Has anyone here looked at using Deep Learning for Magic the Gathering? More specifically for:
- Building a better computer player AI using Deep Reinforcement Learning
- Creating new decks by training GANs on existing winning decks and using them to generate new decks and testing those decks out on the AIs trained in 1)
One of the biggest barriers (besides having countless card combinations to look through) is that there aren’t any good MTG simulators that work in Python (which rules out using Tensorflow, Pytorch, Fast.ai, etc. as the Deep Learning engine.) My favorite simulator is Forge which only works in Java and I guess it can be combined with Deeplearnin4j for this purpose.
What do you think? Has anyone ever attempted to use DL to create better MTG decks?