Language Language English 简体中文

unilab.algos.torch.hora.sac_learner

HORA-owned FastSAC learner.

Functions

derive_priv_info_from_critic_obs(actor_obs, ...)

Return the privileged tail encoded in the critic observation contract.

Classes

HoraSACLearner

FastSAC learner variant whose actor consumes HORA privileged info.

unilab.algos.torch.hora.sac_learner.derive_priv_info_from_critic_obs(actor_obs, critic_obs, *, context)

Return the privileged tail encoded in the critic observation contract.

Parameters:

• actor_obs (Tensor)

• critic_obs (Tensor)

• context (str)

Return type:

Tensor

class unilab.algos.torch.hora.sac_learner.HoraSACLearner

Bases: FastSACLearner

FastSAC learner variant whose actor consumes HORA privileged info.

Parameters:

• obs_dim (int)

• critic_obs_dim (int)

• priv_info_dim (int)

• action_dim (int)

• device (str)

• actor_hidden_dim (int)

• priv_info_embed_dim (int)

• priv_mlp_hidden_dims (Sequence[int])

• log_std_max (float)

• log_std_min (float)

• use_tanh (bool)

• use_layer_norm (bool)

• actor_lr (float)

• weight_decay (float)

• use_symmetry (bool)

• symmetry_augmentation (Any | None)

• kwargs (Any)

__init__(*, obs_dim, critic_obs_dim, priv_info_dim, action_dim, device='cpu', actor_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, actor_lr=0.0003, weight_decay=0.001, use_symmetry=False, symmetry_augmentation=None, **kwargs)

Parameters:

• obs_dim (int)

• critic_obs_dim (int)

• priv_info_dim (int)

• action_dim (int)

• device (str)

• actor_hidden_dim (int)

• priv_info_embed_dim (int)

• priv_mlp_hidden_dims (Sequence[int])

• log_std_max (float)

• log_std_min (float)

• use_tanh (bool)

• use_layer_norm (bool)

• actor_lr (float)

• weight_decay (float)

• use_symmetry (bool)

• symmetry_augmentation (Any | None)

• kwargs (Any)