对 Pyspark 数据帧进行分组和过滤

Question

I have a PySpark data Fram with 3 columns. Some rows are similar in 2 columns but not the third one, see below example.

----------------------------------------
first_name | last_name | requests_ID    |
----------------------------------------
Joe        | Smith     |[2,3]           |
---------------------------------------- 
Joe        | Smith     |[2,3,5,6]       |
---------------------------------------- 
Jim        | Bush      |[9,7]           |
---------------------------------------- 
Jim        | Bush      |[21]            |
---------------------------------------- 
Sarah      | Wood      |[2,3]           |
----------------------------------------   

I want to group-by the rows based on {first_name, last_name} columns and only have the row with the maximum number of {requests_ID}. So the results should be:

----------------------------------------
first_name | last_name | requests_ID    |
----------------------------------------
Joe        | Smith     |[2,3,5,6]       |
---------------------------------------- 
Jim        | Bush      |[9,7]           |
---------------------------------------- 
Sarah      | Wood      |[2,3]           |
---------------------------------------- 

I have tries different things like the following, but it gives me a nested array of both rows in the group-by and not the longest one.

gr_df = filtered_df.groupBy("first_name", "last_name").agg(F.collect_set("requests_ID").alias("requests_ID")) 

Here is the results I get:

----------------------------------------
first_name | last_name | requests_ID    |
----------------------------------------
Joe        | Smith     |[[9,7],[2,3,5,6]]|
---------------------------------------- 
Jim        | Bush      |[[9,7],[21]]    |
---------------------------------------- 
Sarah      | Wood      |[2,3]           |
---------------------------------------- 

Answer 1

To follow through with your current df that looks like this,

----------------------------------------
first_name | last_name | requests_ID    |
----------------------------------------
Joe        | Smith     |[[9,7],[2,3,5,6]]|
---------------------------------------- 
Jim        | Bush      |[[9,7],[21]]    |
---------------------------------------- 
Sarah      | Wood      |[2,3]           |
---------------------------------------- 

try this,

import pyspark.sql.functions as F
from pyspark.sql.types import IntegerType, ArrayType

def myfunc(x):
  temp = []
  for _ in x:
    temp.append(len(x))

  max_ind = temp.index(max(temp))

  return x[max_ind]

udf_extract = F.udf(myfunc, ArrayType(IntegerType()))

df = df.withColumn('new_requests_ID', udf_extract('requests_ID'))

#df.show()

or alternatively, without variable declaration,

import pyspark.sql.functions as F

@F.udf
def myfunc(x):
  temp = []
  for _ in x:
    temp.append(len(x))

  max_ind = temp.index(max(temp))

  return x[max_ind]

df = df.withColumn('new_requests_ID', myfunc('requests_ID'))

#df.show()

Answer 2

You can use size to determine the length of array column and the use window like below:

Imports and create sample DataFrame

import pyspark.sql.functions as f
from pyspark.sql.window import Window

df = spark.createDataFrame([('Joe','Smith',[2,3]),
('Joe','Smith',[2,3,5,6]),
('Jim','Bush',[9,7]),
('Jim','Bush',[21]),
('Sarah','Wood',[2,3])], ('first_name','last_name','requests_ID'))

Define window to get row number of requests_ID column in based on length of column in descending order.

Here, f.size("requests_ID") will give length of requests_ID column and desc() will sort it in descending order.

w_spec = Window().partitionBy("first_name", "last_name").orderBy(f.size("requests_ID").desc())

Apply window function and get first row.

df.withColumn("rn", f.row_number().over(w_spec)).where("rn ==1").drop("rn").show()
+----------+---------+------------+
|first_name|last_name| requests_ID|
+----------+---------+------------+
|       Jim|     Bush|      [9, 7]|
|     Sarah|     Wood|      [2, 3]|
|       Joe|    Smith|[2, 3, 5, 6]|
+----------+---------+------------+