Shared Layers

Building-block layers used across multiple models.

class chronocratic.models.layers.general.BandedFourierLayer(in_channels: int, out_channels: int, band: int, num_bands: int, length: int = 201)

Bases: Module

Banded Fourier Layer for applying banded Fourier transform to the input tensor.

Parameters:

• in_channels (int) – Number of input channels.

• out_channels (int) – Number of output channels.

• band (int) – Index of the band to process.

• num_bands (int) – Total number of bands.

• length (int) – Length of the input sequence. Default is 201.

forward(input_tensor: Tensor) → Tensor

Forward pass for the Banded Fourier Layer.

Parameters:

input_tensor – Input tensor of shape (batch_size, time_steps, in_channels).

Returns:

Output tensor after applying the banded Fourier transform.

reset_parameters() → None

Initialize the layer parameters.

class chronocratic.models.layers.general.LevelModel(latent_dim: int, input_dims: int, sequence_length: int)

Bases: Module

forward(z: Tensor) → Tensor

Return the level component for each latent vector.

class chronocratic.models.layers.general.ResidualConnection(sequence_length: int, input_dims: int, hidden_layer_sizes: Sequence[int], latent_dim: int, encoder_last_dense_dim: int)

Bases: Module

forward(z: Tensor) → Tensor

Return the residual decoder branch for each latent vector.

class chronocratic.models.layers.general.SeasonalLayer(sequence_length: int, input_dims: int, latent_dim: int, custom_seasonality: tuple[tuple[int, int], ...])

Bases: Module

forward(z: Tensor) → Tensor

Return additive seasonal values for each latent vector.

compute_output_shape(input_shape: tuple[int, ...]) → tuple[int, int, int]

Return the output shape for Keras-compatible callers.

class chronocratic.models.layers.general.TrendLayer(sequence_length: int, input_dims: int, latent_dim: int, trend_poly: int)

Bases: Module

forward(z: Tensor) → Tensor

Return polynomial trend values for each latent vector.