Is MLflow really built only for supervised learning, or can it handle something like reinforcement learning? A customer asked me exactly this. They're running a contextual-bandit-style system on-premises and are thinking about moving it to the cloud ...
