dgs.utils.nn.fc_linear

dgs.utils.nn.fc_linear(hidden_layers: list[int], bias: bool | list[bool] = True, act_func: list[str | None | torch.nn.Module] | tuple[str | None | torch.nn.Module, ...] | None = None) → torch.nn.Sequential[source]

Create a Network consisting of one or more fully connected linear layers with input and output sizes given by the hidden_layers.

Parameters:

  • hidden_layers – A list containing the sizes of the input, hidden- and output layers. It is possible to use the set_up_hidden_layer_sizes() function to create this list. The length of the hidden layers is denoted L.

  • bias – Whether to use a bias in every layer. Can be a single value for the whole network or a list of length L - 1 containing one value per layer. Default is True.

  • act_func – A list containing the activation function after each of the fully connected layers. There can be a single activation function after every layer. Therefore, act_func should have a length of L. Every value can either be the torch.nn.Module or the string representing the activation function. E.g. “ReLU” for ReLU Defaults to adding no activation functions.

Returns:

A sequential model containing N-1 fully-connected layers.