Examples

Check available classes

import marl

# Check available agents
print("\n| Agents\t\t", list(marl.agent.available()))

# Check available agents
print("\n| Policies\t\t", list(marl.policy.available()))

# Check available agents
print("\n| Models\t\t", list(marl.model.available()))

# Check available exploration process
print("\n| Expl. Processes\t", list(marl.exploration.available()))

# Check available experience memory
print("\n| Experience Memory\t", list(marl.experience.available()))

Single-agent example

Example for training a single agent with DQN algorithm.

import marl
from marl.agent import DQNAgent
from marl.model.nn import MlpNet

import gym

env = gym.make("LunarLander-v2")

obs_s = env.observation_space
act_s = env.action_space

mlp_model = MlpNet(8,4, hidden_size=[64, 32])

dqn_agent = DQNAgent(mlp_model, obs_s, act_s, experience="ReplayMemory-5000", exploration="EpsGreedy", lr=0.001, name="DQN-LunarLander")

# Train the agent for 100 000 timesteps
dqn_agent.learn(env, nb_timesteps=100000)

# Test the agent for 10 episodes
dqn_agent.test(env, nb_episodes=10)

Multi-agent example

Example for training a system composed of several agents with minimax-Q algorithm.

Warning

Most of the multi-agent algorithms requires external knowledge. It is necessary to specify to each of these agents their multi-agent system (MAS) by using ag.set_mas function.

import marl
from marl import MARL
from marl.agent import MinimaxQAgent
from marl.exploration import EpsGreedy

from soccer import DiscreteSoccerEnv
# Environment available here "https://github.com/blavad/soccer"
env = DiscreteSoccerEnv(nb_pl_team1=1, nb_pl_team2=1)

obs_s = env.observation_space
act_s = env.action_space

# Custom exploration process
expl = EpsGreedy(eps_deb=1.,eps_fin=.3)

# Create two minimax-Q agents
q_agent1 = MinimaxQAgent(obs_s, act_s, act_s, exploration=expl, gamma=0.9, lr=0.001, name="SoccerJ1")
q_agent2 = MinimaxQAgent(obs_s, act_s, act_s, exploration=expl, gamma=0.9, lr=0.001, name="SoccerJ2")

# Create the trainable multi-agent system
mas = MARL(agents_list=[q_agent1, q_agent2])

# Assign MAS to each agent
q_agent1.set_mas(mas)
q_agent2.set_mas(mas)

# Train the agent for 100 000 timesteps
mas.learn(env, nb_timesteps=100000)

# Test the agents for 10 episodes
mas.test(env, nb_episodes=10, time_laps=0.5)

Training two independant DQN agents

The environment HanabiMarlEnv is coming soon.

import marl
from marl.agent import DQNAgent

from hanabi_coop.env import HanabiMarlEnv # coming soon

config_hanabi = {   "players": 2,
                    "random_start_player": True,
                    "hand_size": 4,
                    "max_life_tokens": 3,
                    "max_information_tokens": 8,
                    "vectorized":[True,True]
                }

env = HanabiMarlEnv(config=config_hanabi)

obs_s = env.observation_space
act_s = env.action_space

ag1 = DQNAgent("MlpNet", obs_s, act_s, name="Bob")
ag2 = DQNAgent("MlpNet", obs_s, act_s, name="Jack")

mas = marl.MARL([ag1,ag2])

mas.learn(env, nb_timesteps=100000)