Overview of Windowing FunctionsΒΆ
Let us get an overview of Windowing Functions using Spark.
These are available as part of SQL in most of the traditional databases.
In some databases they are also known as Analytical Functions.
Let us start spark context for this Notebook so that we can execute the code provided. You can sign up for our 10 node state of the art cluster/labs to learn Spark SQL using our unique integrated LMS.
from pyspark.sql import SparkSession
import getpass
username = getpass.getuser()
spark = SparkSession. \
builder. \
config('spark.ui.port', '0'). \
config("spark.sql.warehouse.dir", f"/user/{username}/warehouse"). \
enableHiveSupport(). \
appName(f'{username} | Python - Windowing Functions'). \
master('yarn'). \
getOrCreate()
If you are going to use CLIs, you can use Spark SQL using one of the 3 approaches.
Using Spark SQL
spark2-sql \
--master yarn \
--conf spark.ui.port=0 \
--conf spark.sql.warehouse.dir=/user/${USER}/warehouse
Using Scala
spark2-shell \
--master yarn \
--conf spark.ui.port=0 \
--conf spark.sql.warehouse.dir=/user/${USER}/warehouse
Using Pyspark
pyspark2 \
--master yarn \
--conf spark.ui.port=0 \
--conf spark.sql.warehouse.dir=/user/${USER}/warehouse
First let us understand relevance of these functions using
employeesdata set.
employeesPath = '/public/hr_db/employees'
employees = spark. \
read. \
format('csv'). \
option('sep', '\t'). \
schema('''employee_id INT,
first_name STRING,
last_name STRING,
email STRING,
phone_number STRING,
hire_date STRING,
job_id STRING,
salary FLOAT,
commission_pct STRING,
manager_id STRING,
department_id STRING
'''). \
load(employeesPath)
employees.show()
+-----------+----------+----------+--------+------------------+----------+--------+-------+--------------+----------+-------------+
|employee_id|first_name| last_name| email| phone_number| hire_date| job_id| salary|commission_pct|manager_id|department_id|
+-----------+----------+----------+--------+------------------+----------+--------+-------+--------------+----------+-------------+
| 127| James| Landry| JLANDRY| 650.124.1334|1999-01-14|ST_CLERK| 2400.0| null| 120| 50|
| 128| Steven| Markle| SMARKLE| 650.124.1434|2000-03-08|ST_CLERK| 2200.0| null| 120| 50|
| 129| Laura| Bissot| LBISSOT| 650.124.5234|1997-08-20|ST_CLERK| 3300.0| null| 121| 50|
| 130| Mozhe| Atkinson|MATKINSO| 650.124.6234|1997-10-30|ST_CLERK| 2800.0| null| 121| 50|
| 131| James| Marlow| JAMRLOW| 650.124.7234|1997-02-16|ST_CLERK| 2500.0| null| 121| 50|
| 132| TJ| Olson| TJOLSON| 650.124.8234|1999-04-10|ST_CLERK| 2100.0| null| 121| 50|
| 133| Jason| Mallin| JMALLIN| 650.127.1934|1996-06-14|ST_CLERK| 3300.0| null| 122| 50|
| 134| Michael| Rogers| MROGERS| 650.127.1834|1998-08-26|ST_CLERK| 2900.0| null| 122| 50|
| 135| Ki| Gee| KGEE| 650.127.1734|1999-12-12|ST_CLERK| 2400.0| null| 122| 50|
| 136| Hazel|Philtanker|HPHILTAN| 650.127.1634|2000-02-06|ST_CLERK| 2200.0| null| 122| 50|
| 137| Renske| Ladwig| RLADWIG| 650.121.1234|1995-07-14|ST_CLERK| 3600.0| null| 123| 50|
| 138| Stephen| Stiles| SSTILES| 650.121.2034|1997-10-26|ST_CLERK| 3200.0| null| 123| 50|
| 139| John| Seo| JSEO| 650.121.2019|1998-02-12|ST_CLERK| 2700.0| null| 123| 50|
| 140| Joshua| Patel| JPATEL| 650.121.1834|1998-04-06|ST_CLERK| 2500.0| null| 123| 50|
| 141| Trenna| Rajs| TRAJS| 650.121.8009|1995-10-17|ST_CLERK| 3500.0| null| 124| 50|
| 142| Curtis| Davies| CDAVIES| 650.121.2994|1997-01-29|ST_CLERK| 3100.0| null| 124| 50|
| 143| Randall| Matos| RMATOS| 650.121.2874|1998-03-15|ST_CLERK| 2600.0| null| 124| 50|
| 144| Peter| Vargas| PVARGAS| 650.121.2004|1998-07-09|ST_CLERK| 2500.0| null| 124| 50|
| 145| John| Russell| JRUSSEL|011.44.1344.429268|1996-10-01| SA_MAN|14000.0| 0.40| 100| 80|
| 146| Karen| Partners|KPARTNER|011.44.1344.467268|1997-01-05| SA_MAN|13500.0| 0.30| 100| 80|
+-----------+----------+----------+--------+------------------+----------+--------+-------+--------------+----------+-------------+
only showing top 20 rows
employees.printSchema()
root
|-- employee_id: integer (nullable = true)
|-- first_name: string (nullable = true)
|-- last_name: string (nullable = true)
|-- email: string (nullable = true)
|-- phone_number: string (nullable = true)
|-- hire_date: string (nullable = true)
|-- job_id: string (nullable = true)
|-- salary: float (nullable = true)
|-- commission_pct: string (nullable = true)
|-- manager_id: string (nullable = true)
|-- department_id: string (nullable = true)
employees.count()
107
from pyspark.sql.functions import col
employees. \
select('employee_id',
col('department_id').cast('int').alias('department_id'),
'salary'
). \
orderBy('department_id', 'salary'). \
show()
+-----------+-------------+-------+
|employee_id|department_id| salary|
+-----------+-------------+-------+
| 178| null| 7000.0|
| 200| 10| 4400.0|
| 202| 20| 6000.0|
| 201| 20|13000.0|
| 119| 30| 2500.0|
| 118| 30| 2600.0|
| 117| 30| 2800.0|
| 116| 30| 2900.0|
| 115| 30| 3100.0|
| 114| 30|11000.0|
| 203| 40| 6500.0|
| 132| 50| 2100.0|
| 128| 50| 2200.0|
| 136| 50| 2200.0|
| 135| 50| 2400.0|
| 127| 50| 2400.0|
| 131| 50| 2500.0|
| 140| 50| 2500.0|
| 191| 50| 2500.0|
| 144| 50| 2500.0|
+-----------+-------------+-------+
only showing top 20 rows
Let us say we want to compare individual salary with department wise salary expense.
Here is one of the approach which require self join.
Compute department wise expense usig
groupByandagg.Join with employees again on department_id.
from pyspark.sql.functions import sum, col
department_expense = employees. \
groupBy('department_id'). \
agg(sum('salary').alias('expense'))
department_expense.show()
+-------------+--------+
|department_id| expense|
+-------------+--------+
| 30| 24900.0|
| 110| 20300.0|
| 100| 51600.0|
| 70| 10000.0|
| 90| 58000.0|
| 60| 28800.0|
| 40| 6500.0|
| 20| 19000.0|
| 10| 4400.0|
| 80|304500.0|
| null| 7000.0|
| 50|156400.0|
+-------------+--------+
employees. \
select('employee_id', 'department_id', 'salary'). \
join(department_expense, employees.department_id == department_expense.department_id). \
orderBy(employees.department_id, col('salary')). \
show()
+-----------+-------------+-------+-------------+--------+
|employee_id|department_id| salary|department_id| expense|
+-----------+-------------+-------+-------------+--------+
| 200| 10| 4400.0| 10| 4400.0|
| 113| 100| 6900.0| 100| 51600.0|
| 111| 100| 7700.0| 100| 51600.0|
| 112| 100| 7800.0| 100| 51600.0|
| 110| 100| 8200.0| 100| 51600.0|
| 109| 100| 9000.0| 100| 51600.0|
| 108| 100|12000.0| 100| 51600.0|
| 206| 110| 8300.0| 110| 20300.0|
| 205| 110|12000.0| 110| 20300.0|
| 202| 20| 6000.0| 20| 19000.0|
| 201| 20|13000.0| 20| 19000.0|
| 119| 30| 2500.0| 30| 24900.0|
| 118| 30| 2600.0| 30| 24900.0|
| 117| 30| 2800.0| 30| 24900.0|
| 116| 30| 2900.0| 30| 24900.0|
| 115| 30| 3100.0| 30| 24900.0|
| 114| 30|11000.0| 30| 24900.0|
| 203| 40| 6500.0| 40| 6500.0|
| 132| 50| 2100.0| 50|156400.0|
| 136| 50| 2200.0| 50|156400.0|
+-----------+-------------+-------+-------------+--------+
only showing top 20 rows
However, using this approach is not very efficient and also overly complicated. Windowing functions actually simplify the logic and also runs efficiently
Now let us get into the details related to Windowing functions.
Main package
pyspark.sql.windowIt has classes such as
WindowandWindowSpecWindowhave APIs such aspartitionBy,orderByetcThese APIs (such as
partitionBy) returnWindowSpecobject. We can passWindowSpecobject to over on functions such asrank(),dense_rank(),sum()etcSyntax:
sum().over(spec)wherespec = Window.partitionBy('ColumnName')
from pyspark.sql import window
help(window)
Help on module pyspark.sql.window in pyspark.sql:
NAME
pyspark.sql.window
DESCRIPTION
# Licensed to the Apache Software Foundation (ASF) under one or more
# contributor license agreements. See the NOTICE file distributed with
# this work for additional information regarding copyright ownership.
# The ASF licenses this file to You under the Apache License, Version 2.0
# (the "License"); you may not use this file except in compliance with
# the License. You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
CLASSES
builtins.object
Window
WindowSpec
class Window(builtins.object)
| Utility functions for defining window in DataFrames.
|
| For example:
|
| >>> # ORDER BY date ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW
| >>> window = Window.orderBy("date").rowsBetween(Window.unboundedPreceding, Window.currentRow)
|
| >>> # PARTITION BY country ORDER BY date RANGE BETWEEN 3 PRECEDING AND 3 FOLLOWING
| >>> window = Window.orderBy("date").partitionBy("country").rangeBetween(-3, 3)
|
| .. note:: When ordering is not defined, an unbounded window frame (rowFrame,
| unboundedPreceding, unboundedFollowing) is used by default. When ordering is defined,
| a growing window frame (rangeFrame, unboundedPreceding, currentRow) is used by default.
|
| .. note:: Experimental
|
| .. versionadded:: 1.4
|
| Static methods defined here:
|
| orderBy(*cols)
| Creates a :class:`WindowSpec` with the ordering defined.
|
| .. versionadded:: 1.4
|
| partitionBy(*cols)
| Creates a :class:`WindowSpec` with the partitioning defined.
|
| .. versionadded:: 1.4
|
| rangeBetween(start, end)
| Creates a :class:`WindowSpec` with the frame boundaries defined,
| from `start` (inclusive) to `end` (inclusive).
|
| Both `start` and `end` are relative from the current row. For example,
| "0" means "current row", while "-1" means one off before the current row,
| and "5" means the five off after the current row.
|
| We recommend users use ``Window.unboundedPreceding``, ``Window.unboundedFollowing``,
| and ``Window.currentRow`` to specify special boundary values, rather than using integral
| values directly.
|
| :param start: boundary start, inclusive.
| The frame is unbounded if this is ``Window.unboundedPreceding``, or
| any value less than or equal to max(-sys.maxsize, -9223372036854775808).
| :param end: boundary end, inclusive.
| The frame is unbounded if this is ``Window.unboundedFollowing``, or
| any value greater than or equal to min(sys.maxsize, 9223372036854775807).
|
| .. versionadded:: 2.1
|
| rowsBetween(start, end)
| Creates a :class:`WindowSpec` with the frame boundaries defined,
| from `start` (inclusive) to `end` (inclusive).
|
| Both `start` and `end` are relative positions from the current row.
| For example, "0" means "current row", while "-1" means the row before
| the current row, and "5" means the fifth row after the current row.
|
| We recommend users use ``Window.unboundedPreceding``, ``Window.unboundedFollowing``,
| and ``Window.currentRow`` to specify special boundary values, rather than using integral
| values directly.
|
| :param start: boundary start, inclusive.
| The frame is unbounded if this is ``Window.unboundedPreceding``, or
| any value less than or equal to -9223372036854775808.
| :param end: boundary end, inclusive.
| The frame is unbounded if this is ``Window.unboundedFollowing``, or
| any value greater than or equal to 9223372036854775807.
|
| .. versionadded:: 2.1
|
| ----------------------------------------------------------------------
| Data descriptors defined here:
|
| __dict__
| dictionary for instance variables (if defined)
|
| __weakref__
| list of weak references to the object (if defined)
|
| ----------------------------------------------------------------------
| Data and other attributes defined here:
|
| currentRow = 0
|
| unboundedFollowing = 9223372036854775807
|
| unboundedPreceding = -9223372036854775808
class WindowSpec(builtins.object)
| A window specification that defines the partitioning, ordering,
| and frame boundaries.
|
| Use the static methods in :class:`Window` to create a :class:`WindowSpec`.
|
| .. note:: Experimental
|
| .. versionadded:: 1.4
|
| Methods defined here:
|
| __init__(self, jspec)
| Initialize self. See help(type(self)) for accurate signature.
|
| orderBy(self, *cols)
| Defines the ordering columns in a :class:`WindowSpec`.
|
| :param cols: names of columns or expressions
|
| .. versionadded:: 1.4
|
| partitionBy(self, *cols)
| Defines the partitioning columns in a :class:`WindowSpec`.
|
| :param cols: names of columns or expressions
|
| .. versionadded:: 1.4
|
| rangeBetween(self, start, end)
| Defines the frame boundaries, from `start` (inclusive) to `end` (inclusive).
|
| Both `start` and `end` are relative from the current row. For example,
| "0" means "current row", while "-1" means one off before the current row,
| and "5" means the five off after the current row.
|
| We recommend users use ``Window.unboundedPreceding``, ``Window.unboundedFollowing``,
| and ``Window.currentRow`` to specify special boundary values, rather than using integral
| values directly.
|
| :param start: boundary start, inclusive.
| The frame is unbounded if this is ``Window.unboundedPreceding``, or
| any value less than or equal to max(-sys.maxsize, -9223372036854775808).
| :param end: boundary end, inclusive.
| The frame is unbounded if this is ``Window.unboundedFollowing``, or
| any value greater than or equal to min(sys.maxsize, 9223372036854775807).
|
| .. versionadded:: 1.4
|
| rowsBetween(self, start, end)
| Defines the frame boundaries, from `start` (inclusive) to `end` (inclusive).
|
| Both `start` and `end` are relative positions from the current row.
| For example, "0" means "current row", while "-1" means the row before
| the current row, and "5" means the fifth row after the current row.
|
| We recommend users use ``Window.unboundedPreceding``, ``Window.unboundedFollowing``,
| and ``Window.currentRow`` to specify special boundary values, rather than using integral
| values directly.
|
| :param start: boundary start, inclusive.
| The frame is unbounded if this is ``Window.unboundedPreceding``, or
| any value less than or equal to max(-sys.maxsize, -9223372036854775808).
| :param end: boundary end, inclusive.
| The frame is unbounded if this is ``Window.unboundedFollowing``, or
| any value greater than or equal to min(sys.maxsize, 9223372036854775807).
|
| .. versionadded:: 1.4
|
| ----------------------------------------------------------------------
| Data descriptors defined here:
|
| __dict__
| dictionary for instance variables (if defined)
|
| __weakref__
| list of weak references to the object (if defined)
DATA
__all__ = ['Window', 'WindowSpec']
FILE
/opt/spark-2.4.7-bin-hadoop2.7/python/pyspark/sql/window.py
Functions |
API or Function |
|---|---|
Aggregate Functions |
|
Ranking Functions |
|
Analytic Functions |
|