Struct GraphLRP 

pub struct GraphLRP {
    pub alpha: f32,
    pub beta: f32,
    pub epsilon: f32,
    pub activations: HashMap<String, Tensor>,
    pub relevances: HashMap<String, Tensor>,
}

Layer-wise Relevance Propagation for Graph Neural Networks

Implements LRP-based explainability methods adapted for graph structures, providing node-level and edge-level importance scores.

Fields

alpha: f32

Alpha parameter for LRP-alpha-beta rule

beta: f32

Beta parameter for LRP-alpha-beta rule

epsilon: f32

Epsilon parameter for numerical stability

activations: HashMap<String, Tensor>

Stored activations for each layer

relevances: HashMap<String, Tensor>

Stored relevance scores for each layer

Implementations

impl GraphLRP

pub fn new() -> Self

Create a new GraphLRP instance with default parameters

pub fn with_alpha_beta(alpha: f32, beta: f32) -> Self

Create with custom alpha-beta parameters

pub fn store_activation(&mut self, layer_name: String, activation: Tensor)

Store activations from a forward pass

pub fn compute_relevance_epsilon( &self, input: &Tensor, _output: &Tensor, weight: &Tensor, output_relevance: &Tensor, ) -> Result<Tensor, Box<dyn Error>>

Compute relevance scores using LRP-epsilon rule

pub fn compute_relevance_alpha_beta( &self, input: &Tensor, weight: &Tensor, output_relevance: &Tensor, ) -> Result<Tensor, Box<dyn Error>>

Compute relevance scores using LRP-alpha-beta rule

pub fn compute_graph_relevance( &self, graph: &GraphData, node_relevance: &Tensor, layer_name: &str, ) -> Result<GraphRelevanceResult, Box<dyn Error>>

Compute graph-aware relevance propagation considering edge structure

pub fn analyze_relevance_patterns( &self, _graph: &GraphData, relevance_result: &GraphRelevanceResult, ) -> RelevanceAnalysis

Analyze relevance patterns across the entire graph

Trait Implementations

impl Clone for GraphLRP

fn clone(&self) -> GraphLRP

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for GraphLRP

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

Formats the value using the given formatter.

impl Default for GraphLRP

fn default() -> Self

Returns the “default value” for a type.