def __init__(
self,
network_type,
env,
timesteps_per_actorbatch=1000,
gamma=0.99,
actor_batch_size=4,
epochs=1000,
lr_policy=0.01,
lr_value=0.0005,
policy_copy_interval=20,
pretrained=None,
layers=(32, 32),
tensorboard_log=None,
seed=None,
render=False,
device="cpu",
run_num=None,
save_model=None,
save_interval=50,
):
self.network_type = network_type
self.env = env
self.timesteps_per_actorbatch = timesteps_per_actorbatch
self.gamma = gamma
self.actor_batch_size = actor_batch_size
self.epochs = epochs
self.lr_policy = lr_policy
self.lr_value = lr_value
self.tensorboard_log = tensorboard_log
self.seed = seed
self.render = render
self.policy_copy_interval = policy_copy_interval
self.evaluate = evaluate
self.save_interval = save_interval
self.pretrained = pretrained
self.layers = layers
self.run_num = run_num
self.save_model = save_model
self.save = save_params
self.load = load_params
# Assign device
if "cuda" in device and torch.cuda.is_available():
self.device = torch.device(device)
else:
self.device = torch.device("cpu")
# Assign seed
if seed is not None:
set_seeds(seed, self.env)
# init writer if tensorboard
self.writer = None
if self.tensorboard_log is not None: # pragma: no cover
from torch.utils.tensorboard import SummaryWriter
self.writer = SummaryWriter(log_dir=self.tensorboard_log)
self.create_model()
def __init__(
self,
network_type: str,
env: Union[gym.Env, VecEnv],
gamma: float = 0.99,
replay_size: int = 1000000,
batch_size: int = 100,
lr_p: float = 0.0001,
lr_q: float = 0.001,
polyak: float = 0.995,
epochs: int = 100,
start_steps: int = 10000,
steps_per_epoch: int = 4000,
noise: Optional[Any] = None,
noise_std: float = 0.1,
max_ep_len: int = 1000,
start_update: int = 1000,
update_interval: int = 50,
layers: Tuple = (32, 32),
seed: Optional[int] = None,
render: bool = False,
device: Union[torch.device, str] = "cpu",
):
self.network_type = network_type
self.env = env
self.gamma = gamma
self.replay_size = replay_size
self.batch_size = batch_size
self.lr_p = lr_p
self.lr_q = lr_q
self.polyak = polyak
self.epochs = epochs
self.start_steps = start_steps
self.steps_per_epoch = steps_per_epoch
self.noise = noise
self.noise_std = noise_std
self.max_ep_len = max_ep_len
self.start_update = start_update
self.update_interval = update_interval
self.layers = layers
self.seed = seed
self.render = render
# Assign device
if "cuda" in device and torch.cuda.is_available():
self.device = torch.device(device)
else:
self.device = torch.device("cpu")
# Assign seed
if seed is not None:
set_seeds(seed, self.env)
# Setup tensorboard writer
self.writer = None
self.empty_logs()
self.create_model()
def __init__(
self,
network_type: str,
env: Union[gym.Env, VecEnv],
gamma: float = 0.99,
replay_size: int = 1000000,
batch_size: int = 256,
lr: float = 3e-4,
alpha: float = 0.01,
polyak: float = 0.995,
entropy_tuning: bool = True,
epochs: int = 1000,
start_steps: int = 0,
steps_per_epoch: int = 1000,
max_ep_len: int = 1000,
start_update: int = 256,
update_interval: int = 1,
layers: Tuple = (256, 256),
seed: Optional[int] = None,
render: bool = False,
device: Union[torch.device, str] = "cpu",
):
self.network_type = network_type
self.env = env
self.gamma = gamma
self.replay_size = replay_size
self.batch_size = batch_size
self.lr = lr
self.alpha = alpha
self.polyak = polyak
self.entropy_tuning = entropy_tuning
self.epochs = epochs
self.start_steps = start_steps
self.steps_per_epoch = steps_per_epoch
self.max_ep_len = max_ep_len
self.start_update = start_update
self.update_interval = update_interval
self.layers = layers
self.seed = seed
self.render = render
# Assign device
if "cuda" in device and torch.cuda.is_available():
self.device = torch.device(device)
else:
self.device = torch.device("cpu")
# Assign seed
if seed is not None:
set_seeds(seed, self.env)
# Setup tensorboard writer
self.writer = None
self.empty_logs()
self.create_model()
def __init__(
self,
agent,
env,
logger,
buffer=None,
off_policy=False,
save_interval=0,
render=False,
max_ep_len=1000,
distributed=False,
ckpt_log_name="experiment",
steps_per_epoch=4000,
epochs=10,
device="cpu",
log_interval=10,
batch_size=50,
seed=None,
deterministic_actions=False,
transform=None,
history_length=4,
):
self.agent = agent
self.env = env
self.logger = logger
self.off_policy = off_policy
if self.off_policy and buffer is None:
if self.agent.replay_buffer is None:
raise Exception("Off Policy Training requires a Replay Buffer")
else:
self.buffer = self.agent.replay_buffer
self.save_interval = save_interval
self.render = render
self.max_ep_len = max_ep_len
self.ckpt_log_name = ckpt_log_name
self.steps_per_epoch = steps_per_epoch
self.epochs = epochs
self.device = device
self.log_interval = log_interval
self.batch_size = batch_size
self.deterministic_actions = deterministic_actions
self.transform = transform
self.history_length = history_length
if seed is not None:
set_seeds(seed, self.env)
def __init__(
self,
network_type,
env,
gamma=0.99,
replay_size=1000000,
batch_size=256,
lr=3e-4,
alpha=0.01,
polyak=0.995,
entropy_tuning=True,
epochs=1000,
start_steps=0,
steps_per_epoch=1000,
max_ep_len=1000,
start_update=256,
update_interval=1,
layers=(256, 256),
pretrained=None,
tensorboard_log=None,
seed=None,
render=False,
device="cpu",
run_num=None,
save_model=None,
save_interval=5000,
):
self.network_type = network_type
self.env = env
self.gamma = gamma
self.replay_size = replay_size
self.batch_size = batch_size
self.lr = lr
self.alpha = alpha
self.polyak = polyak
self.entropy_tuning = entropy_tuning
self.epochs = epochs
self.start_steps = start_steps
self.steps_per_epoch = steps_per_epoch
self.max_ep_len = max_ep_len
self.start_update = start_update
self.update_interval = update_interval
self.save_interval = save_interval
self.layers = layers
self.pretrained = pretrained
self.tensorboard_log = tensorboard_log
self.seed = seed
self.render = render
self.run_num = run_num
self.save_model = save_model
self.save = save_params
self.load = load_params
self.evaluate = evaluate
# Assign device
if "cuda" in device and torch.cuda.is_available():
self.device = torch.device(device)
else:
self.device = torch.device("cpu")
# Assign seed
if seed is not None:
set_seeds(seed, self.env)
# Setup tensorboard writer
self.writer = None
if self.tensorboard_log is not None: # pragma: no cover
from torch.utils.tensorboard import SummaryWriter
self.writer = SummaryWriter(log_dir=self.tensorboard_log)
self.create_model()
def __init__(
self,
network_type,
env,
double_dqn=False,
dueling_dqn=False,
noisy_dqn=False,
categorical_dqn=False,
prioritized_replay=False,
epochs=100,
max_iterations_per_epoch=100,
max_ep_len=1000,
gamma=0.99,
lr=0.001,
batch_size=32,
replay_size=100,
prioritized_replay_alpha=0.6,
max_epsilon=1.0,
min_epsilon=0.01,
epsilon_decay=1000,
num_atoms=51,
Vmin=-10,
Vmax=10,
tensorboard_log=None,
seed=None,
render=False,
device="cpu",
save_interval=5000,
pretrained=None,
run_num=None,
save_model=None,
transform=None,
):
self.env = env
self.double_dqn = double_dqn
self.dueling_dqn = dueling_dqn
self.noisy_dqn = noisy_dqn
self.categorical_dqn = categorical_dqn
self.prioritized_replay = prioritized_replay
self.max_epochs = epochs
self.max_iterations_per_epoch = max_iterations_per_epoch
self.max_ep_len = max_ep_len
self.replay_size = replay_size
self.prioritized_replay_alpha = prioritized_replay_alpha
self.lr = lr
self.gamma = gamma
self.batch_size = batch_size
self.num_atoms = num_atoms
self.Vmin = Vmin
self.Vmax = Vmax
self.tensorboard_log = tensorboard_log
self.render = render
self.loss_hist = []
self.reward_hist = []
self.max_epsilon = max_epsilon
self.min_epsilon = min_epsilon
self.epsilon_decay = epsilon_decay
self.evaluate = evaluate
self.run_num = run_num
self.save_model = save_model
self.save_interval = save_interval
self.save = save_params
self.load = load_params
self.pretrained = pretrained
self.network_type = network_type
self.history_length = None
self.transform = transform
# Assign device
if "cuda" in device and torch.cuda.is_available():
self.device = torch.device(device)
else:
self.device = torch.device("cpu")
# Assign seed
if seed is not None:
set_seeds(seed, self.env)
# Setup tensorboard writer
self.writer = None
if self.tensorboard_log is not None: # pragma: no cover
from torch.utils.tensorboard import SummaryWriter
self.writer = SummaryWriter(log_dir=self.tensorboard_log)
self.create_model()
def __init__(
self,
network_type: str,
env: Union[gym.Env, venv],
gamma: float = 0.99,
actor_batch_size: int = 64,
lr_actor: float = 0.01,
lr_critic: float = 0.1,
num_episodes: int = 100,
timesteps_per_actorbatch: int = 4000,
max_ep_len: int = 1000,
layers: Tuple = (32, 32),
noise: Any = None,
noise_std: float = 0.1,
tensorboard_log: str = None,
seed: Optional[int] = None,
render: bool = False,
device: Union[torch.device, str] = "cpu",
run_num: int = None,
save_model: str = None,
save_interval: int = 1000,
):
self.network_type = network_type
self.env = env
self.gamma = gamma
self.actor_batch_size = actor_batch_size
self.lr_actor = lr_actor
self.lr_critic = lr_critic
self.num_episodes = num_episodes
self.timesteps_per_actorbatch = timesteps_per_actorbatch
self.max_ep_len = max_ep_len
self.layers = layers
self.noise = noise
self.noise_std = noise_std
self.tensorboard_log = tensorboard_log
self.seed = seed
self.render = render
self.run_num = run_num
self.save_interval = save_interval
self.save_model = None
self.save = save_params
self.load = load_params
# Assign device
if "cuda" in device and torch.cuda.is_available():
self.device = torch.device("cuda")
else:
self.device = torch.device("cpu")
# Assign seed
if seed is not None:
set_seeds(seed, self.env)
# Setup tensorboard writer
self.writer = None
if self.tensorboard_log is not None: # pragma: no cover
from torch.utils.tensorboard import SummaryWriter
self.writer = SummaryWriter(log_dir=self.tensorboard_log)
self.create_model()
def __init__(
self,
network_type,
env,
gamma=0.99,
replay_size=1000000,
batch_size=100,
lr_p=0.001,
lr_q=0.001,
polyak=0.995,
policy_frequency=2,
epochs=100,
start_steps=10000,
steps_per_epoch=4000,
noise=None,
noise_std=0.1,
pretrained=None,
max_ep_len=1000,
start_update=1000,
update_interval=50,
layers=(256, 256),
tensorboard_log=None,
seed=None,
render=False,
device="cpu",
run_num=None,
save_model=None,
save_interval=5000,
):
self.network_type = network_type
self.env = env
self.gamma = gamma
self.replay_size = replay_size
self.batch_size = batch_size
self.lr_p = lr_p
self.lr_q = lr_q
self.polyak = polyak
self.policy_frequency = policy_frequency
self.epochs = epochs
self.start_steps = start_steps
self.steps_per_epoch = steps_per_epoch
self.noise = noise
self.noise_std = noise_std
self.max_ep_len = max_ep_len
self.start_update = start_update
self.update_interval = update_interval
self.save_interval = save_interval
self.layers = layers
self.tensorboard_log = tensorboard_log
self.pretrained = pretrained
self.seed = seed
self.render = render
self.evaluate = evaluate
self.run_num = run_num
self.save_model = save_model
self.save = save_params
self.load = load_params
# Assign device
if "cuda" in device and torch.cuda.is_available():
self.device = torch.device(device)
else:
self.device = torch.device("cpu")
# Assign seed
if seed is not None:
set_seeds(seed, self.env)
# Setup tensorboard writer
self.writer = None
if self.tensorboard_log is not None: # pragma: no cover
from torch.utils.tensorboard import SummaryWriter
self.writer = SummaryWriter(log_dir=self.tensorboard_log)
self.create_model()
self.checkpoint = self.get_hyperparams()
def __init__(
self,
network_type: str,
env: Union[gym.Env, VecEnv],
double_dqn: bool = False,
dueling_dqn: bool = False,
noisy_dqn: bool = False,
categorical_dqn: bool = False,
prioritized_replay: bool = False,
epochs: int = 100,
max_iterations_per_epoch: int = 100,
max_ep_len: int = 1000,
gamma: float = 0.99,
lr: float = 0.001,
batch_size: int = 32,
replay_size: int = 100,
prioritized_replay_alpha: float = 0.6,
max_epsilon: float = 1.0,
min_epsilon: float = 0.01,
epsilon_decay: int = 1000,
num_atoms: int = 51,
vmin: int = -10,
vmax: int = 10,
seed: Optional[int] = None,
render: bool = False,
device: Union[torch.device, str] = "cpu",
):
self.env = env
self.double_dqn = double_dqn
self.dueling_dqn = dueling_dqn
self.noisy_dqn = noisy_dqn
self.categorical_dqn = categorical_dqn
self.prioritized_replay = prioritized_replay
self.max_epochs = epochs
self.max_iterations_per_epoch = max_iterations_per_epoch
self.max_ep_len = max_ep_len
self.replay_size = replay_size
self.prioritized_replay_alpha = prioritized_replay_alpha
self.lr = lr
self.gamma = gamma
self.batch_size = batch_size
self.num_atoms = num_atoms
self.Vmin = vmin
self.Vmax = vmax
self.render = render
self.reward_hist = []
self.max_epsilon = max_epsilon
self.min_epsilon = min_epsilon
self.epsilon_decay = epsilon_decay
self.network_type = network_type
# Assign device
if "cuda" in device and torch.cuda.is_available():
self.device = torch.device(device)
else:
self.device = torch.device("cpu")
# Assign seed
if seed is not None:
set_seeds(seed, self.env)
# Setup tensorboard writer
self.writer = None
self.empty_logs()
self.create_model()