Business
Jobs
  • About Us
  • Solutions
    • Job Postings
      Post your job and receive qualified candidates in 48h.
    • Candidate Assessments
      500+ technical and psychological tests, plus anti-fraud.
    • Headhunting
      Tailor-made executive search from start to finish.
    • Payroll + EOR
      Payroll dispersal and EOR across 15+ LATAM countries.
  • Pricing
  • Jobs

0

1.4K
Views
GroupBy column and filter rows with maximum value in Pyspark

I am almost certain this has been asked before, but a search through stackoverflow did not answer my question. Not a duplicate of [2] since I want the maximum value, not the most frequent item. I am new to pyspark and trying to do something really simple: I want to groupBy column "A" and then only keep the row of each group that has the maximum value in column "B". Like this:

df_cleaned = df.groupBy("A").agg(F.max("B"))

Unfortunately, this throws away all other columns - df_cleaned only contains the columns "A" and the max value of B. How do I instead keep the rows? ("A", "B", "C"...)

over 4 years ago · Santiago Trujillo
3 answers
Answer question

0

You can do this without a udf using a Window.

Consider the following example:

import pyspark.sql.functions as f
data = [
    ('a', 5),
    ('a', 8),
    ('a', 7),
    ('b', 1),
    ('b', 3)
]
df = sqlCtx.createDataFrame(data, ["A", "B"])
df.show()
#+---+---+
#|  A|  B|
#+---+---+
#|  a|  5|
#|  a|  8|
#|  a|  7|
#|  b|  1|
#|  b|  3|
#+---+---+

Create a Window to partition by column A and use this to compute the maximum of each group. Then filter out the rows such that the value in column B is equal to the max.

from pyspark.sql import Window
w = Window.partitionBy('A')
df.withColumn('maxB', f.max('B').over(w))\
    .where(f.col('B') == f.col('maxB'))\
    .drop('maxB')\
    .show()
#+---+---+
#|  A|  B|
#+---+---+
#|  a|  8|
#|  b|  3|
#+---+---+

Or equivalently using pyspark-sql:

df.registerTempTable('table')
q = "SELECT A, B FROM (SELECT *, MAX(B) OVER (PARTITION BY A) AS maxB FROM table) M WHERE B = maxB"
sqlCtx.sql(q).show()
#+---+---+
#|  A|  B|
#+---+---+
#|  b|  3|
#|  a|  8|
#+---+---+
over 4 years ago · Santiago Trujillo Report

0

Another possible approach is to apply join the dataframe with itself specifying "leftsemi". This kind of join includes all columns from the dataframe on the left side and no columns on the right side.

For example:

import pyspark.sql.functions as f
data = [
    ('a', 5, 'c'),
    ('a', 8, 'd'),
    ('a', 7, 'e'),
    ('b', 1, 'f'),
    ('b', 3, 'g')
]
df = sqlContext.createDataFrame(data, ["A", "B", "C"])
df.show()
+---+---+---+
|  A|  B|  C|
+---+---+---+
|  a|  5|  c|
|  a|  8|  d|
|  a|  7|  e|
|  b|  1|  f|
|  b|  3|  g|
+---+---+---+

Max value of column B by by column A can be selected doing:

df.groupBy('A').agg(f.max('B')
+---+---+
|  A|  B|
+---+---+
|  a|  8|
|  b|  3|
+---+---+

Using this expression as a right side in a left semi join, and renaming the obtained column max(B) back to its original name B, we can obtain the result needed:

df.join(df.groupBy('A').agg(f.max('B').alias('B')),on='B',how='leftsemi').show()
+---+---+---+
|  B|  A|  C|
+---+---+---+
|  3|  b|  g|
|  8|  a|  d|
+---+---+---+

The physical plan behind this solution and the one from accepted answer are different and it is still not clear to me which one will perform better on large dataframes.

The same result can be obtained using spark SQL syntax doing:

df.registerTempTable('table')
q = '''SELECT *
FROM table a LEFT SEMI
JOIN (
    SELECT 
        A,
        max(B) as max_B
    FROM table
    GROUP BY A
    ) t
ON a.A=t.A AND a.B=t.max_B
'''
sqlContext.sql(q).show()
+---+---+---+
|  A|  B|  C|
+---+---+---+
|  b|  3|  g|
|  a|  8|  d|
+---+---+---+
over 4 years ago · Santiago Trujillo Report

0

just want to add scala spark version of @ndricca´s answer in case anyone needs it:

val data = Seq(("a", 5,"c"), ("a",8,"d"),("a",7,"e"),("b",1,"f"),("b",3,"g"))
val df = data.toDF("A","B","C")
df.show()
+---+---+---+
|  A|  B|  C|
+---+---+---+
|  a|  5|  c|
|  a|  8|  d|
|  a|  7|  e|
|  b|  1|  f|
|  b|  3|  g|
+---+---+---+

val rightdf = df.groupBy("A").max("B")
rightdf.show()
+---+------+
|  A|max(B)|
+---+------+
|  b|     3|
|  a|     8|
+---+------+

val resdf = df.join(rightdf, df("B") === rightdf("max(B)"), "leftsemi")
resdf.show()
+---+---+---+
|  A|  B|  C|
+---+---+---+
|  a|  8|  d|
|  b|  3|  g|
+---+---+---+

over 4 years ago · Santiago Trujillo Report
Answer question
Find remote jobs

Discover the new way to find a job!

Top jobs
Top job categories
Business
Post vacancy Pricing Sales
Legal
Terms and conditions Privacy policy
© 2026 PeakU Inc. All Rights Reserved.
Andres GPT
Show me some job opportunities
There's an error!