d3rlpy.algos.AWAC¶
-
class
d3rlpy.algos.
AWAC
(*, actor_learning_rate=0.0003, critic_learning_rate=0.0003, actor_optim_factory=<d3rlpy.models.optimizers.AdamFactory object>, critic_optim_factory=<d3rlpy.models.optimizers.AdamFactory object>, actor_encoder_factory='default', critic_encoder_factory='default', q_func_factory='mean', batch_size=1024, n_frames=1, n_steps=1, gamma=0.99, tau=0.005, lam=1.0, n_action_samples=1, max_weight=20.0, n_critics=2, bootstrap=False, share_encoder=False, update_actor_interval=1, use_gpu=False, scaler=None, action_scaler=None, augmentation=None, generator=None, impl=None, **kwargs)[source]¶ Advantage Weighted Actor-Critic algorithm.
AWAC is a TD3-based actor-critic algorithm that enables efficient fine-tuning where the policy is trained with offline datasets and is deployed to online training.
The policy is trained as a supervised regression.
\[J(\phi) = \mathbb{E}_{s_t, a_t \sim D} [\log \pi_\phi(a_t|s_t) \exp(\frac{1}{\lambda} A^\pi (s_t, a_t))]\]where \(A^\pi (s_t, a_t) = Q_\theta(s_t, a_t) - Q_\theta(s_t, a'_t)\) and \(a'_t \sim \pi_\phi(\cdot|s_t)\)
The key difference from AWR is that AWAC uses Q-function trained via TD learning for the better sample-efficiency.
References
- Parameters
actor_learning_rate (float) – learning rate for policy function.
critic_learning_rate (float) – learning rate for Q functions.
actor_optim_factory (d3rlpy.models.optimizers.OptimizerFactory) – optimizer factory for the actor.
critic_optim_factory (d3rlpy.models.optimizers.OptimizerFactory) – optimizer factory for the critic.
actor_encoder_factory (d3rlpy.models.encoders.EncoderFactory or str) – encoder factory for the actor.
critic_encoder_factory (d3rlpy.models.encoders.EncoderFactory or str) – encoder factory for the critic.
q_func_factory (d3rlpy.models.q_functions.QFunctionFactory or str) – Q function factory.
batch_size (int) – mini-batch size.
n_frames (int) – the number of frames to stack for image observation.
n_steps (int) – N-step TD calculation.
gamma (float) – discount factor.
tau (float) – target network synchronization coefficiency.
lam (float) – \(\lambda\) for weight calculation.
n_action_samples (int) – the number of sampled actions to calculate \(A^\pi(s_t, a_t)\).
max_weight (float) – maximum weight for cross-entropy loss.
n_critics (int) – the number of Q functions for ensemble.
bootstrap (bool) – flag to bootstrap Q functions.
share_encoder (bool) – flag to share encoder network.
update_actor_interval (int) – interval to update policy function.
use_gpu (bool, int or d3rlpy.gpu.Device) – flag to use GPU, device ID or device.
scaler (d3rlpy.preprocessing.Scaler or str) – preprocessor. The available options are [‘pixel’, ‘min_max’, ‘standard’]
action_scaler (d3rlpy.preprocessing.ActionScaler or str) – action preprocessor. The available options are
['min_max']
.augmentation (d3rlpy.augmentation.AugmentationPipeline or list(str)) – augmentation pipeline.
generator (d3rlpy.algos.base.DataGenerator) – dynamic dataset generator (e.g. model-based RL).
impl (d3rlpy.algos.torch.sac_impl.SACImpl) – algorithm implementation.
Methods
-
build_with_dataset
(dataset)¶ Instantiate implementation object with MDPDataset object.
- Parameters
dataset (d3rlpy.dataset.MDPDataset) – dataset.
- Return type
-
build_with_env
(env)¶ Instantiate implementation object with OpenAI Gym object.
- Parameters
env (gym.core.Env) – gym-like environment.
- Return type
-
create_impl
(observation_shape, action_size)[source]¶ Instantiate implementation objects with the dataset shapes.
This method will be used internally when fit method is called.
-
fit
(episodes, n_epochs=1000, save_metrics=True, experiment_name=None, with_timestamp=True, logdir='d3rlpy_logs', verbose=True, show_progress=True, tensorboard=True, eval_episodes=None, save_interval=1, scorers=None, shuffle=True)¶ Trains with the given dataset.
algo.fit(episodes)
- Parameters
episodes (List[d3rlpy.dataset.Episode]) – list of episodes to train.
n_epochs (int) – the number of epochs to train.
save_metrics (bool) – flag to record metrics in files. If False, the log directory is not created and the model parameters are not saved during training.
experiment_name (Optional[str]) – experiment name for logging. If not passed, the directory name will be {class name}_{timestamp}.
with_timestamp (bool) – flag to add timestamp string to the last of directory name.
logdir (str) – root directory name to save logs.
verbose (bool) – flag to show logged information on stdout.
show_progress (bool) – flag to show progress bar for iterations.
tensorboard (bool) – flag to save logged information in tensorboard (additional to the csv data)
eval_episodes (Optional[List[d3rlpy.dataset.Episode]]) – list of episodes to test.
save_interval (int) – interval to save parameters.
scorers (Optional[Dict[str, Callable[[Any, List[d3rlpy.dataset.Episode]], float]]]) – list of scorer functions used with eval_episodes.
shuffle (bool) – flag to shuffle transitions on each epoch.
- Return type
-
fit_batch_online
(env, buffer=None, explorer=None, n_epochs=1000, n_steps_per_epoch=1000, n_updates_per_epoch=1000, eval_interval=10, eval_env=None, eval_epsilon=0.0, save_metrics=True, save_interval=1, experiment_name=None, with_timestamp=True, logdir='d3rlpy_logs', verbose=True, show_progress=True, tensorboard=True, timelimit_aware=True)¶ Start training loop of batch online deep reinforcement learning.
- Parameters
env (d3rlpy.envs.batch.BatchEnv) – gym-like environment.
buffer (Optional[d3rlpy.online.buffers.BatchBuffer]) – replay buffer.
explorer (Optional[d3rlpy.online.explorers.Explorer]) – action explorer.
n_epochs (int) – the number of epochs to train.
n_steps_per_epoch (int) – the number of steps per epoch.
update_interval – the number of steps per update.
n_updates_per_epoch (int) – the number of updates per epoch.
eval_interval (int) – the number of epochs before evaluation.
eval_env (Optional[gym.core.Env]) – gym-like environment. If None, evaluation is skipped.
eval_epsilon (float) – \(\epsilon\)-greedy factor during evaluation.
save_metrics (bool) – flag to record metrics. If False, the log directory is not created and the model parameters are not saved.
save_interval (int) – the number of epochs before saving models.
experiment_name (Optional[str]) – experiment name for logging. If not passed, the directory name will be
{class name}_online_{timestamp}
.with_timestamp (bool) – flag to add timestamp string to the last of directory name.
logdir (str) – root directory name to save logs.
verbose (bool) – flag to show logged information on stdout.
show_progress (bool) – flag to show progress bar for iterations.
tensorboard (bool) – flag to save logged information in tensorboard (additional to the csv data)
timelimit_aware (bool) – flag to turn
terminal
flagFalse
whenTimeLimit.truncated
flag isTrue
, which is designed to incorporate withgym.wrappers.TimeLimit
.
- Return type
-
fit_online
(env, buffer=None, explorer=None, n_steps=1000000, n_steps_per_epoch=10000, update_interval=1, update_start_step=0, eval_env=None, eval_epsilon=0.0, save_metrics=True, save_interval=1, experiment_name=None, with_timestamp=True, logdir='d3rlpy_logs', verbose=True, show_progress=True, tensorboard=True, timelimit_aware=True)¶ Start training loop of online deep reinforcement learning.
- Parameters
env (gym.core.Env) – gym-like environment.
buffer (Optional[d3rlpy.online.buffers.Buffer]) – replay buffer.
explorer (Optional[d3rlpy.online.explorers.Explorer]) – action explorer.
n_steps (int) – the number of total steps to train.
n_steps_per_epoch (int) – the number of steps per epoch.
update_interval (int) – the number of steps per update.
update_start_step (int) – the steps before starting updates.
eval_env (Optional[gym.core.Env]) – gym-like environment. If None, evaluation is skipped.
eval_epsilon (float) – \(\epsilon\)-greedy factor during evaluation.
save_metrics (bool) – flag to record metrics. If False, the log directory is not created and the model parameters are not saved.
save_interval (int) – the number of epochs before saving models.
experiment_name (Optional[str]) – experiment name for logging. If not passed, the directory name will be
{class name}_online_{timestamp}
.with_timestamp (bool) – flag to add timestamp string to the last of directory name.
logdir (str) – root directory name to save logs.
verbose (bool) – flag to show logged information on stdout.
show_progress (bool) – flag to show progress bar for iterations.
tensorboard (bool) – flag to save logged information in tensorboard (additional to the csv data)
timelimit_aware (bool) – flag to turn
terminal
flagFalse
whenTimeLimit.truncated
flag isTrue
, which is designed to incorporate withgym.wrappers.TimeLimit
.
- Return type
-
classmethod
from_json
(fname, use_gpu=False)¶ Returns algorithm configured with json file.
The Json file should be the one saved during fitting.
from d3rlpy.algos import Algo # create algorithm with saved configuration algo = Algo.from_json('d3rlpy_logs/<path-to-json>/params.json') # ready to load algo.load_model('d3rlpy_logs/<path-to-model>/model_100.pt') # ready to predict algo.predict(...)
-
get_params
(deep=True)¶ Returns the all attributes.
This method returns the all attributes including ones in subclasses. Some of scikit-learn utilities will use this method.
params = algo.get_params(deep=True) # the returned values can be used to instantiate the new object. algo2 = AlgoBase(**params)
-
load_model
(fname)¶ Load neural network parameters.
algo.load_model('model.pt')
-
predict
(x)¶ Returns greedy actions.
# 100 observations with shape of (10,) x = np.random.random((100, 10)) actions = algo.predict(x) # actions.shape == (100, action size) for continuous control # actions.shape == (100,) for discrete control
- Parameters
x (Union[numpy.ndarray, List[Any]]) – observations
- Returns
greedy actions
- Return type
-
predict_value
(x, action, with_std=False)¶ Returns predicted action-values.
# 100 observations with shape of (10,) x = np.random.random((100, 10)) # for continuous control # 100 actions with shape of (2,) actions = np.random.random((100, 2)) # for discrete control # 100 actions in integer values actions = np.random.randint(2, size=100) values = algo.predict_value(x, actions) # values.shape == (100,) values, stds = algo.predict_value(x, actions, with_std=True) # stds.shape == (100,)
- Parameters
x (Union[numpy.ndarray, List[Any]]) – observations
action (Union[numpy.ndarray, List[Any]]) – actions
with_std (bool) – flag to return standard deviation of ensemble estimation. This deviation reflects uncertainty for the given observations. This uncertainty will be more accurate if you enable
bootstrap
flag and increasen_critics
value.
- Returns
predicted action-values
- Return type
Union[numpy.ndarray, Tuple[numpy.ndarray, numpy.ndarray]]
-
sample_action
(x)¶ Returns sampled actions.
The sampled actions are identical to the output of predict method if the policy is deterministic.
- Parameters
x (Union[numpy.ndarray, List[Any]]) – observations.
- Returns
sampled actions.
- Return type
-
save_model
(fname)¶ Saves neural network parameters.
algo.save_model('model.pt')
-
save_params
(logger)¶ Saves configurations as params.json.
- Parameters
logger (d3rlpy.logger.D3RLPyLogger) – logger object.
- Return type
-
save_policy
(fname, as_onnx=False)¶ Save the greedy-policy computational graph as TorchScript or ONNX.
# save as TorchScript algo.save_policy('policy.pt') # save as ONNX algo.save_policy('policy.onnx', as_onnx=True)
The artifacts saved with this method will work without d3rlpy. This method is especially useful to deploy the learned policy to production environments or embedding systems.
See also
https://pytorch.org/tutorials/beginner/Intro_to_TorchScript_tutorial.html (for Python).
https://pytorch.org/tutorials/advanced/cpp_export.html (for C++).
https://onnx.ai (for ONNX)
-
set_params
(**params)¶ Sets the given arguments to the attributes if they exist.
This method sets the given values to the attributes including ones in subclasses. If the values that don’t exist as attributes are passed, they are ignored. Some of scikit-learn utilities will use this method.
algo.set_params(batch_size=100)
- Parameters
params (Any) – arbitrary inputs to set as attributes.
- Returns
itself.
- Return type
d3rlpy.base.LearnableBase
-
update
(epoch, total_step, batch)[source]¶ Update parameters with mini-batch of data.
- Parameters
epoch (int) – the current number of epochs.
total_step (int) – the current number of total iterations.
batch (d3rlpy.dataset.TransitionMiniBatch) – mini-batch data.
- Returns
loss values.
- Return type
Attributes
-
action_scaler
¶ Preprocessing action scaler.
- Returns
preprocessing action scaler.
- Return type
Optional[ActionScaler]
-
impl
¶ Implementation object.
- Returns
implementation object.
- Return type
Optional[ImplBase]
-
n_frames
¶ Number of frames to stack.
This is only for image observation.
- Returns
number of frames to stack.
- Return type
-
observation_shape
¶ Observation shape.
- Returns
observation shape.
- Return type
Optional[Sequence[int]]
-
scaler
¶ Preprocessing scaler.
- Returns
preprocessing scaler.
- Return type
Optional[Scaler]