common.encoder

Please Reference ding/ding/docs/source/api_doc/model/common/encoder.py for usage

ConvEncoder

class ding.model.common.encoder.ConvEncoder(obs_shape: ding.utils.type_helper.SequenceType, hidden_size_list: ding.utils.type_helper.SequenceType = [32, 64, 64, 128], activation: Optional[torch.nn.modules.module.Module] = ReLU(), norm_type: Optional[str] = None)

Overview:

The Convolution Encoder used in models. Used to encoder raw 2-dim observation.

Interfaces:

__init__, forward

__init__(obs_shape: ding.utils.type_helper.SequenceType, hidden_size_list: ding.utils.type_helper.SequenceType = [32, 64, 64, 128], activation: Optional[torch.nn.modules.module.Module] = ReLU(), norm_type: Optional[str] = None) → None

Overview:

Init the Convolution Encoder according to arguments.

Arguments:

• obs_shape (SequenceType): Sequence of in_channel, some output size

• hidden_size_list (SequenceType): The collection of hidden_size

• activation (nn.Module):

  The type of activation to use in the conv layers and ResBlock, if None then default set to nn.ReLU()

• norm_type (str):

  The type of normalization to use, see ding.torch_utils.ResBlock for more details

_get_flatten_size() → int

Overview:

Get the encoding size after self.main to get the number of in-features to feed to nn.Linear.

Arguments:

• x (torch.Tensor): Encoded Tensor after self.main

Returns:

• outputs (torch.Tensor): Size int, also number of in-feature

forward(x: torch.Tensor) → torch.Tensor

Overview:

Return embedding tensor of the env observation

Arguments:

• x (torch.Tensor): Env raw observation

Returns:

• outputs (torch.Tensor): Embedding tensor

FCEncoder

class ding.model.common.encoder.FCEncoder(obs_shape: int, hidden_size_list: ding.utils.type_helper.SequenceType, res_block: bool = False, activation: Optional[torch.nn.modules.module.Module] = ReLU(), norm_type: Optional[str] = None)

Overview:

The FCEncoder used in models. Used to encoder raw 1-dim observation.

Interfaces:

__init__, forward

__init__(obs_shape: int, hidden_size_list: ding.utils.type_helper.SequenceType, res_block: bool = False, activation: Optional[torch.nn.modules.module.Module] = ReLU(), norm_type: Optional[str] = None) → None

Overview:

Init the FC Encoder according to arguments.

Arguments:

• obs_shape (int): Observation shape

• hidden_size_list (SequenceType): The collection of hidden_size

• res_block (bool): Whether use res_block.

• activation (nn.Module):

  The type of activation to use in the ResFCBlock, if None then default set to nn.ReLU()

• norm_type (str):

  The type of normalization to use, see ding.torch_utils.ResFCBlock for more details

forward(x: torch.Tensor) → torch.Tensor

Overview:

Return embedding tensor of the env observation

Arguments:

• x (torch.Tensor): Env raw observation

Returns:

• outputs (torch.Tensor): Embedding tensor

StructEncoder

class ding.model.common.encoder.StructEncoder

__init__() → None

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