OnlineDTActor¶
- class torchrl.modules.OnlineDTActor(state_dim: int, action_dim: int, transformer_config: Optional[Union[Dict, DTConfig]] = None, device: Optional[Union[device, str, int]] = None)[source]¶
Online Decision Transformer Actor class.
Actor class for the Online Decision Transformer to sample actions from gaussian distribution as presented inresented in “Online Decision Transformer”.
Returns the mean and standard deviation for the gaussian distribution to sample actions from.
- Parameters:
state_dim (int) – state dimension.
action_dim (int) – action dimension.
transformer_config (Dict or
DecisionTransformer.DTConfig
) – config for the GPT2 transformer. Defaults todefault_config()
.device (torch.device, optional) – device to use. Defaults to None.
Examples
>>> model = OnlineDTActor(state_dim=4, action_dim=2, ... transformer_config=OnlineDTActor.default_config()) >>> observation = torch.randn(32, 10, 4) >>> action = torch.randn(32, 10, 2) >>> return_to_go = torch.randn(32, 10, 1) >>> mu, std = model(observation, action, return_to_go) >>> mu.shape torch.Size([32, 10, 2]) >>> std.shape torch.Size([32, 10, 2])
- classmethod default_config()[source]¶
Default configuration for
OnlineDTActor
.
- forward(observation: Tensor, action: Tensor, return_to_go: Tensor) Tuple[Tensor, Tensor] [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.