DuelingCnnDQNet

class torchrl.modules.DuelingCnnDQNet(out_features: int, out_features_value: int = 1, cnn_kwargs: Optional[dict] = None, mlp_kwargs: Optional[dict] = None, device: Optional[Union[device, str, int]] = None)

Dueling CNN Q-network.

Presented in https://arxiv.org/abs/1511.06581

Parameters:

• out_features (int) – number of features for the advantage network

• out_features_value (int) – number of features for the value network

• cnn_kwargs (dict, optional) –

  kwargs for the feature network. Default is

  >>> cnn_kwargs = {
  ...     'num_cells': [32, 64, 64],
  ...     'strides': [4, 2, 1],
  ...     'kernels': [8, 4, 3],
  ... }
  

• mlp_kwargs (dict, optional) –

  kwargs for the advantage and value network. Default is

  >>> mlp_kwargs = {
  ...     "depth": 1,
  ...     "activation_class": nn.ELU,
  ...     "num_cells": 512,
  ...     "bias_last_layer": True,
  ... }
  

• device (Optional[DEVICE_TYPING]) – device to create the module on.

forward(x: Tensor) → Tensor

Defines 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.