Basic Usage

Installation

To install the DRAMA library: pip install drama-wrapper.

Note that the PyPI package does not include the examples and their corresponding dependencies. To run the reference implementations, clone this GitHub repository, execute pip install -r requirements.txt, and click through the Jupyter notebooks.

Initializing Restrictors and Environments

Using environments with DRAMA is similar to to PettingZoo but with an additional wrapper. The environment may stay unaltered and the wrapper is highly configurable. The restrictor is initialized similarly to agents in most common reinforcement learning frameworks.

from drama.restrictors import Restrictor
from drama.wrapper import RestrictionWrapper

env = ...
restrictor = Restrictor(...)
wrapper = RestrictionWrapper(env, restrictor)

Interacting With Environments

The interaction with the environment stays unaltered:

wrapper.reset()
for agent in wrapper.agent_iter():
    observation, reward, termination, truncation, info = wrapper.last()
    action = policies[agent](observation)
    wrapper.step(action)

However, note that the policies in the cycle must be extended by the restrictors which compute the restrictions as actions prior to the original agents.

Please refer to getting-started.ipynb for a first full example.