Struct GnnLayer 

pub struct GnnLayer {
    pub in_dim: usize,
    pub out_dim: usize,
    pub w_self: Vec<f64>,
    pub w_neigh: Vec<f64>,
    pub bias: Vec<f64>,
    pub activation: ExtActivation,
}

A single graph neural network layer implementing the sum-aggregation message passing update:

h_i^(l+1) = σ(W_self * h_i^(l) + W_neigh * Σ_{j ∈ N(i)} h_j^(l) + b)

All nodes share the same weight matrices.

Fields

in_dim: usize

Input feature dimension.

out_dim: usize

Output feature dimension.

w_self: Vec<f64>

Self-loop weight matrix W_self [out_dim × in_dim].

w_neigh: Vec<f64>

Neighbour aggregation weight matrix W_neigh [out_dim × in_dim].

bias: Vec<f64>

Bias vector [out_dim].

activation: ExtActivation

Activation function.

Implementations

impl GnnLayer

pub fn new(in_dim: usize, out_dim: usize, activation: ExtActivation) -> Self

Create a new GNN layer with zero-initialised weights.

pub fn forward( &self, node_feats: &[f64], n_nodes: usize, adj: &[Vec<usize>], ) -> Vec<f64>

Forward pass.

• node_feats – [n_nodes × in_dim] flat row-major node feature matrix.
• adj – adjacency list: adj[i] contains the neighbour indices of node i.

Returns [n_nodes × out_dim] flat row-major.

pub fn num_params(&self) -> usize

Total number of trainable parameters.

Trait Implementations

impl Clone for GnnLayer

fn clone(&self) -> GnnLayer

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for GnnLayer

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

Formats the value using the given formatter.