autoqild.mi_estimators.neural_networks_torch

Neural Nwtowkr implementations for running the PC-softmax and Mine MI estimator.

Classes

ClassNet(in_dim, out_dim, n_units, n_hidden, ...)

ClassNet is a fully connected neural network used for classification tasks.

StatNet(in_dim[, cls_enc, n_units, ...])

StatNet is a fully connected neural network used for statistical modeling in MINE (Mutual Information Neural Estimation) tasks to estimate mutual information.
class autoqild.mi_estimators.neural_networks_torch.ClassNet(in_dim, out_dim, n_units, n_hidden, device, is_pc_softmax=True)[source]

Bases: Module

ClassNet is a fully connected neural network used for classification tasks.

This class implements a simple feedforward neural network with a configurable number of hidden layers and units. It supports a custom softmax function (PC-softmax) for handling imbalanced data.

Parameters:

  • in_dim (int) – Number of input features.

  • out_dim (int) – Number of output classes.

  • n_units (int) – Number of units in each hidden layer.

  • n_hidden (int) – Number of hidden layers.

  • device (torch.device) – Device to run the network on (CPU or GPU).

  • is_pc_softmax (bool, optional, default=True) – If True, use the custom PC-softmax function in the final layer.

input

Input layer.

Type:

torch.nn.Linear

hidden_layers

Hidden layers.

Type:

list of torch.nn.Linear

output

Output layer.

Type:

torch.nn.Linear

is_pc_softmax

Whether to use the PC-softmax function.

Type:

bool

device

Device used for computation.

Type:

torch.device

forward(x_in, label_proportions)[source]

Forward pass through the network.

Parameters:

  • x_in (torch.Tensor) – Input tensor.

  • label_proportions (list or torch.Tensor) – Proportions of each class in the dataset.

Returns:

x_in – Output tensor after applying the network layers.

Return type:

torch.Tensor

score(x_in, label_proportions)[source]

Compute class probabilities for the input samples.

Parameters:

  • x_in (torch.Tensor) – Input tensor.

  • label_proportions (list or torch.Tensor) – Proportions of each class in the dataset.

Returns:

x_in – Output tensor with class probabilities.

Return type:

torch.Tensor

class autoqild.mi_estimators.neural_networks_torch.StatNet(in_dim, cls_enc=1, n_units=100, n_hidden=1, device='cpu')[source]

Bases: Module

StatNet is a fully connected neural network used for statistical modeling in MINE (Mutual Information Neural Estimation) tasks to estimate mutual information.

This class implements a simple feedforward neural network with a configurable number of hidden layers and units. It is typically used to model the joint distribution of input features and class labels for MI estimation.

Parameters:

  • in_dim (int) – Number of input features.

  • cls_enc (int, optional, default=1) – Number of classes in the one-hot encoded target variable.

  • n_units (int, optional, default=100) – Number of units in each hidden layer.

  • n_hidden (int, optional, default=1) – Number of hidden layers.

  • device (torch.device, optional, default=`cpu`) – Device to run the network on (CPU or GPU).

input

Input layer.

Type:

torch.nn.Linear

hidden_layers

Hidden layers.

Type:

list of torch.nn.Linear

output

Output layer.

Type:

torch.nn.Linear

forward(x_in)[source]

Forward pass through the network.

Parameters:

x_in (torch.Tensor) – Input tensor.

Returns:

x_in – Output tensor after applying the network layers.

Return type:

torch.Tensor