Struct ForwardPushSolver 

pub struct ForwardPushSolver {
    pub alpha: f64,
    pub epsilon: f64,
}

Forward Push solver for Personalized PageRank.

Given a graph encoded as a CsrMatrix<f64> (adjacency list in CSR format), computes the PPR vector from a single source vertex.

Parameters

• alpha – teleport probability (fraction absorbed per push). Default: 0.85.
• epsilon – push threshold. Vertices with residual[u] > epsilon * degree(u) are eligible for a push. Smaller values yield more accurate results at the cost of more work.

Complexity

O(1 / epsilon) pushes in total, independent of |V| or |E|.

Fields

alpha: f64

Teleportation probability (alpha).

epsilon: f64

Approximation tolerance (epsilon).

Implementations

impl ForwardPushSolver

pub fn new(alpha: f64, epsilon: f64) -> Self

Create a new forward-push solver.

Parameters are validated lazily at the start of each computation (see validate_params).

pub fn default_params() -> Self

Create a solver with default parameters (alpha = 0.85, epsilon = 1e-6).

pub fn ppr_from_source( &self, graph: &CsrMatrix<f64>, source: usize, ) -> Result<Vec<(usize, f64)>, SolverError>

Compute PPR from source returning sparse (vertex, score) pairs sorted by score descending.

Errors

• SolverError::InvalidInput if source >= graph.rows.
• SolverError::NumericalInstability if the mass invariant is violated after convergence.

pub fn top_k( &self, graph: &CsrMatrix<f64>, source: usize, k: usize, ) -> Result<Vec<(usize, f64)>, SolverError>

Compute PPR from source and return only the top-k entries.

Convenience wrapper around ppr_from_source.

Trait Implementations

impl Clone for ForwardPushSolver

fn clone(&self) -> ForwardPushSolver

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ForwardPushSolver

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

Formats the value using the given formatter.

impl SolverEngine for ForwardPushSolver

fn solve( &self, matrix: &CsrMatrix<f64>, rhs: &[f64], _budget: &ComputeBudget, ) -> Result<SolverResult, SolverError>

Adapt forward-push PPR to the generic solver interface.

The rhs vector is interpreted as a source indicator: the index of the first non-zero entry is taken as the source vertex. If rhs is all zeros, vertex 0 is used. The returned SolverResult.solution contains the dense PPR vector.

fn estimate_complexity( &self, _profile: &SparsityProfile, _n: usize, ) -> ComplexityEstimate

Estimate the computational cost of solving the given system without actually performing the solve.

fn algorithm(&self) -> Algorithm

Return the algorithm identifier for this engine.

impl SublinearPageRank for ForwardPushSolver

fn ppr( &self, matrix: &CsrMatrix<f64>, source: usize, alpha: f64, epsilon: f64, ) -> Result<Vec<(usize, f64)>, SolverError>

Compute a sparse approximate PPR vector from a single source node.

fn ppr_multi_seed( &self, matrix: &CsrMatrix<f64>, seeds: &[(usize, f64)], alpha: f64, epsilon: f64, ) -> Result<Vec<(usize, f64)>, SolverError>

Compute PPR from a distribution over seed nodes rather than a single source.