def __init__(self, args):
self.n_actions = args.n_actions
self.n_agents = args.n_agents
self.state_shape = args.state_shape
self.obs_shape = args.obs_shape
if args.alg == 'vdn':
self.policy = VDN(args)
elif args.alg == 'qmix':
self.policy = QMIX(args)
elif args.alg == 'coma':
self.policy = COMA(args)
elif args.alg == 'qtran_alt':
self.policy = QtranAlt(args)
elif args.alg == 'qtran_base':
self.policy = QtranBase(args)
elif args.alg == 'maven':
self.policy = MAVEN(args)
elif args.alg == 'central_v':
self.policy = CentralV(args)
elif args.alg == 'reinforce':
self.policy = Reinforce(args)
else:
raise Exception("No such algorithm")
self.args = args
print('Init Agents')
def __init__(self, args):
self.n_actions = args.n_actions
self.n_agents = args.n_agents
self.state_shape = args.state_shape
self.obs_shape = args.obs_shape
if args.alg == 'vdn':
from policy.vdn import VDN
self.policy = VDN(args)
elif args.alg == 'iql':
from policy.iql import IQL
self.policy = IQL(args)
elif args.alg == 'qmix':
from policy.qmix import QMIX
self.policy = QMIX(args)
elif args.alg == 'coma':
from policy.coma import COMA
self.policy = COMA(args)
elif args.alg == 'qtran_alt':
from policy.qtran_alt import QtranAlt
self.policy = QtranAlt(args)
elif args.alg == 'qtran_base':
from policy.qtran_base import QtranBase
self.policy = QtranBase(args)
elif args.alg == 'maven':
from policy.maven import MAVEN
self.policy = MAVEN(args)
elif args.alg == 'central_v':
from policy.central_v import CentralV
self.policy = CentralV(args)
elif args.alg == 'reinforce':
from policy.reinforce import Reinforce
self.policy = Reinforce(args)
else:
raise Exception("No such algorithm")
self.args = args
def __init__(self, args):
self.n_actions = args.n_actions
self.n_agents = args.n_agents
self.state_shape = args.state_shape
self.obs_shape = args.obs_shape
self.policy = COMA(args)
self.args = args
def __init__(self, args):
self.n_actions = args.n_actions
self.n_agents = args.n_agents
self.state_shape = args.state_shape
self.obs_shape = args.obs_shape
if args.alg == 'vdn':
self.policy = VDN(args)
elif args.alg == 'qmix':
self.policy = QMIX(args)
else:
self.policy = COMA(args)
self.args = args
def __init__(self, args):
self.n_actions = args.n_actions
self.n_agents = args.n_agents * 2
self.state_shape = args.state_shape
self.obs_shape = args.obs_shape
self.idact_shape = args.id_dim + args.n_actions
self.search_actions = np.eye(args.n_actions)
self.search_ids = np.zeros(self.n_agents)
if args.alg == 'vdn':
self.policy = VDN(args)
elif args.alg == 'qmix':
self.policy = QMIX(args)
elif args.alg == 'ours':
self.policy = OURS(args)
elif args.alg == 'coma':
self.policy = COMA(args)
elif args.alg == 'qtran_alt':
self.policy = QtranAlt(args)
elif args.alg == 'qtran_base':
self.policy = QtranBase(args)
elif args.alg == 'maven':
self.policy = MAVEN(args)
elif args.alg == 'central_v':
self.policy = CentralV(args)
elif args.alg == 'reinforce':
self.policy = Reinforce(args)
else:
raise Exception("No such algorithm")
if args.use_fixed_model:
args_goal_a = get_common_args()
args_goal_a.load_model = True
args_goal_a = get_mixer_args(args_goal_a)
args_goal_a.learn = False
args_goal_a.epsilon = 0 # 1
args_goal_a.min_epsilon = 0
args_goal_a.map = 'battle'
args_goal_a.n_actions = args.n_actions
args_goal_a.episode_limit = args.episode_limit
args_goal_a.n_agents = args.n_agents
args_goal_a.state_shape = args.state_shape
args_goal_a.feature_shape = args.feature_shape
args_goal_a.view_shape = args.view_shape
args_goal_a.obs_shape = args.obs_shape
args_goal_a.real_view_shape = args.real_view_shape
args_goal_a.load_num = args.load_num
args_goal_a.use_ja = False
args_goal_a.mlp_hidden_dim = [512, 512]
self.fixed_policy = VDN_F(args_goal_a)
self.args = args
print('Init Agents')
def __init__(self, args):
self.n_actions = args.n_actions
self.n_agents = args.n_agents
self.state_shape = args.state_shape
self.obs_shape = args.obs_shape
alg = args.alg
if alg.find('reinforce') > -1:
self.policy = Reinforce(args)
elif alg.find('coma') > -1:
self.policy = COMA(args)
elif alg.find('central_v') > -1:
self.policy = CentralV(args)
else:
raise Exception("No such algorithm")
self.args = args
print('Init CommAgents')
def __init__(self, args):
self.n_actions = args.n_actions
self.n_agents = args.n_agents
self.state_shape = args.state_shape
self.obs_shape = args.obs_shape
if args.alg == 'vdn':
self.policy = VDN(args)
elif args.alg == 'qmix':
self.policy = QMIX(args)
elif args.alg == 'coma':
self.policy = COMA(args)
elif args.alg == 'qtran_alt':
self.policy = QtranAlt(args)
elif args.alg == 'qtran_base':
self.policy = QtranBase(args)
else:
raise Exception("No such algorithm")
self.args = args