen"), 
    ("2016-10-02","Jhon","milk")))

#group by (date, user), sort by key & reduce by user & concatenate products
val dtusrGrpRDD = ordersRDD.map(rec => ((rec._2, rec._1), rec._3))
   .sortByKey().map(x=>(x._1._1, x._2))
   .reduceByKey((acc, v) => acc+","+v)

#if needed, make it to DF
scala> dtusrGrpRDD.toDF("user", "product").show()
+----+-------------------+
|user|            product|
+----+-------------------+
| Tom|computer,iphone,pen|
|Jhon|          book,milk|
+----+-------------------+

Answer 2

If you are using a HiveContext (which you should be):

Example using python:

from pyspark.sql.functions import collect_set

df = ... load your df ...
new_df = df.groupBy("user").agg(collect_set("product").alias("products"))

If you don't want the resulting list in products deduped, you can use collect_list instead.

Answer 3

For dataframes it is two-liner:

import org.apache.spark.sql.functions.collect_list
//collect_set nistead of collect_list if you don't want duplicates
val output =  join.groupBy("user").agg(collect_list($"product"))

GroupBy will give you a grouped user set post which you can iterate and collect_list or collect_set on the grouped dataset.