Gym Starter Kit

Gym Starter Kit simplifies Agent development for OpenAI Gym. Using gym-start command, you can run your agent quickly(do not have to write codes for environment). Therefore, you can foucus on agent development.

Getting Started

(Preparetion) Install OpenAI Gym

See Gym Installation.

Install gym-starter-kit

$ pip install gym-starter-kit

If you want to install development version, run as below:

$ git clone https://github.com/marevol/gym-starter-kit.git
$ cd gym-starter-kit
$ pip install .

Create Your Agent

Gym Starter Kit runs your agent in OpenAI Gym. For example, you can write the following agent in my_agent.py:

class MyAgent(object):
    def __init__(self, env):
        self.env = env

    # Return your action for the observation
    def act(self, observation):
        return self.env.action_space.sample()

    # Learn your action for the observation of act method
    def fit(self, observation, reward, done, info):
        pass

    # Called on Episode start
    def __enter__(self):
        return self

    # Called on Episode end
    def __exit__(self, exception_type, exception_value, traceback):
        pass

Run Your Agent

$ gym-start --agent my_agent.MyAgent

To specify an environment, use --env-id option:

$ gym-start --env-id Pong-v0

To check available options, run as below:

$ gym-start -h

Examples

See Gym Starter Kit Example.