def env_creator():
if args.game.__package__.endswith('atari'):
if (args.game_name.startswith('foozpong') or
args.game_name.startswith('basketball_pong') or
args.game_name.startswith('volleyball_pong')
):
env = args.game.env(obs_type=args.atari_obs_type,
max_cycles=args.max_steps['atari'],
full_action_space=False,
num_players=2)
else:
env = args.game.env(obs_type=args.atari_obs_type,
full_action_space=False,
max_cycles=args.max_steps['atari'])
env = frame_skip_v0(env, args.atari_frame_skip_num)
env = frame_stack_v1(env, args.atari_frame_stack_num)
else:
env = args.game.env()
if args.game_name.startswith('rps'):
env = one_hot_obs_wrapper(env)
env = dtype_v0(env, dtype=float32)
env = pad_observations_v0(env)
env = pad_action_space_v0(env)
if args.game_name.startswith('connect_four') or args.game_name.startswith('tictactoe'):
env = FlattenEnvWrapper(env)
GAUSSIAN_STD = 1.0
assert abs(GAUSSIAN_STD - 1.0) < 1e-5, "must be 1.0, otherwise simple ensemble implementation is wrong"
env = LatentGaussianAugmentedEnvWrapper(env,
latent_parameter_dim=args.latent_para_dim,
gaussian_std=1.0,
use_dict_obs_space=args.use_dict_obs_space)
return env
def env_fn():
#env = gym.make("CartPole-v0")#
env = simple_spread_v2.parallel_env()
# print(env.action_spaces.values())
# exit(0)
env = supersuit.pad_observations_v0(env)
env = supersuit.pad_action_space_v0(env)
#env = supersuit.aec_wrappers.continuous_actions(env)
venv = pettingzoo_env_to_vec_env_v0(env)
return venv
def comm_env(**kwargs):
raw_env = base_env.raw_env(**kwargs)
# Set all agents to silent
for agent in raw_env.world.agents:
agent.silent = True
env = AssertOutOfBoundsWrapper(raw_env)
env = OrderEnforcingWrapper(env)
env = CommWrapper(env, comm_dict)
env = pad_observations_v0(env)
env = pad_action_space_v0(env)
env = _PettingZooEnv(env)
return env
def env_creator(args):
env = env_constr.env(
) #killable_knights=False, killable_archers=False)
resize_size = 84 if model == None else 32
env = supersuit.resize_v0(env,
resize_size,
resize_size,
linear_interp=True)
env = supersuit.color_reduction_v0(env)
env = supersuit.pad_action_space_v0(env)
env = supersuit.pad_observations_v0(env)
# env = supersuit.frame_stack_v0(env,2)
env = supersuit.dtype_v0(env, np.float32)
env = supersuit.normalize_obs_v0(env)
if model == "MLPModelV2":
env = supersuit.flatten_v0(env)
env = PettingZooEnv(env)
return env
def unwrapped_check(env):
env.reset()
agents = env.agents
if image_observation(env, agents):
env = max_observation_v0(env, 2)
env = color_reduction_v0(env, mode="full")
env = normalize_obs_v0(env)
if box_action(env, agents):
env = clip_actions_v0(env)
env = scale_actions_v0(env, 0.5)
if observation_homogenizable(env, agents):
env = pad_observations_v0(env)
env = frame_stack_v1(env, 2)
env = agent_indicator_v0(env)
env = black_death_v3(env)
if (not_dict_observation(env, agents)
and not_discrete_observation(env, agents)
and not_multibinary_observation(env, agents)):
env = dtype_v0(env, np.float16)
env = flatten_v0(env)
env = frame_skip_v0(env, 2)
if action_homogenizable(env, agents):
env = pad_action_space_v0(env)
env = clip_reward_v0(env, lower_bound=-1, upper_bound=1)
env = delay_observations_v0(env, 2)
env = sticky_actions_v0(env, 0.5)
env = nan_random_v0(env)
env = nan_zeros_v0(env)
assert env.unwrapped.__class__ == DummyEnv, f"Failed to unwrap {env}"
base_obs_space = {"a_{}".format(idx): Box(low=np.float32(0.0), high=np.float32(10.0), shape=[8, 8, 3]) for idx in range(2)}
base_act_spaces = {"a_{}".format(idx): Discrete(5) for idx in range(2)}
return PettingzooWrap(DummyEnv(base_obs, base_obs_space, base_act_spaces))
wrappers = [
supersuit.color_reduction_v0(new_dummy(), "R"),
supersuit.resize_v0(dtype_v0(new_dummy(), np.uint8), x_size=5, y_size=10),
supersuit.resize_v0(dtype_v0(new_dummy(), np.uint8), x_size=5, y_size=10, linear_interp=True),
supersuit.dtype_v0(new_dummy(), np.int32),
supersuit.flatten_v0(new_dummy()),
supersuit.reshape_v0(new_dummy(), (64, 3)),
supersuit.normalize_obs_v0(new_dummy(), env_min=-1, env_max=5.0),
supersuit.frame_stack_v1(new_dummy(), 8),
supersuit.pad_observations_v0(new_dummy()),
supersuit.pad_action_space_v0(new_dummy()),
supersuit.agent_indicator_v0(new_dummy(), True),
supersuit.agent_indicator_v0(new_dummy(), False),
supersuit.reward_lambda_v0(new_dummy(), lambda x: x / 10),
supersuit.clip_reward_v0(new_dummy()),
supersuit.clip_actions_v0(new_continuous_dummy()),
supersuit.frame_skip_v0(new_dummy(), 4),
supersuit.sticky_actions_v0(new_dummy(), 0.75),
supersuit.delay_observations_v0(new_dummy(), 3),
]
@pytest.mark.parametrize("env", wrappers)
def test_basic_wrappers(env):
env.seed(5)
supersuit.resize_v0(dtype_v0(knights_archers_zombies_v4.env(), np.uint8),
x_size=5,
y_size=10),
supersuit.resize_v0(dtype_v0(knights_archers_zombies_v4.env(), np.uint8),
x_size=5,
y_size=10,
linear_interp=True),
supersuit.dtype_v0(knights_archers_zombies_v4.env(), np.int32),
supersuit.flatten_v0(knights_archers_zombies_v4.env()),
supersuit.reshape_v0(knights_archers_zombies_v4.env(), (512 * 512, 3)),
supersuit.normalize_obs_v0(dtype_v0(knights_archers_zombies_v4.env(),
np.float32),
env_min=-1,
env_max=5.0),
supersuit.frame_stack_v1(knights_archers_zombies_v4.env(), 8),
supersuit.pad_observations_v0(knights_archers_zombies_v4.env()),
supersuit.pad_action_space_v0(knights_archers_zombies_v4.env()),
supersuit.black_death_v0(knights_archers_zombies_v4.env()),
supersuit.agent_indicator_v0(knights_archers_zombies_v4.env(), True),
supersuit.agent_indicator_v0(knights_archers_zombies_v4.env(), False),
supersuit.reward_lambda_v0(knights_archers_zombies_v4.env(),
lambda x: x / 10),
supersuit.clip_reward_v0(knights_archers_zombies_v4.env()),
supersuit.clip_actions_v0(prison_v2.env(continuous=True)),
supersuit.frame_skip_v0(knights_archers_zombies_v4.env(), 4),
supersuit.sticky_actions_v0(knights_archers_zombies_v4.env(), 0.75),
supersuit.delay_observations_v0(knights_archers_zombies_v4.env(), 3),
]
@pytest.mark.parametrize("env", wrappers)
def create_env(args):
if args[env_name] == "simple_speaker_listener":
env = simple_speaker_listener_v3.env()
env = supersuit.pad_action_space_v0(env)
env = supersuit.pad_observations_v0(env)
return env
def create_env(args):
env = simple_speaker_listener_v3.env()
env = pad_action_space_v0(env)
env = pad_observations_v0(env)
return env
def env(**kwargs):
env = base_env.parallel_env(**kwargs)
env = pad_observations_v0(env)
env = pad_action_space_v0(env)
env = _ParallelPettingZooEnv(env)
return env
0.75),
supersuit.delay_observations_v0(
generated_agents_parallel_v0.parallel_env(), 3),
supersuit.max_observation_v0(generated_agents_parallel_v0.parallel_env(),
3),
]
@pytest.mark.parametrize("env", parallel_wrappers)
def test_pettingzoo_parallel_api_gen(env):
parallel_test.parallel_api_test(env, num_cycles=50)
wrapper_fns = [
lambda: supersuit.pad_action_space_v0(generated_agents_parallel_v0.env()),
lambda: supersuit.pad_observations_v0(generated_agents_parallel_v0.env()),
lambda: supersuit.agent_indicator_v0(generated_agents_parallel_v0.env()),
lambda: supersuit.vectorize_aec_env_v0(generated_agents_parallel_v0.env(),
2),
lambda: supersuit.pad_action_space_v0(generated_agents_parallel_v0.
parallel_env()),
lambda: supersuit.pad_observations_v0(generated_agents_parallel_v0.
parallel_env()),
lambda: supersuit.agent_indicator_v0(generated_agents_parallel_v0.
parallel_env()),
lambda: supersuit.pettingzoo_env_to_vec_env_v1(generated_agents_parallel_v0
.parallel_env()),
]
@pytest.mark.parametrize("wrapper_fn", wrapper_fns)
dtype_v0(knights_archers_zombies_v10.env(vector_state=False),
np.uint8),
x_size=5,
y_size=10,
linear_interp=True,
),
supersuit.dtype_v0(knights_archers_zombies_v10.env(), np.int32),
supersuit.flatten_v0(knights_archers_zombies_v10.env()),
supersuit.reshape_v0(knights_archers_zombies_v10.env(vector_state=False),
(512 * 512, 3)),
supersuit.normalize_obs_v0(dtype_v0(knights_archers_zombies_v10.env(),
np.float32),
env_min=-1,
env_max=5.0),
supersuit.frame_stack_v1(combined_arms_v6.env(), 8),
supersuit.pad_observations_v0(simple_world_comm_v2.env()),
supersuit.pad_action_space_v0(simple_world_comm_v2.env()),
supersuit.black_death_v3(combined_arms_v6.env()),
supersuit.agent_indicator_v0(knights_archers_zombies_v10.env(), True),
supersuit.agent_indicator_v0(knights_archers_zombies_v10.env(), False),
supersuit.reward_lambda_v0(knights_archers_zombies_v10.env(),
lambda x: x / 10),
supersuit.clip_reward_v0(combined_arms_v6.env()),
supersuit.nan_noop_v0(knights_archers_zombies_v10.env(), 0),
supersuit.nan_zeros_v0(knights_archers_zombies_v10.env()),
supersuit.nan_random_v0(chess_v5.env()),
supersuit.nan_random_v0(knights_archers_zombies_v10.env()),
supersuit.frame_skip_v0(combined_arms_v6.env(), 4),
supersuit.sticky_actions_v0(combined_arms_v6.env(), 0.75),
supersuit.delay_observations_v0(combined_arms_v6.env(), 3),
supersuit.max_observation_v0(knights_archers_zombies_v10.env(), 3),