scimba_torch.neural_nets.coordinates_based_nets.res_net

Residual Network (ResNet) architectures.

Classes

GenericResNet(in_size, out_size, **kwargs)

A general Residual Network (ResNet) architecture.

class GenericResNet(in_size, out_size, **kwargs)

Bases: ScimbaModule

A general Residual Network (ResNet) architecture.

The layer structure is defined by the layer_structure parameter, and specifies the width, depth, and skip connections. layer_structure is a list, where:

• the first element is the width of the hidden layers,

• the second element is the number of layers,

• the remaining elements are list of pairs of integers representing the skip connections.

For instance, the default value [10, 6, [1, 3], [4, 6]] means:

• 10 hidden units in each layer,

• 6 layers,

• skip connection from layer 1 to layer 3,

• skip connection from layer 4 to layer 6.

Parameters:

• in_size (int) – Dimension of the input

• out_size (int) – Dimension of the output

• **kwargs –

  Additional keyword arguments:

  • activation_type (str, default=”tanh”): The activation function type to use in hidden layers.

  • activation_output (str, default=”id”): The activation function type to use in the output layer.

  • layer_structure (list, default=[10, 6, [1, 3], [4, 6]]): A list representing the layer structure of the ResNet.

  • weights_norm_bool (bool, default=False): If True, applies weight normalization to the layers.

Example

>>> model = ResNet(
...     4, 1, activation_type='tanh',
...     layer_structure=[20, 6, [1, 3], [4, 6]]
... )

hidden_layers

A list of hidden linear layers.

output_layer

The final output linear layer.

forward(inputs, with_last_layer=True)

Apply the network to the inputs.

Parameters:

• inputs (Tensor) – Input tensor

• with_last_layer (bool) – Whether to apply the final output layer

Return type:

Tensor

Returns:

The result of the network