Aggregate functions in SQL are super dope. When combining these functions with clauses such as
GROUP BY and
HAVING, we discover ways to view our data from completely new perspectives. Instead of looking at the same old endless flat table, we can use these functions to give us entirely new insights; aggregate functions help us to understand bigger-picture things. Those things might include finding outliers in datasets, or simply figuring out which employee with a family to feed should be terminated, based on some arbitrary metric such as sales numbers.
With the basics of
JOINs under our belts, this is when SQL starts feel really, really powerful. Our plain two-dimensional tables suddenly gain this power to be combined, aggregated, folded on to themselves, expand infinitely outward as the universe itself, and even transcend into the fourth dimension.*
Our Base Aggregation Functions
First up, let's see what we mean by "aggregate functions" anyway. These simple functions provide us with a way to mathematically quantify what exactly is in our database. Aggregate functions are performed on table columns to give us the make-up of said column. On their own, they seem quite simple:
AVG: The average of a set of values in a column.
COUNT: Number of rows a column contains in a specified table or view.
MIN: The minimum value in a set of values.
MAX: The maximum value in a set of values.
SUM: The sum of values.
A particularly useful way of using aggregate functions on their own is when we'd like to know the number of
DISTINCT values. While aggregate values take all records into account, using
DISTINCT limits the data returned to specifically refer to unique values.
COUNT(column_name) will return the number of all records in a column, where
COUNT(DISTINCT column_name) will ignore counting records where the value in the counted column is repeated.
Using GROUP BY
GROUP BY statement is often used with aggregate functions (COUNT, MAX, MIN, SUM, AVG) to group the result-set by one or more columns.
To demonstrate how aggregate functions work moving forward, I'll be using a familiar database: the database which contains all the content for this very blog. Let's get a quick preview of what we're working with:
|Welcome to Hackers and Slackersemail@example.com||Welcome to Hackers and Slackers | Hackers and Slackers||Technology for badasses||2017-11-17 20:29:13||2018-07-25 02:06:02||2017-11-13 20:37:00||Technology for badasses.|
|Generating Tree Hierarchies with Treelibfirstname.lastname@example.org||Tree Hierarchies with Treelib | Hackers and Slackers||Treelib is a Python library that allows you to create a visual tree hierarchy: a simple plaintext representation of parent-child relationships.||2017-11-17 20:45:10||2019-03-28 09:02:39||2017-11-17 20:56:40||Using Python to visualize file hierarchies as trees.|
|About the Squad||about||https://email@example.com||About | Hackers and Slackers||Hackers and Slackers is a community which values technology, life, and improving the latter with the former.||2017-11-17 20:58:42||2019-04-22 08:47:02||2017-11-17 20:58:46||Hackers and Slackers is a community which values technology, life, and improving the latter with the former.|
|Join||join||https://firstname.lastname@example.org||Join | Hackers and Slackers||2017-11-17 20:59:05||2018-07-25 02:06:02||2017-11-17 21:03:06|
|Merge Sets of Data in Python Using Pandasemail@example.com||Merging Dataframes with Pandas | Hackers and Slackers||Perform merges of data similar to SQL JOINs using Python's Pandas library: the essential library for data analysis in Oython.||2017-11-18 00:09:32||2018-12-26 09:29:22||2017-11-18 00:22:25||Perform SQL-like merges of data using Python's Pandas.|
Item 1 on our agenda: we're going to use aggregates to find which authors have been posting most frequently:
SELECT COUNT(title), author_id FROM posts GROUP BY author_id;
And the result:
Oh look, a real-life data problem to solve! It seems like authors are represented in Ghost's posts table simply by their IDs. This isn't very useful. Luckily, we've already learned enough about JOINs to know we can fill in the missing information from the users table!
SELECT COUNT(posts.title), users.name FROM posts LEFT JOIN users ON (posts.author_id = users.id) GROUP BY users.id ORDER BY COUNT(posts.title) DESC;
Let's see the results this time around:
Now that's more like it! Matt is crushing the game with his Lynx Roundup series, with myself in second place. Max had respectable numbers for a moment but has presumably moved on to other hobbies, such as living his life.
For the remainder, well, I've got nothing to say other than we're hiring. We don't pay though. In fact, there's probably zero benefits to joining us.
Conditional Grouping With "HAVING"
HAVING is like the
WHERE of aggregations. We can't use
WHERE on aggregate values, so that's why
HAVING can't accept any conditional value, but instead it must accept a numerical conditional derived from a
GROUP BY. Perhaps this would be easier to visualize in a query:
SELECT tags.name, COUNT(DISTINCT posts_tags.post_id) FROM posts_tags LEFT JOIN tags ON tags.id = posts_tags.tag_id LEFT JOIN posts ON posts.id = posts_tags.post_id GROUP BY tags.id HAVING COUNT(DISTINCT posts_tags.post_id) > 10 ORDER BY COUNT(DISTINCT posts_tags.post_id) DESC;
In this scenario, we want to see which tags on our blog have the highest number of associated posts. The query is very similar to the one we made previously, only this time we have a special guest:
HAVING COUNT(DISTINCT posts_tags.post_id) > 10
This usage of
HAVING only gives us tags which have ten posts or more. This should clean up our report by letting Darwinism takes its course. Here's how it worked out:
|#Adventures in Excel||16|
As expected, Matt's roundup posts take the lead (and if we compare this to previous data, we can see Matt has made a total of 17 non-Lynx posts: meaning Max and Matt are officially TIED).
If we hadn't included our
HAVING statement, this list would be much longer, filled with tags nobody cares about. Thanks to explicit omission, now we don't need to experience the dark depression that comes when confronting those sad pathetic tags. Out of sight, out of mind.
To explore some of the other aggregates, we're going to switch datasets. This time, we're going to look at wind speeds across US cities:
|datetime||Vancouver||Portland||San Francisco||Seattle||Los Angeles||San Diego||Las Vegas||Phoenix||Albuquerque||Denver||San Antonio||Dallas||Houston||Kansas City||Minneapolis||Saint Louis||Chicago||Nashville||Indianapolis||Atlanta||Detroit||Jacksonville||Charlotte||Miami||Pittsburgh||Toronto||Philadelphia||New York||Montreal||Boston||Beersheba||Tel Aviv District||Eilat||Haifa||Nahariyya||Jerusalem|
Let's figure our if Chicago really is the windy city, shall we?
SELECT AVG(Chicago), AVG(`San Francisco`), AVG(`Los Angeles`), AVG (Seattle), AVG(`New York`), AVG(`Boston`), AVG(Vancouver), AVG(Miami) FROM wind_speed;
...Aaand the results!:
|AVG(Chicago)||AVG(`San Francisco`)||AVG(`Los Angeles`)||AVG (Seattle)||AVG(`New York`)||AVG(`Boston`)||AVG(Vancouver)||AVG(Miami)|
Wow, so it looks like (at first glance), Chicago really is the windiest city! I'm... not sure if I was expecting that, for some reason. Let's see the range of wind speeds in Chicago as well:
SELECT AVG(Chicago), MIN(Chicago), MAX(Chicago) FROM wind_speed;
Then lowest wind speed in our dataset for Chicago appears to be 0 (not shocking). On the flip side, the highest wind speed we have recorded for Chicago was 25mph! Wow! Isn't that... dangerous?
Aggregate functions aren't just about counting values or finding averages. Especially in Data Science, these functions are critical to drawing any statistical conclusions from data. That said, attention spans only last so long, and I'm not a scientist. Perhaps that can be your job.
Hackers and Slackers Newsletter
Join the newsletter to receive the latest updates in your inbox.