Last year, I downloaded data on a bunch of League of Legends (wildly popular video game) games , with the intention of doing something cool with it. I thought through a few different things, and eventually decided to see if I could create a sort of visualization of how different champions (playable characters) are used.
I used a neural network that is really closely related to NLP word embeddings. Instead of predicting a word given the surrounding words, I tried to predict one champion given the other four on the team.
Check it out here.
I think that many skilled players could probably group champions similarly by hand, but this uses nothing but team compositions. With the right data, it could be run separately for different regions, skill levels, patches, etc.
I was light on technical details to keep it readable by different audiences, but let me know if you have any questions about it or any other comments!
[link][2 comments]