Struct JointPenaltyBundle 

pub struct JointPenaltyBundle {
    pub specs: Arc<Vec<JointPenaltySpec>>,
    pub log_lambdas: Vec<f64>,
}

Per-evaluation bundle of cross-block penalties paired with their current log-smoothing parameters.

The outer optimizer concatenates joint penalty log λ values onto the per-block ρ vector; the inner solver receives this bundle via [crate::families::custom_family::BlockwiseFitOptions::joint_penalties] and adds the full-width quadratic / matvec / preconditioner / Hessian contributions to the joint-Newton primitives.

Fields

specs: Arc<Vec<JointPenaltySpec>>

log_lambdas: Vec<f64>

Implementations

impl JointPenaltyBundle

pub fn new( specs: Arc<Vec<JointPenaltySpec>>, log_lambdas: Vec<f64>, total_compiled: usize, ) -> Result<Self, String>

Build a bundle, validating the per-penalty log λ count and dimension agreement against total_compiled.

pub fn len(&self) -> usize

pub fn is_empty(&self) -> bool

pub fn quadratic(&self, beta: ArrayView1<'_, f64>) -> f64

Total joint-penalty contribution to the objective: ½ Σ_j exp(ρ_j) · βᵀ S_j β.

pub fn add_apply_into(&self, vector: ArrayView1<'_, f64>, out: &mut Array1<f64>)

Accumulate Σ_j exp(ρ_j) · S_j · v into out (additive).

pub fn add_diag(&self, diag: &mut Array1<f64>)

Accumulate Σ_j exp(ρ_j) · diag(S_j) into diag (additive).

pub fn add_to_matrix(&self, matrix: &mut Array2<f64>)

Accumulate Σ_j exp(ρ_j) · S_j into the full matrix (additive).

pub fn rho_objective_gradient(&self, beta: ArrayView1<'_, f64>, out: &mut [f64])

Per-penalty ρ-gradient contribution to the outer objective term: ∂/∂ρ_j [½ exp(ρ_j) βᵀ S_j β] = exp(ρ_j) · ½ βᵀ S_j β.

Trait Implementations

impl Clone for JointPenaltyBundle

fn clone(&self) -> JointPenaltyBundle

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for JointPenaltyBundle

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

Formats the value using the given formatter.