def multi_agent_make_env(scenario_name, benchmark=False):
from multiagent import scenarios
from multiagent.environment import MultiAgentEnv
scenario = scenarios.load(scenario_name + '.py').Scenario()
world = scenarios.make_world()
if benchmark:
env = MultiAgentEnv(world, scenario.reset_world, scenario.reward,
scenario.observation, scenario.benchmark_data)
else:
env = MultiAgentEnv(world, scenario.reset_world, scenario.reward,
scenario.observation)
return env
'''define the model name'''
model_name = 'independent_ddpg'
'''define the scenario name'''
scenario_name = 'simple_spread'
'''define the special property'''
# independentArgs = namedtuple( 'independentArgs', [] )
aux_args = AuxArgs[model_name]()
alias = '_new_6'
'''load scenario from script'''
scenario = scenario.load(scenario_name+".py").Scenario()
'''create world'''
world = scenario.make_world()
'''create multiagent environment'''
env = MultiAgentEnv(world, scenario.reset_world, scenario.reward, scenario.observation, info_callback=None, shared_viewer=True)
env = GymWrapper(env)
MergeArgs = namedtuple('MergeArgs', Args._fields+AuxArgs[model_name]._fields)
# under offline trainer if set batch_size=replay_buffer_size=update_freq -> epoch update
args = Args(model_name=model_name,
agent_num=env.get_num_of_agents(),
hid_size=32,
obs_size=np.max(env.get_shape_of_obs()),
continuous=False,
action_dim=np.max(env.get_output_shape_of_act()),
init_std=0.1,
