SharkNice walks us through a simple setup for training unit micro with reinforcement learning. Using a Stalker kiting a Roach as a proof of concept, he breaks down how to define observations and balance reward functions—providing an ML blueprint you can apply to any language or framework.