RyanSaxe/mtg

Employs transformer-based neural networks for both draft pick selection and deck construction, trained on Magic Arena gameplay data. Provides modular Python components for drafting and deckbuilding that can be independently trained and evaluated, though pretrained models must be generated from scratch using the included training scripts. Integrates with Magic Arena through a separate UI layer and leverages 17Lands dataset for analysis and model development.

179 stars. No commits in the last 6 months.