Linear Layers#

Classes#

class Activation(*, activation='ELU', shapes=None, activation_kwargs=None, **kwargs)[source]#

Bases: BaseSingleLayer

Activation layer constructor.

Attributes:

groupint: Layer group, if 0 it will always be used, else it will only be used if its group matches the Networks
descriptionstr: Description of the layer
layersSequential: Layers to loop through in the forward pass

Parameters:

activationstr, Optional: Which activation function to use from PyTorch, default = ‘ELU’
shapesShapes | NoneType, Optional: Shape of the outputs from each layer, only required if tracking layer outputs is necessary
activation_kwargsdict[str, Any] | NoneType, Optional: Additional keyword arguments to pass to the activation function
**kwargs: Leftover parameters to pass to base layer for checking

class Linear(net_out, shapes, *, features=None, layer=None, factor=None, batch_norm=False, flatten_target=False, dropout=0, activation='SELU', **kwargs)[source]#

Bases: BaseSingleLayer

Linear layer constructor.

Attributes:

groupint: Layer group, if 0 it will always be used, else it will only be used if its group matches the Networks
descriptionstr: Description of the layer
layersSequential: Layers to loop through in the forward pass

Parameters:

net_outlist[int]: Shape of the network’s output
shapesShapes: Shape of the outputs from each layer
featuresint | NoneType, Optional: Number of output features for the layer, if factor is provided, features will not be used
layerint | NoneType, Optional: If factor is not None, which layer for factor to be relative to, if None, network output will be used
factorfloat | NoneType, Optional: Output features is equal to the factor of the network’s output, or if layer is provided, which layer to be relative to, will be used if provided, else features will be used
batch_normbool, Optional: If batch normalisation should be used, default = False
flatten_targetbool, Optional: If the target should be flattened so that features is equal to the product of the target multiplied by factor, if factor is provided, default = False
dropoutfloat, Optional: Probability of dropout, default = 0
activationstr | NoneType, Optional: Which activation function to use from PyTorch, default = ‘SELU’
**kwargs: Leftover parameters to pass to base layer for checking

class OrderedBottleneck(shapes, *, min_size=0, **kwargs)[source]#

Bases: BaseLayer

Information-ordered bottleneck to randomly change the size of the bottleneck in an autoencoder to encode the most important information in the first values of the latent space.

See Information-Ordered Bottlenecks for Adaptive Semantic Compression by Ho et al. (2023).

Attributes:

groupint: Layer group, if 0 it will always be used, else it will only be used if its group matches the Networks
min_sizeint: Minimum gate size
descriptionstr: Description of the layer

Parameters:

shapesShapes: Shape of the outputs from each layer
min_sizeint, Optional: Minimum gate size, default = 0
**kwargs: Leftover parameters to pass to base layer for checking

extra_repr()[source]#

Displays layer parameters when printing the network

Returns:

str: Layer parameters

forward(x, *_, **__)[source]#

Forward pass of the information-ordered bottleneck layer

Parameters:

xTensor: Input tensor with shape (N,…,Z) and type float, where N is the batch size and Z is the latent space where Z > min_size

Returns:

Tensor: Input zeroed from a random index to the last value along the dimension Z with shape (N,…,Z) and type float

class Sample(idx, shapes, **kwargs)[source]#

Bases: BaseLayer

Samples random values from a Gaussian distribution for a variational autoencoder, mean and standard deviation are the first and second half of the input channels with the last channel ignored if there are an odd number.

Attributes:

groupint: Layer group, if 0 it will always be used, else it will only be used if its group matches the Networks
descriptionstr: Description of the layer
sample_layerNormal: Layer to sample values from a Gaussian distribution

Parameters:

idxint: Layer number
shapesShapes: Shape of the outputs from each layer
**kwargs: Leftover parameters to pass to base layer for checking

forward(x, net, *_, **__)[source]#

Forward pass of the sampling layer for a variational autoencoder

Parameters:

xTensor: Input tensor with shape (N,C,…) | (N,Z) and type float, where N is the batch size, and either the channels dimension, C, or latent, Z, containing the mean and standard deviation
netNetwork: Parent network that this layer is part of

Returns:

Tensor: Output tensor sampled from the input tensor split into mean and standard deviation with shape (N,C/2,…) | (N,Z/2) and type float

to(*args, **kwargs)[source]#: See torch.nn.Module.to().

class Upsample(idx, shapes, *, shape=None, scale=2, mode='nearest', **kwargs)[source]#

Bases: BaseSingleLayer

Constructs an upsampler.

Attributes:

groupint: Layer group, if 0 it will always be used, else it will only be used if its group matches the Networks
descriptionstr: Description of the layer
layersSequential: Layers to loop through in the forward pass

Parameters:

idxint: Layer number
shapesShapes: Shape of the outputs from each layer
shapelist[int] | NoneType, Optional: Shape of the output, will be used if provided, else scale will be used
scalefloat | tuple[float] | tuple[float, …], Optional: Factor to upscale all or individual dimensions, first dimension is ignored, won’t be used if shape is provided, default = 2
mode{‘nearest’, ‘linear’, ‘bilinear’, ‘bicubic’, ‘trilinear’}: What interpolation method to use for upsampling
**kwargs: Leftover parameters to pass to base layer for checking