My introduction to SQL window functions

Last Monday, I started working @ Opinary (Berlin) as a Data Scientist. Wednesday, I was introduced to MySQL / BigQuery window functions. They’re likely to be one of the most useful (and one of my favourite) functions from now on. Let me very quickly describe what window functions are.

Imagine you have a database that looks like this:

name, age, salary, department
Brian, 24, 1000, Sales
Jonathan, 32, 2800, Engineering
Melissa, 40, 2500, Engineering
John, 25, 3000, Engineering
Lucy, 30, 2800, Sales
James, 28, 1500, Marketing

and you’re asked to calculate which are the best paid employees per department. To me, this called to GROUP BY deparment find the MAXimum salary and JOIN the resulting table back to the original one. In MySQL this corresponds to:

SELECT name, age, salary, department, max_department_salary = salary as
is_best_paid_department_employee FROM employees JOIN (SELECT MAX(salary) as max_department_salary
FROM employees GROUP BY deparment) USING(department)

This is all good and fine and would result in something like this:

query result
name, age, salary, department, is_best_paid_department_employee
Brian, 24, 1000, Sales, False
Jonathan, 32, 2800, Engineering, False
Melissa, 40, 2500, Engineering, False
John, 25, 3000, Engineering, True
Lucy, 30, 2800, Sales, True
James, 28, 1500, Marketing, True

Now, imagine your colleague asks you this (artificial) follow-up question: “Oh, can you calculate the second best paid employees?”. You think “Well, I could remove best paid employees and repeat the query again. It’s not super nice, but it should work”. Your invested colleague continues and asks: “Actually, can you tell me how each employee ranks in their department?”. That to me, sounded very complex… until I heard about window functions.

Window functions perform a calculation on a group of related rows. The simplest one: ROW_NUMBER(). This function simply assigns a row number to every row in the table or to a subset of them. We can, for example, assign row numbers to subgroups defined by a column like in a GROUP_BY using the PARTITION BY clause. The rows belonging to a group can also be sorted according to some criteria and then assigned a row number. In our case, we could do:

SELECT name, age, salary, department, ROW_NUMBER() OVER (PARTITION BY deparment ORDER BY salary DESC)
as nth_best_paid_department_employee FROM employees

which would return:

query result
name, age, salary, department, nth_best_paid_department_employee
Brian, 24, 1000, Sales, 2
Jonathan, 32, 2800, Engineering, 2
Melissa, 40, 2500, Engineering, 3
John, 25, 3000, Engineering, 1
Lucy, 30, 2800, Sales, 1
James, 28, 1500, Marketing, 1

ROW_NUMBER is a non-aggregate window function as it assigns an individual value to the rows that have been grouped (partitioned) as oppossed to aggregate ones using GROUP BY that are used to calculate summary statistics (max, min, count, etc) over the different groups. Here’s a link to other non-aggregate window functions and what they do!

EXTRA: What if several employees earned the same maximum amount?

Well, in this case, ROW_NUMBER would assign number 1 to an arbitrary employee and subsequent numbers to the rest. However, we can do better than this using another window function called RANK. RANK will assign the same number to employees with the same salary. For example, if our top paid employees earn [3000, 3000, 2500, 2500, 2300], RANK will return [1, 1, 3, 3, 4]. DENSE_RANK, a variation of this window function, will return [1, 1, 2, 2, 3].

Written on February 9, 2020