Periodic Tab DL

Periodic Tab DL provides Fourier-based feature encoding and periodic neural networks for tabular data processing.

Functions

class FourierEncoder(nn.Module)

Fourier Feature Encoder Layer with optional Random Fourier Features (RFF).

Parameters:

• input_size (int) - Number of input features.

• num_features_per_input (int) - Number of Fourier features per input feature.

• kernel_size (int, optional, Default is 1) - Size of the kernel for local feature extraction.

• scale_input (bool, optional, Default is True) - Whether to scale the input with learnable parameters.

• init_frequency_range (tuple, optional, Default is (0.0, 5.0)) - Range for frequency initialization.

• use_feature_scaling (bool, optional, Default is True) - Whether to add learnable scaling for Fourier features.

• use_convolution (bool, optional, Default is True) - Whether to apply convolution over input features.

• activation (str, optional, Default is ‘sin_cos’) - Activation function (‘sin_cos’, ‘sin’, ‘cos’, ‘tanh’).

• frequency_init (str, optional, Default is ‘log’) - Method for initializing frequencies.

• frequency_scale (float, optional, Default is None) - Scaling factor for frequencies.

• use_phase_shift (bool, optional, Default is True) - Whether to use learnable phase shifts.

• use_learnable_amplitude (bool, optional, Default is True) - Whether to use learnable amplitude scaling.

• use_rff (bool, optional, Default is True) - Whether to use Random Fourier Features.

• rff_sigma (float, optional, Default is 1.0) - Bandwidth parameter for RFF.

Input:

• x (Tensor) - Input tensor of shape (batch_size, input_size).

Output:

• Tensor - Fourier encoded features.

class FourierBlock(nn.Module)

Fourier block for periodic neural networks.

Parameters:

• input_dim (int) - Input dimension.

• hidden_dim (int) - Hidden dimension.

• output_dim (int) - Output dimension.

• activation (str, optional, Default is ‘relu’) - Activation function.

Input:

• x (Tensor) - Input tensor.

Output:

• Tensor - Block output.

class FourierNet(nn.Module)

Fourier-based neural network for tabular data.

Parameters:

• input_dim (int) - Input dimension.

• hidden_dims (List[int]) - List of hidden dimensions.

• output_dim (int) - Output dimension.

• activation (str, optional, Default is ‘relu’) - Activation function.

Input:

• x (Tensor) - Input tensor.

Output:

• Tensor - Network output.

class TabFourierNet(nn.Module)

Tabular Fourier network with feature encoding.

Parameters:

• input_dim (int) - Input dimension.

• hidden_dims (List[int]) - List of hidden dimensions.

• output_dim (int) - Output dimension.

• fourier_params (Dict) - Fourier encoder parameters.

Input:

• x (Tensor) - Input tensor.

Output:

• Tensor - Network output.

class OrthogonalPolynomial

Orthogonal polynomial basis functions.

Parameters:

• degree (int) - Polynomial degree.

• basis_type (str) - Type of orthogonal basis.

Methods:

• evaluate(self, x) - Evaluate polynomial at given points.

class PnPBlock(nn.Module)

Plug-and-Play neural network block.

Parameters:

• input_dim (int) - Input dimension.

• output_dim (int) - Output dimension.

• activation (str) - Activation function.

Input:

• x (Tensor) - Input tensor.

Output:

• Tensor - Block output.

class AutoPnP(nn.Module)

Automatic Plug-and-Play network architecture.

Parameters:

• input_dim (int) - Input dimension.

• hidden_dims (List[int]) - List of hidden dimensions.

• output_dim (int) - Output dimension.

Input:

• x (Tensor) - Input tensor.

Output:

• Tensor - Network output.