def generate_experiments():
exp_list = list()
obs_config = SimpleObservationConfig(max_depth=2)
env_config = FlatlandConfig(
height=35,
width=35,
n_agents=1,
n_cities=3,
grid_distribution_of_cities=False,
max_rails_between_cities=2,
max_rail_in_cities=4,
observation_builder_config=obs_config,
random_seed=RANDOM_SEED,
)
controller_config = ActorCriticConfig(
actor_optimizer_config=AdamConfig(lr=1e-4),
critic_optimizer_config=AdamConfig(lr=1e-4),
gamma=0.99,
entropy_coeff=0.01,
actor_layers_sz=[256, 128],
critic_layers_sz=[256, 128],
)
torch.save(exp_list, "generated/exp_list")
def HordeOfAgents(random_seed):
return FlatlandConfig(
height=120,
width=80,
n_agents=50,
n_cities=10,
grid_distribution_of_cities=False,
max_rails_between_cities=2,
max_rail_in_cities=4,
observation_builder_config=None,
reward_config=None,
malfunction_rate=1. / 1000,
random_seed=random_seed,
greedy=True,
)
def SeveralAgents(random_seed):
return FlatlandConfig(
height=35,
width=35,
n_agents=5,
n_cities=3,
grid_distribution_of_cities=False,
max_rails_between_cities=2,
max_rail_in_cities=4,
observation_builder_config=None,
reward_config=None,
malfunction_rate=1. / 200,
random_seed=random_seed,
greedy=True,
)
def env6(random_seed):
return FlatlandConfig(
height=40,
width=60,
n_agents=80,
n_cities=9,
grid_distribution_of_cities=False,
max_rails_between_cities=2,
max_rail_in_cities=4,
observation_builder_config=None,
reward_config=None,
malfunction_rate=1. / 800,
random_seed=random_seed,
greedy=True,
)
n_cities = 2
x_dim = 25
y_dim = 25
else:
n_agents = n_agents + math.ceil(10**(len(str(n_agents)) - 1) * 0.75)
n_cities = n_agents // 10 + 2
x_dim = math.ceil(math.sqrt(150 * n_cities)) + 7
y_dim = x_dim
_test_env.append(
FlatlandConfig(
height=x_dim,
width=y_dim,
n_agents=n_agents,
n_cities=n_cities,
grid_distribution_of_cities=False,
max_rails_between_cities=2,
max_rail_in_cities=4,
observation_builder_config=None,
reward_config=None,
malfunction_rate=1. / 1500,
random_seed=None,
greedy=True,
))
def test_env(random_seed, i):
env = _test_env[i]
env.random_seed = random_seed
return env
def FewAgents(random_seed):
return FlatlandConfig()