League of Legends

Champ Recommendation Tool using a Graph Network

LeagueofLegends12 - Champ Recommendation Tool using a Graph Network

vttcji955p331 - Champ Recommendation Tool using a Graph Network

Graph Network of Champions

EDIT: Thanks for the Silver!

La Introduction

There are a handful of Champion recommendation tools out there, but a majority use "defined" relationships. This means that if you play an Assassin like Zed, it will recommend you an Assassin like Talon. This can be fairly useful for intra-lane recommendations, but what if you play Zed and want to know which ADC to play? Defined relationships become harder to make.

This is where a "Graph Network Recommendation" will reign supreme. Rather than taking "defined" relationships, like play-style, it instead finds what other Zed players are playing as an ADC – regardless of the "defined" relationships between those Champs. It then uses this connected graph to drive the recommendation engine.

Ooh Pretty Graph

I've visualised an interactive version of the graph to make it clearer:


However, at this point I should make it clear that this exact graph was not the one used in the recommendation engine, although the data is the same. I explain why I had to change it, as well as a short summary of the technical side in this medium article:


La Tool

Once the graph has been made we can use the Riot API to find your top played Champs and find which new (your higher Mastery Champs will be removed) Champs you might enjoy in each lane based on how close they are in the graph!

Try it for yourself here*:



I take no responsibility for your success/failures on these recommended Champions. Try a Normal game or five first.)

Closing Thoughts

I'm a Data Scientist by training and this was my first attempt at Web design – hence not the most visually appeasing site. For that I can only apologise!

Another approach was taken by Redditor JabrZer0 a few months ago. This was based on your performance on each Champ (win rate %) as opposed to frequency of play. You can check out the alternative method here: https://www.reddit.com/r/leagueoflegends/comments/av4at9/i_built_a_champion_recommender_to_help_you_branch/

As always, feedback is always appreciated. I'm happy to answer any questions here, although do check the Medium article first as I may have already answered it there.

JackWillz (EUW)

Bug Log

The tool seems to be recommending almost everyone Teemo top?! He has clearly found a bush to hide in within the code. Temporarily disabled Teemo recommendations until I've found out where he is hiding.

Error-1: Aware of a '500 Internal Serve Error' haunting some players with certain Champion pools. Looking into it (@16:30 BST). FIXED (@16:44 BST),

Error-2: Internal errors found caused by:

  • Pure one-tricks (Less than 10,000 Mastery on every Champion except 1)
  • Newer players (<10,000 Mastery on every Champion)
  • ADC mains with 25k+ experience on every ADC – the code is designed to remove Champions you already play often to give you "new" recommendations. Doesn't work if you've played them all already!

These aren't quick-fixes as I'll need to re-write a chunk to handle each of these scenarios. Afraid I have a works conference this evening so those suffering from the above, I must apologise and ask you check back later today (Hopefully around 01:00 BST) – hopefully all fixed by then!

Source: Original link

© Post "Champ Recommendation Tool using a Graph Network" for game League of Legends.

Top 10 Most Anticipated Video Games of 2020

2020 will have something to satisfy classic and modern gamers alike. To be eligible for the list, the game must be confirmed for 2020, or there should be good reason to expect its release in that year. Therefore, upcoming games with a mere announcement and no discernible release date will not be included.

Top 15 NEW Games of 2020 [FIRST HALF]

2020 has a ton to look forward to...in the video gaming world. Here are fifteen games we're looking forward to in the first half of 2020.

You Might Also Like

Leave a Reply

Your email address will not be published. Required fields are marked *