Inspiration
Our journey into the world of League of Legends (LoL) began as a challenge—a challenge to explore an unfamiliar domain. As a team that had never played LoL before, our inspiration was rooted in the desire to push our boundaries and learn something new.
What it does
The LoLPRMaster project delves into two exciting aspects: classification model training and the optimization of the Elo rating system parameters. Our system assesses player performance, and with the power of Particle Swarm Optimization (PSO) in AWS SageMaker, it fine-tunes the Elo rating parameters to provide an accurate representation of team skills.
How we built it
Our journey commenced with data exploration and model training. We crafted a classification model to evaluate player performance. Next, we harnessed AWS SageMaker for the implementation of Particle Swarm Optimization (PSO) to refine the Elo rating system's parameters. By bringing these two components together, we paved the way for the future of LoL ranking.
Challenges we ran into
While our project experienced numerous triumphs, challenges awaited, primarily in the deployment process, which presented the hardest problems we needed to overcome.
Accomplishments that we're proud of
Our proudest moments lie in the simplicity and effectiveness of our solution. We managed to take an unfamiliar domain and transform it into something understandable and actionable. This project gave us the chance to dig into research and obtain promising results.
What we learned
This journey was a valuable learning experience. We absorbed the intricacies of LoL, data science and AWS. We learned the possibilities that open up when you apply classification models and optimization techniques to real-world scenarios.
What's next for LoLPRMaster
During the project, we also developed a player performance estimation model. However, the complexity associated with implementing such an Elo system required more time than we had available when we decided to participate in the hackathon. Therefore, we opted to leave it for the next version.
Despite the fact that the prediction error may be slightly larger than expected, this project holds significant potential due to its decoupled structure. The estimator model can be easily replaced with any other statistical method you prefer to measure teams' and players' performances. Once these performances are measured, the weights for these performances (and the weights for tournament value and the weights for the days since the match, which weren't optimized in this project version) assigned to the Elo rating system can be periodically optimized, either globally or locally as needed. This adaptability allows the system to respond effectively to the ever-changing environment of League of Legends.
Log in or sign up for Devpost to join the conversation.