Pyspark Spark DataFrame - Aggregate and filter columns in map type column

问题

My DataFrame looks like:

| c1 | c2|  c3  |
|----+---+-------
| A  | b | 22:00| 
| A  | b | 23:00|
| A  | b | 09:00|
| A  | c | 22:00|
| B  | c | 09:30|

I would like to perform some aggregations and create a second DataFrame with 3 columns:

c1: is the column that I want to group by.

map_category_room_date: map type, key the c2 and value the lower/min value in c3.

cnt_orig: is the count on how many rows the original group had.

Result

|    c1    |  map_category_room_date | cnt_orig |
|----------+-------------------------+----------|
|   'A'    |{'b': 09:00, 'C': 22:00} |    4     |
|   'B'    |{'c': 09:30}             |    1     |

What aggregate functions can I use to archive this is the most simple way?

Thanks

回答1:

You can window function to generate the count, then use inbuilt functions to get the final dataframe you desire by doing to following

from pyspark.sql import Window
windowSpec = Window.partitionBy("c1")

from pyspark.sql import functions as F
df.withColumn("cnt_orig", count('c1').over(windowSpec)).orderBy('c3').groupBy("c1", "c2", "cnt_orig").agg(first('c3').as('c3'))
    .withColumn("c2", F.regexp_replace(F.regexp_replace(F.array($"c2", $"c3").cast(StringType), "[\\[\\]]", ""), ",", " : "))
      .groupBy("c1", "cnt_orig").agg(F.collect_list("c2").as('map_category_room_date'))

You should get the following result

+---+--------+----------------------+
|c1 |cnt_orig|map_category_room_date|
+---+--------+----------------------+
|A  |4       |[b : 09:00, c : 22:00]|
|b  |1       |[c : 09:00]           |
+---+--------+----------------------+

Scala way

working code to get the desired output in scala is

val windowSpec = Window.partitionBy("c1")

df.withColumn("cnt_orig", count("c1").over(windowSpec)).orderBy("c3").groupBy("c1", "c2", "cnt_orig").agg(first("c3").as("c3"))
    .withColumn("c2", regexp_replace(regexp_replace(array($"c2", $"c3").cast(StringType), "[\\[\\]]", ""), ",", " : "))
      .groupBy("c1", "cnt_orig").agg(collect_list("c2").as("map_category_room_date"))

来源：https://stackoverflow.com/questions/45445077/pyspark-spark-dataframe-aggregate-and-filter-columns-in-map-type-column

标签

python

apache-spark

dataframe

aggregate-functions