GitHub for the following article can be found here

As of today (October 2020) there are 151 “Champions” (playable characters) in the online game, League of Legends, each offering a unique and individual play-style that has made it the single most popular e-sports in the world. Although the variety provides for an engaging competitive environment, it creates complexity for both new players and analysts alike. If every Champion is unique, how is a new player expected to understand the intricacies of every match-up or an analyst meant to summarise the performance of a player?

In April 2016 Riot (the company behind the game) attempted to help this issue by introducing “Champion Classes”. This consisted of 12 sub-classes that fit into 6 classes. These were hand-crafted by the development team at Riot and provided a good starting point for new players to become accustomed to the game. You’re loading into one of your first few games and see a character called “Thresh”, a quick check would indicate his class is a “Catcher” and so is another character: “Blitzcrank”, who you’ve played against before. Although you miss some finer points, you understand the general idea is to avoid being caught by them.

Copy of the original article on Classes

Updated Class List

However, these Classes quickly fell by the wayside and it is rare to find them mentioned in any content nowadays. This prompted me to answer a question: how would AI class each Champion? Would this agree with Riots interpretation, and if so what else can we learn about them? To do so, we will split the approach into four stages.

1. Data gathering, cleaning and feature creation.
2. Reduce the dimensionality of the data by extracting key signals.
3. Use a clustering algorithm to split the Champions into classes.
4. Analyse the Classes to determine trends & themes.

Pre-Note on Game Context (ignore if you’re familiar with the game):

If you are not familiar with League of Legends you will miss some context. If you’d like to become more familiar then you can read this introduction to the game. Otherwise, as a very minimum it will help to know it’s a 5v5 game where each time has a player in one of these roles: Top Laner, Middle Laner, Jungler, ADC, Support.

If you wish to apply a business spin to the article then translating “Champions” into “Customers” and data such as “Kills” and “Deaths” to “Purchases” and “Website Bounces” may help. You can jump to the bottom of the article to get some examples of it used in “the Real World”.

#data-science #dimensionality-reduction #clustering #league-of-legends #umap