README.md

Gym Server

A remote gym-like training environment.

Inspired by Google Football Research Environment, this OpenAI Gym-like training environment can function like a local one.

This is designed to be extensible for many types of environments and supports an instance-based experimentation model that encapsulates model evaluation into separate environments.

[Definitely a work-in-progress.]

How It Works

This Gym server comes in two parts: the provision-er which manages experiment instances and the runtime, which actually runs them.

Eventually, this remote environment will serve as a testbed for asynchronous communication between an agent and an environment, only sending back new states after certain periods of time. This will allow RL agents to respond to having imperfect models of the world. (Like in "Learning to Predict Without Looking Ahead: World Models Without Forward Prediction")

Setup/Development

Install dependencies using pipenv:

pipenv install

Now activate the virtual environment:

pipenv shell

Now you're all set up. Run the Flask development server (Windows/Bash):

cd gym-server/gym_server
flask run

Deployment

This uses a Docker container. Deploy it to a solution that can handle containers like Google Kubernetes engine and provision static IPs.

An example is up at gym.williecubed.me. (Use the development server for now.)

Tidbits on Cloud Providers

If not deploying on Google Cloud, one can safely remove the google-cloud-logging client libraries from the dependencies using pipenv.