Struct BlockwiseInnerResult 

pub struct BlockwiseInnerResult {
    pub block_states: Vec<ParameterBlockState>,
    pub active_sets: Vec<Option<Vec<usize>>>,
    pub log_likelihood: f64,
    pub penalty_value: f64,
    pub cycles: usize,
    pub converged: bool,
    pub block_logdet_h: f64,
    pub block_logdet_s: f64,
    pub s_lambdas: Vec<ArrayBase<OwnedRepr<f64>, Dim<[usize; 2]>>>,
    pub joint_workspace: Option<Arc<dyn ExactNewtonJointHessianWorkspace>>,
    pub kkt_residual: Option<ProjectedKktResidual>,
    pub active_constraints: Option<Arc<ActiveLinearConstraintBlock>>,
}

Fields

block_states: Vec<ParameterBlockState>

active_sets: Vec<Option<Vec<usize>>>

log_likelihood: f64

penalty_value: f64

cycles: usize

converged: bool

block_logdet_h: f64

block_logdet_s: f64

s_lambdas: Vec<ArrayBase<OwnedRepr<f64>, Dim<[usize; 2]>>>

Cached assembled penalty matrices S(ρ) = Σ_k exp(ρ_k) S_k per block. Avoids redundant re-assembly in the outer objective evaluation.

joint_workspace: Option<Arc<dyn ExactNewtonJointHessianWorkspace>>

kkt_residual: Option<ProjectedKktResidual>

Projected KKT residual at the converged inner iterate, propagated to the unified evaluator’s InnerAssembly::kkt_residual for the outer REML/LAML scoring path. None when the solver path doesn’t produce a typed KKT diagnostic (blockwise NR fallback, eager-stop).

active_constraints: Option<Arc<ActiveLinearConstraintBlock>>

Active linear-inequality constraint rows at the converged inner iterate. When Some, the unified evaluator builds the constraint-aware kernel K_T = K_S − K_S Aᵀ (A K_S Aᵀ)⁻¹ A K_S for per-coordinate mode responses v_k = ∂β/∂ρ_k.

Trait Implementations

impl Clone for BlockwiseInnerResult

fn clone(&self) -> BlockwiseInnerResult

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for BlockwiseInnerResult

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

Formats the value using the given formatter.