unilab.algos.torch.hora.sac_models

HORA SAC actor models.

Classes

HoraSACActor

Privileged HORA teacher actor for SAC.
class unilab.algos.torch.hora.sac_models.HoraSACActor[source]

Bases: Module

Privileged HORA teacher actor for SAC.

The named modules below are intentionally stable for later distillation: priv_encoder, actor_trunk, action_mean_head, and action_logstd_head.

Parameters:
action_scale: torch.Tensor
action_bias: torch.Tensor
__init__(obs_dim, priv_info_dim, action_dim, *, hidden_dim=512, priv_info_embed_dim=9, priv_mlp_hidden_dims=(256, 128, 9), log_std_max=0.0, log_std_min=-5.0, use_tanh=True, use_layer_norm=True, device='cpu', action_scale=None, action_bias=None)[source]

Initialize internal Module state, shared by both nn.Module and ScriptModule.

Parameters:
encode_privileged_info(priv_info)[source]
Parameters:

priv_info (Tensor)

Return type:

Tensor

forward(obs, priv_info)[source]

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

Parameters:
Return type:

Tuple[Tensor, Tensor, Tensor]

as_export_module()[source]

Return a wrapper with explicit actor/priv inputs for ONNX export.

Return type:

Module

get_actions_and_log_probs(obs, priv_info)[source]
Parameters:
Return type:

Tuple[Tensor, Tensor, Tensor]

explore(obs, priv_info, deterministic=False)[source]
Parameters:
Return type:

Tensor