Building RL environment simulators as a service

This recipe will walk you through the process of converting your RL training environment/simulator into a service. This will allow you to offer Simulation-as-a-Service for training RL agents!

So far, we have trained several RL agents in a variety of environments using different simulators depending on the task to be solved. The training scripts used the Open AI Gym interface to talk to the environment running in the same process, or locally in a different process. This recipe will guide you through the process of converting any OpenAI Gym-compatible training environment (including your custom RL training environments) into a service that can be deployed locally or remotely as a service. Once built and deployed, an agent training client can connect to the sim server and train one or more agents remotely.

As a concrete example, we will take our tradegym library, which is a collection of the RL training environments for cryptocurrency...