sakura.models.modules.FCCompressor

class sakura.models.modules.FCCompressor(input_dim: int, output_dim: int, hidden_neurons: int = 50, hidden_layers: int = 1, config=None)

Bases: Module

Fully connected compressor module class

This module is designed to compress outputs from pre-encoder to a lower dimension with configurable hidden layers and neurons.

Architecture details:

  • When config is None, default 1 layer compressor: Input → Linear → CELU → Output

  • Hidden layer neurons can be uniform (single neuron count) or varied (list)

Parameters:

  • input_dim (int) – The dimensionality of the input data

  • output_dim – The dimensionality of the output data

  • hidden_neurons (int, optional) – The number of neurons in each hidden layer, defaults to 50

  • hidden_layers (int, optional) – The number of layer(s) in the network, defaults to 1

  • config (list[dict], optional) – A list of the module layer configuration dictionaries

Methods

forward

Sequentially forward through all modules in model_list to transform input tensor.

Attributes