parser.add_argument("-e",
"--env",
default=ENV_ID,
help="Environment id, default=" + ENV_ID)
args = parser.parse_args()
device = torch.device("cuda" if args.cuda else "cpu")
save_path = os.path.join("saves", "sac-" + args.name)
os.makedirs(save_path, exist_ok=True)
env = gym.make(args.env)
test_env = gym.make(args.env)
act_net = model.ModelActor(env.observation_space.shape[0],
env.action_space.shape[0]).to(device)
crt_net = model.ModelCritic(env.observation_space.shape[0]).to(device)
twinq_net = model.ModelSACTwinQ(env.observation_space.shape[0],
env.action_space.shape[0]).to(device)
print(act_net)
print(crt_net)
print(twinq_net)
tgt_crt_net = ptan.agent.TargetNet(crt_net)
writer = SummaryWriter(comment="-sac_" + args.name)
agent = model.AgentDDPG(act_net, device=device)
exp_source = ptan.experience.ExperienceSourceFirstLast(env,
agent,
gamma=GAMMA,
steps_count=1)
buffer = ptan.experience.ExperienceReplayBuffer(exp_source,
parser.add_argument("-e",
"--env",
default=ENV_ID,
help="Environment id, default=" + ENV_ID)
args = parser.parse_args()
device = torch.device("cuda" if args.cuda else "cpu")
save_path = os.path.join("saves", "a2c-" + args.name)
os.makedirs(save_path, exist_ok=True)
envs = [gym.make(args.env) for _ in range(ENVS_COUNT)]
test_env = gym.make(args.env)
net_act = model.ModelActor(envs[0].observation_space.shape[0],
envs[0].action_space.shape[0]).to(device)
net_crt = model.ModelCritic(envs[0].observation_space.shape[0]).to(device)
print(net_act)
print(net_crt)
writer = SummaryWriter(comment="-a2c_" + args.name)
agent = model.AgentA2C(net_act, device=device)
exp_source = ptan.experience.ExperienceSourceFirstLast(
envs, agent, GAMMA, steps_count=REWARD_STEPS)
opt_act = optim.Adam(net_act.parameters(), lr=LEARNING_RATE_ACTOR)
opt_crt = optim.Adam(net_crt.parameters(), lr=LEARNING_RATE_CRITIC)
batch = []
best_reward = None
with ptan.common.utils.RewardTracker(writer) as tracker:
with ptan.common.utils.TBMeanTracker(writer,