espnet2.gan_codec.shared.encoder.seanet_2d.get_activation

source

espnet2.gan_codec.shared.encoder.seanet_2d.get_activation(activation: str | None = None, channels=None, **kwargs)

Get the specified activation function.

This function returns an activation function as a PyTorch module based on the provided name. It supports custom activation functions such as ‘snake’ which requires the number of channels to be specified.

espnet2.gan_codec.shared.encoder.seanet_2d.activation

The name of the activation function to retrieve.

• Type: str

espnet2.gan_codec.shared.encoder.seanet_2d.channels

The number of channels required for specific activation functions (e.g., ‘snake’).

• Type: Optional[int]

espnet2.gan_codec.shared.encoder.seanet_2d.kwargs

Additional parameters for the activation function.

• Type: Any

• Parameters:

  • activation (str) – The name of the activation function to use. Common options include ‘ReLU’, ‘ELU’, ‘LeakyReLU’, etc.
  • channels (Optional *[*int ]) – The number of input channels for activation functions that require it.
  • **kwargs – Additional keyword arguments for the activation function.

• Returns: The corresponding activation function as a PyTorch module.

• Return type: nn.Module

• Raises:AssertionError – If ‘snake’ is specified without providing the number of channels.

Examples

>>> relu = get_activation('ReLU')
>>> snake_activation = get_activation('snake', channels=64)

NOTE

The function uses getattr to dynamically retrieve the activation function from the torch.nn module. Make sure to pass valid activation names.