Struct EdgeEncoding 

pub struct EdgeEncoding {
    pub edge_embed: Array2<f64>,
    pub projection: Array1<f64>,
    pub max_edge_types: usize,
    pub hidden_dim: usize,
}

Edge encoding along shortest paths between node pairs.

For each pair (i, j), takes the edges along the shortest path from i to j and averages the learnable edge embeddings. This provides additional structural information to the attention mechanism.

Fields

edge_embed: Array2<f64>

Learnable edge embedding: [max_edge_types, hidden_dim]

projection: Array1<f64>

Projection from hidden_dim to scalar bias

max_edge_types: usize

Maximum number of edge types

hidden_dim: usize

Hidden dimension for edge embedding

Implementations

impl EdgeEncoding

pub fn new(max_edge_types: usize, hidden_dim: usize) -> Self

Create a new edge encoding module.

Arguments

• max_edge_types - Maximum number of distinct edge feature types
• hidden_dim - Dimension for edge embeddings

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

Compute edge bias matrix.

For simplicity, uses edge weights discretized to integer types. Each edge on the shortest path contributes its embedding, which is averaged and projected to a scalar.

Arguments

• adj - Adjacency matrix with edge weights
• spd - Shortest path distance matrix from SpatialEncoding::compute_spd_matrix

Returns

Edge bias matrix [n, n]

Trait Implementations

impl Clone for EdgeEncoding

fn clone(&self) -> EdgeEncoding

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for EdgeEncoding

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

Formats the value using the given formatter.