def main():
parser = argparse.ArgumentParser(
description="Run Extended Q-Learning with given config")
parser.add_argument(
"-c",
"--config",
type=str,
metavar="",
required=True,
help="Config file name - file must be available as .json in ./configs")
args = parser.parse_args()
with open(os.path.join(".", "configs", args.config), "r") as read_file:
config = json.load(read_file)
np.random.seed(101)
agent = CrossEntropyAgent(config)
if config["is_training"]:
total_reward = session.train(agent, config)
agent.save()
helper.plot_scores(total_reward)
else:
env = gym.make(config["env_name"])
agent.load()
session.evaluate(agent, env, config)
env.env.close()
env.close()
def main():
parser = argparse.ArgumentParser(description="Run Extended Q-Learning with given config")
parser.add_argument("-c",
"--config",
type=str,
metavar="",
required=True,
help="Config file name - file must be available as .json in ./configs")
args = parser.parse_args()
# load config files
with open(os.path.join(".", "configs", args.config), "r") as read_file:
config = json.load(read_file)
env = UnityEnvironment(file_name=os.path.join(*config["env_path"]))
normalizer = OnlineNormalizer(config["network"]["observation_size"])
if config["run_training"]:
elite_net = session.train(env, normalizer, config)
checkpoint_dir = os.path.join(".", *config["checkpoint_dir"], config["env_name"])
utils.save_state_dict(os.path.join(checkpoint_dir), elite_net.state_dict())
else:
trained_net = getattr(network, config["network"]["type"])(config["network"]).to(torch.device(config["device"]))
checkpoint_dir = os.path.join(".", *config["checkpoint_dir"], config["env_name"])
trained_net.load_state_dict(utils.load_latest_available_state_dict(os.path.join(checkpoint_dir, "*")))
session.evaluate(trained_net, env, normalizer, config, num_test_runs=50)
env.close()
def main():
parser = argparse.ArgumentParser(
description="Run Extended Q-Learning with given config")
parser.add_argument(
"-c",
"--config",
type=str,
metavar="",
required=True,
help="Config file name - file must be available as .json in ./configs")
args = parser.parse_args()
# load config files
with open(os.path.join(".", "configs", args.config), "r") as read_file:
config = json.load(read_file)
env = UnityEnvironment(file_name=os.path.join(*config["env_path"]))
noise = OrnsteinUhlenbeckNoise(config["n_actions"], config["mu"],
config["theta"], config["sigma"],
config["seed"])
replay_buffer = ReplayBuffer(config["buffer_size"], config["device"],
config["seed"])
agent = DDPGAgent(config, noise, replay_buffer)
if config["run_training"]:
session.train(agent, env, config)
checkpoint_dir = os.path.join(".", *config["checkpoint_dir"],
config["env_name"])
utils.save_state_dict(os.path.join(checkpoint_dir, "actor"),
agent.actor.state_dict())
utils.save_state_dict(os.path.join(checkpoint_dir, "critic"),
agent.critic.state_dict())
else:
checkpoint_dir = os.path.join(".", *config["checkpoint_dir"],
config["env_name"])
agent.actor.load_state_dict(
utils.load_latest_available_state_dict(
os.path.join(checkpoint_dir, "actor", "*")))
agent.critic.load_state_dict(
utils.load_latest_available_state_dict(
os.path.join(checkpoint_dir, "critic", "*")))
session.evaluate(agent, env, num_test_runs=1)
env.close()
def main():
parser = argparse.ArgumentParser(
description=
"Run twin delayed deep deterministic policy gradient with given config"
)
parser.add_argument(
"-c",
"--config",
type=str,
metavar="",
required=True,
help="Config file name - file must be available as .json in ./configs")
args = parser.parse_args()
# load config files
with open(os.path.join(".", "configs", args.config), "r") as read_file:
config = json.load(read_file)
env = gym.make(config["env_name"])
env.seed(config["seed"])
torch.manual_seed(config["seed"])
np.random.seed(config["seed"])
random.seed(config["seed"])
state_dim = env.observation_space.shape[0]
action_dim = env.action_space.shape[0]
max_action = float(env.action_space.high[0])
agent = TwinDelayedDDPG(config, state_dim, action_dim, max_action)
if config["run_training"]:
session.train(agent, env, config)
agent.save()
else:
if config["create_video"]:
env = utils.set_up_monitoring(env, config)
agent.load()
session.evaluate(agent, env, config["eval_episodes"])
env.close()
def main():
parser = argparse.ArgumentParser(
description="Run Extended Q-Learning with given config")
parser.add_argument(
"-c",
"--config",
type=str,
metavar="",
required=True,
help="Config file name - file must be available as .json in ./configs")
args = parser.parse_args()
with open(os.path.join(".", "configs", args.config), "r") as read_file:
config = json.load(read_file)
batch_handler = BatchHandler(config)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
generator_noise_input = torch.FloatTensor(16, 100, 1,
1).normal_(0, 1).to(device)
atari_gan = AtariGan(config)
session.train(atari_gan, generator_noise_input, batch_handler, config)