Magic: The Gathering is full of interesting ML challenges (derekrodriguez.dev)

The blog post by Derek Rodriguez highlights the challenges and opportunities in applying machine learning (ML) to enhance the Magic: The Gathering gaming experience. Rodriguez discusses three main projects, starting with the development of a recommender system for Commander decks. Utilizing a simple Non-negative Matrix Factorization (NMF) model, he aims to identify "good" cards that synergize well with given deck components. While initial results fall short of existing tools like EDHRec, Rodriguez is exploring a more complex ~100M parameter Set Transformer model, though it's facing difficulties with obscure cards. This shows the ongoing challenge of developing ML solutions that can interpret complex game dynamics effectively. Additionally, Rodriguez explores automating card inventory scanning with smartphone cameras, recognizing the potential of existing iOS technologies to enhance this functionality. While he reports challenges with traditional approaches, he sees promise in utilizing APIs like Google Cloud's Vertex AI for Optical Character Recognition (OCR) to improve card identification. Lastly, he addresses optimizing land curves in decks using probabilistic programming, suggesting that both casual and competitive players could benefit from enhanced deck-building tools. Overall, these initiatives underscore the significant potential for machine learning applications in gaming, pushing the boundaries of user engagement and experience in collectible card games.