sakura.models.modules.FCDecoder

class sakura.models.modules.FCDecoder(input_dim, output_dim, hidden_neurons=50, hidden_layers=3, output_activation_function='relu', dropout=False, dropout_input=False, dropout_input_p=0.5, dropout_hidden=False, dropout_hidden_p=0.5, config=None)

Bases: Module

Fully connected decoder module class

Module supports configurable hidden layers and neurons, output activation functions, and dropout regularization.

Architecture details:

  • When config is None, default 3 hidden layers with structure: Input → Linear → CELU → Linear → CELU → Linear → Output

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

  • Optional dropout placement after the input layer

  • Optional but uniform dropout placement after hidden layers (if #hidden_layer > 1)

  • Final layer always has output_dim neurons with optional activation

Parameters:

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

  • output_dim (int) – The dimensionality of the output data

  • hidden_neurons (int or list[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 3

  • output_activation_function (Literal['relu', 'softmax','identity'], optional) – The activation function for the output layer, defaults to ‘relu’

  • dropout (bool, optional) – Whether to apply dropout regularization, defaults to False

  • dropout_input (bool, optional) – Whether to apply dropout to the input layer, defaults to False

  • dropout_input_p (float, optional) – The probability of dropout for the input layer, defaults to 0.5

  • dropout_hidden (bool, optional) – Whether to apply dropout to the hidden layer, defaults to False

  • dropout_hidden_p (float, optional) – The probability of dropout for the hidden layer, defaults to 0.5

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

Methods

forward

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

Attributes