The gym example demonstrates the AEA framework's flexibility with respect to Reinforcement Learning using OpenAI's gym framework.


There is no immediate use case for this example as you can train an RL agent without the AEA proxy layer just fine (and faster).

However, the example decouples the RL agent from the gym.Env allowing them to run in separate execution environments, potentially owned by different entities.

Preparation instructions


Follow the Preliminaries and Installation sections from the AEA quick start.

Download the necessary directories into your working directory:

svn export
svn export

Install the gym and numpy library.

pip install numpy gym

Demo instructions

Run the example

python examples/gym_ex/

Notice the usual RL setup, i.e. the fit method of the RL agent has the typical signature and a familiar implementation.

Note how demonstrates how easy it is to use an AEA agent as a proxy layer between an OpenAI gym.Env and a standard RL agent.

It is just one line of code to introduce the proxy agent and proxy environment!

from gyms.env import BanditNArmedRandom
from proxy.env import ProxyEnv
from rl.agent import RLAgent

if __name__ == "__main__":
    NB_GOODS = 10
    NB_STEPS = 4000

    # Use any gym.Env compatible environment:
    gym_env = BanditNArmedRandom(nb_bandits=NB_GOODS, nb_prices_per_bandit=NB_PRICES_PER_GOOD)

    # Pass the gym environment to a proxy environment:
    proxy_env = ProxyEnv(gym_env)

    # Use any RL agent compatible with the gym environment and call the fit method:
    rl_agent = RLAgent(nb_goods=NB_GOODS), nb_steps=NB_STEPS)