def __init__(self, env, args, *other_args):
super(ACAgent, self).__init__(env, args, other_args)
self.num_env = args.num_env
optimizer_fn = get_optimizer_func(args.opt_fn)()
# policy net
self.policy = get_policy_net(env, args)
self.target_policy = get_policy_net(env, args)
self.policy_opt = optimizer_fn(params=filter(lambda p: p.requires_grad,
self.policy.parameters()),
lr=self.policy.learning_rate)
# value net
self.value = get_value_net(env, args)
self.target_value = get_value_net(env, args)
self.value_opt = optimizer_fn(params=filter(lambda p: p.requires_grad,
self.value.parameters()),
lr=self.args.lr)
if self.num_env > 1:
self.queue = mp.Queue()
self.envs = []
self.states = []
self.memorys = []
for i in range(self.num_env):
env = make_env(args=args)
state = env.reset()
self.envs.append(env)
self.states.append(state)
self.memorys.append(ReplayBuffer(1e5))
def __init__(self, env, args, *other_args):
super(ACAgent, self).__init__(env, args, other_args)
self.num_env = args.num_env
optimizer_fn = get_optimizer_func(args.opt_fn)()
# policy net
self.policy = get_policy_net(env, args)
self.target_policy = get_policy_net(env, args)
self.policy_opt = optimizer_fn(params=filter(lambda p: p.requires_grad,
self.policy.parameters()),
lr=self.policy.learning_rate)
# value net
self.value = get_value_net(env, args)
self.target_value = get_value_net(env, args)
self.value_opt = optimizer_fn(params=filter(lambda p: p.requires_grad,
self.value.parameters()),
lr=self.args.lr)
if self.num_env > 1:
self.queue = mp.Queue()
self.envs = []
self.states = []
self.memorys = []
for i in range(self.num_env):
env = make_env(args=args)
state = env.reset()
self.envs.append(env)
self.states.append(state)
self.memorys.append(ReplayBuffer(1e5))
with open(self.args.progress_bar_path, "w") as progress_file:
if self.args.eval or self.args.infer:
progress_file.write("%d,%d\n" % (0, 1))
else:
progress_file.write("%d,%d\n" % (0, self.args.max_iter_num))
def run(self):
logger.info("%s is running" % self.name)
args = common_arg_parser()
if args.debug:
import logging
logger.setLevel(logging.DEBUG)
set_global_seeds(args.seed)
logger.info("tbase.run set global_seeds: %s" % str(args.seed))
if torch.cuda.is_available():
if args.num_env > 1 and args.device != 'cpu':
set_start_method('spawn')
env = make_env(args=args)
print("\n" + "*" * 80)
logger.info("Initializing agent by parameters:")
logger.info(str(args))
agent = get_agent(env, args)
if not args.eval and not args.infer:
logger.info("Training agent")
agent.learn()
logger.info("Finished, tensorboard --logdir=%s" %
args.tensorboard_dir)
# eval models
if not args.infer:
eval_env = make_eval_env(args=args)
agent.eval(eval_env, args)
# infer actions
if args.infer:
infer_env = make_infer_env(args=args)
agent.infer(infer_env)
def __init__(self, env, args, *other_args):
# change to random policy
args.policy_net = "Random"
super(Agent, self).__init__(env, args, other_args)
self.policy = get_policy_net(env, args)
self.num_env = args.num_env
self.envs = []
self.states = []
self.memorys = []
for i in range(self.num_env):
env = make_env(args=args)
state = env.reset()
self.envs.append(env)
self.states.append(state)
self.memorys.append(ReplayBuffer(1e5))
self.queue = mp.Queue()
def main():
args = common_arg_parser()
if args.debug:
import logging
logger.setLevel(logging.DEBUG)
set_global_seeds(args.seed)
logger.info("tbase.run set global_seeds: %s" % str(args.seed))
if torch.cuda.is_available() and args.num_env > 1 and args.device != 'cpu':
set_start_method('spawn')
env = make_env(args=args)
print("\n" + "*" * 80)
logger.info("Initializing agent by parameters:")
logger.info(str(args))
agent = get_agent(env, args)
if not args.eval:
logger.info("Training agent")
agent.learn()
logger.info("Finished, check details by run tensorboard --logdir=%s" %
args.tensorboard_dir)
# eval models
eval_env = make_eval_env(args=args)
agent.eval(eval_env, args)