Home > Enterprise >  Best approach to apply RL to pre-built simulator?
Best approach to apply RL to pre-built simulator?

Time:11-15

I've built a simulator in python that emulates the process of accepting students into a university. It is fairly complex and I've based it off of real-world data so that I can run model-free reinforcement learning algorithms on it.

I've explored employing OpenAI gym, but it doesn't seem to make sense to use it with my already-built simulator, as I'd have to build an entirely new class reflecting the behavior of what I've already built, essentially re-writing it (and as I mentioned, it's fairly detailed).

I'm considering employing DQN in the simulator itself, but since it's not a graphic simulator, there are few resources available. Is there an easier way to do this with OpenAI gym, or is there another library/tool that might be of better use here?

CodePudding user response:

If you have to employ RL methods then you need to have some form of step(), reset(), observation, reward, done implemented. The best way would be to go with OpenAI gym. You can always leave the env.render() empty if it's not a graphic simulator.

  • Related