I'm looking for an example-based answer, whether that's code directly in the answer or a link to a tutorial, but regardless more than a text-only answer.
I'm curious- how would one define an arbitrary Markov Decision Process in OpenAI Gym for purposes of reinforcement learning solutions? The sort of problem I see frequently in my role are traveling salesman, vehicle routing, and inventory optimization. Typically, I've used optimization techniques like genetic algorithms and bayesian optimization to find near optimal solutions. However, in this question, I'd like to see a practical/feasible RL approach to such problems. It's my understanding that OpenAI Gym is the simplest tool for defining an agent/environment for RL. (If this is not true, please answer this question with same level of detail using your preferred alternative.)