So, if you ever find yourself working with multiple window functions in BigQuery, leverage the named windows specification for tidier, leaner code.
Say you have a function like:
DENSE_RANK() OVER (PARTITION BY country ORDER BY hire_date).
You can extract the (PARTITION ... ORDER...) part to a named window and then reuse this partitioning window for other calculations, thus having to write less code.
Let's look at a practical example of how it's done.
Here's our input data:
+-------------+------------+-----------+--------------+------------+--------------------+----------+---------+
| employee_id | first_name | last_name | gross_salary | hire_date | position | store_id | country |
+-------------+------------+-----------+--------------+------------+--------------------+----------+---------+
| F2 | Jane | Doe | 75000 | 2020-01-01 | Senior Salesperson | F100 | France |
| F4 | Callum | Blake | 55000 | 2023-06-01 | Junior Salesperson | F100 | France |
| F1 | Jack | Dew | 77000 | 2019-03-01 | Salesperson | F200 | France |
| F3 | Emily | Scott | 80000 | 2021-01-01 | Salesperson | F200 | France |
| G4 | Josh | Woods | 85000 | 2020-01-01 | Senior Salesperson | G100 | Germany |
| G5 | Callum | Blake | 65000 | 2022-09-01 | Salesperson | G100 | Germany |
| G2 | Johann | Frost | 87000 | 2017-06-01 | Senior Salesperson | G200 | Germany |
| G3 | Francis | Drake | 80000 | 2018-04-01 | Senior Salesperson | G200 | Germany |
| G1 | Germain | Muller | 90000 | 2017-04-01 | Senior Salesperson | G200 | Germany |
+-------------+------------+-----------+--------------+------------+--------------------+----------+---------+
Now, there's a series of metrics we'd need to compute for comparison with each employee. We'd ideally need WINDOW functions for these:
an average gross salary per store
a count of employees per store
a maximum gross salary paid for the same position in the respective country
an ordering of employees in each country based on hire date
Instead of writing multiple time something like OVER (PARTITION BY store_id) , we can give this partitioning window a name - in this case store_window and define it in the WINDOW specification, as shown below.
Note that calculations that use the same window definition can re-use the same named window.
SELECT
first_name, last_name, gross_salary, hire_date, position, store_id, country,
AVG(gross_salary) OVER store_window AS avg_salary_per_store,
COUNT(DISTINCT employee_id) OVER store_window AS no_employees_in_store,
MAX(gross_salary) OVER country_position_window AS max_salary_country_same_position,
DENSE_RANK() OVER country_tenure_window AS order_of_employement_country
FROM employees
WINDOW store_window AS (PARTITION BY store_id),
country_store_window AS (PARTITION BY country, store_id),
country_position_window AS (PARTITION BY country, position),
country_tenure_window AS (PARTITION BY country ORDER BY hire_date)
See below the output of this query.
+------------+-----------+--------------+------------+--------------------+----------+---------+----------------------+-----------------------+----------------------------------+-----------------------------+
| first_name | last_name | gross_salary | hire_date | position | store_id | country | avg_salary_per_store | no_employees_in_store | max_salary_country_same_position | order_of_employment_country |
+------------+-----------+--------------+------------+--------------------+----------+---------+----------------------+-----------------------+----------------------------------+-----------------------------+
| Jack | Dew | 77000 | 2019-03-01 | Salesperson | F200 | France | 78500.0 | 2 | 80000 | 1 |
| Jane | Doe | 75000 | 2020-01-01 | Senior Salesperson | F100 | France | 65000.0 | 2 | 75000 | 2 |
| Emily | Scott | 80000 | 2021-01-01 | Salesperson | F200 | France | 78500.0 | 2 | 80000 | 3 |
| Callum | Blake | 55000 | 2023-06-01 | Junior Salesperson | F100 | France | 65000.0 | 2 | 55000 | 4 |
| Germain | Muller | 90000 | 2017-04-01 | Senior Salesperson | G200 | Germany | 85666.666666666672 | 3 | 90000 | 1 |
| Johann | Frost | 87000 | 2017-06-01 | Senior Salesperson | G200 | Germany | 85666.666666666672 | 3 | 90000 | 2 |
| Francis | Drake | 80000 | 2018-04-01 | Senior Salesperson | G200 | Germany | 85666.666666666672 | 3 | 90000 | 3 |
| Josh | Woods | 85000 | 2020-01-01 | Senior Salesperson | G100 | Germany | 75000.0 | 2 | 90000 | 4 |
| Callum | Blake | 65000 | 2022-09-01 | Salesperson | G100 | Germany | 75000.0 | 2 | 65000 | 5 |
+------------+-----------+--------------+------------+--------------------+----------+---------+----------------------+-----------------------+----------------------------------+-----------------------------+
As you've seen from the above, the named WINDOW approach is pretty handy and clean, as opposed to individually specifying the window in the function call.
Thanks for reading!
Found it useful? Subscribe to my Analytics newsletter at notjustsql.com.