Struct SpatialEncoding 

pub struct SpatialEncoding {
    pub spatial_bias: Array1<f64>,
    pub max_distance: usize,
}

Spatial encoding using shortest-path distances (SPD) between node pairs.

Computes the all-pairs shortest path distance matrix via BFS on unweighted graphs, then provides learnable bias terms indexed by distance.

Fields

spatial_bias: Array1<f64>

Learnable bias for each distance value: [max_distance + 1] Index 0 = self-loop, index k = distance k

max_distance: usize

Maximum distance to encode (larger distances use max_distance bias)

Implementations

impl SpatialEncoding

pub fn new(max_distance: usize) -> Self

Create a new spatial encoding module.

Arguments

• max_distance - Maximum SPD to encode distinctly

pub fn compute_spd_matrix(&self, adj: &CsrMatrix) -> Array2<usize>

Compute all-pairs shortest path distances via BFS.

Returns a matrix [n, n] where entry (i, j) is the shortest path distance from node i to node j. Unreachable pairs get distance max_distance + 1.

pub fn forward(&self, adj: &CsrMatrix) -> Array2<f64>

Get the spatial bias matrix [n, n] for attention.

Arguments

• adj - Adjacency matrix

Returns

Bias matrix where entry (i, j) is the learnable bias for SPD(i, j)

Trait Implementations

impl Clone for SpatialEncoding

fn clone(&self) -> SpatialEncoding

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for SpatialEncoding

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.