Learning PettingZoo Environments

API

Example of initializing environment:

from pettingzoo.butterfly import pistonball_v4
env = pistonball_v4.env()

Interacting with env:

env.reset()
for agent in env.agent_iter():
    observation, reward, done, info = env.last()
    action = policy(observation, agent)
    env.step(action)

Parallel API:

parallel_env = pistonball_v4.parallel_env()
observations = parallel_env.reset()
max_cycles = 500
for step in range(max_cycles):
    actions = {agent: policy(observations[agent], agent) for agent in parallel_env.agents}
    observations, rewards, dones, infos = parallel_env.step(actions)

Full API:

Attributes: agents, num_agents, possible_agents, max_num_agents, observation_spaces, action_spaces

Methods: render(mode='human'), seed(seed=None), close(), observation_space(agent), action_space(agent)

step(actions): receives a dictionary of actions keyed by the agent name. Returns the observation dictionary, reward dictionary, done dictionary, and info dictionary, where each dictionary is keyed by the agent.

reset(): resets the environment and returns a dictionary of observations (keyed by the agent name)

Manual control:

Often, you want to be able to play before trying to learn it to get a better feel for it. Some of our games directly support this:

from pettingzoo.butterfly import prison_v3
prison_v3.manual_control(<environment parameters>)

Random demo:

You can also easily get a quick impression of them by watching a random policy control all the actions:

from pettingzoo.utils import random_demo
random_demo(env, render=True, episodes=1)

Enviroonment Creation

Environments

• Atari
• Butterfly
• Classic
• MAgent
• MPE
• SISL

Credits

• pettingzoo website
• TDS | pettingzoo tutorial 1
• TDS | pettingzoo tutorial 2
• SOF | install box2d