unilab.algos.torch.offpolicy.runtime
Runtime resolution helpers for off-policy script assembly.
Functions
Classes
OffPolicyRuntime
|
Optional runtime overrides for the generic off-policy SAC path. |
-
class unilab.algos.torch.offpolicy.runtime.OffPolicyRuntime[source]
Bases: object
Optional runtime overrides for the generic off-policy SAC path.
All fields are optional so custom runtimes only declare the behaviour they
need to change from standard SAC.
- Parameters:
-
-
learner_cls:
type[Any] | None = None
-
algo_type:
str | None = None
-
actor_kwargs:
dict[str, Any]
-
supports_symmetry:
bool = True
-
build_model_kwargs(*, obs_dim, critic_obs_dim)[source]
Build kwargs shared by learner construction and collector actor construction.
- Parameters:
obs_dim (int)
critic_obs_dim (int)
- Return type:
dict[str, Any]
-
__init__(learner_cls=None, algo_type=None, actor_kwargs=<factory>, supports_symmetry=True)
- Parameters:
-
-
unilab.algos.torch.offpolicy.runtime.resolve_custom_offpolicy_runtime(rl_cfg)[source]
Resolve an optional custom off-policy runtime from owner config.
- Parameters:
rl_cfg (dict[str, Any])
- Return type:
OffPolicyRuntime | None