Layers

class rff.layers.BasicEncoding

Layer for mapping coordinates using the basic encoding

Initializes internal Module state, shared by both nn.Module and ScriptModule.

forward(v: Tensor) → Tensor

Computes \(\gamma(\mathbf{v}) = (\cos{2 \pi \mathbf{v}} , \sin{2 \pi \mathbf{v}})\)

Parameters: v (Tensor) – input tensor of shape \((N, *, \text{input_size})\)
Returns: mapped tensor of shape \((N, *, 2 \cdot \text{input_size})\)
Return type: Tensor

class rff.layers.GaussianEncoding(sigma: Optional[float] = None, input_size: Optional[float] = None, encoded_size: Optional[float] = None, b: Optional[Tensor] = None)

Layer for mapping coordinates using random Fourier features

Parameters

sigma (Optional[float]) – standard deviation
input_size (Optional[float]) – the number of input dimensions
encoded_size (Optional[float]) – the number of dimensions the b matrix maps to
b (Optional[Tensor], optional) – Optionally specify a b matrix already sampled

Raises

ValueError – If b is provided and one of sigma, input_size, or encoded_size is provided. If b is not provided and one of sigma, input_size, or encoded_size is not provided.

forward(v: Tensor) → Tensor

Computes \(\gamma(\mathbf{v}) = (\cos{2 \pi \mathbf{B} \mathbf{v}} , \sin{2 \pi \mathbf{B} \mathbf{v}})\)

Parameters: v (Tensor) – input tensor of shape \((N, *, \text{input_size})\)
Returns: Tensor mapping using random fourier features of shape \((N, *, 2 \cdot \text{encoded_size})\)
Return type: Tensor

class rff.layers.PositionalEncoding(sigma: float, m: int)

Layer for mapping coordinates using the positional encoding

Parameters

sigma (float) – frequency constant
m (int) – number of frequencies to map to

forward(v: Tensor) → Tensor

Computes \(\gamma(\mathbf{v}) = (\dots, \cos{2 \pi \sigma^{(j/m)} \mathbf{v}} , \sin{2 \pi \sigma^{(j/m)} \mathbf{v}}, \dots)\)

Parameters: v (Tensor) – input tensor of shape \((N, *, \text{input_size})\)
Returns: mapped tensor of shape \((N, *, 2 \cdot m \cdot \text{input_size})\)
Return type: Tensor