Struct KhatriRaoConeConstraints 

pub struct KhatriRaoConeConstraints { /* private fields */ }

Nonnegativity cone (e_k ⊗ ψ_i)ᵀ β ≥ 0 for a row-major Khatri-Rao block.

The coefficient block is β = vec(A) with A reshaped row-major as p_left × p_cov (coefficient A[k, j] = β[k · p_cov + j]). The cone constrains the factored linear functionals α_k(x_i) = ψ_iᵀ A_{k,:} to be non-negative for every observation row i of factor and every k ∈ coupled_rows.

Row identifiers are stable and dense: row r = s · n + i where s indexes into coupled_rows and i is the observation row. Active-set warm starts therefore survive across iterations exactly as with the dense system.

Implementations

impl KhatriRaoConeConstraints

pub fn new( factor: Arc<Array2<f64>>, coupled_rows: Vec<usize>, p_left: usize, ) -> Result<Self, String>

pub fn factor(&self) -> &Array2<f64>

pub fn coupled_rows(&self) -> &[usize]

pub fn p_left(&self) -> usize

pub fn single_coupled_slot(&self, slot: usize) -> Result<Self, String>

One coupled response-row slot as a standalone cone over a single p_cov coefficient block. The covariate factor remains shared by Arc; only the small row-norm vector and this slot’s optional bounds are copied. This is the exact block decomposition of an identity-Hessian projection, not a reduced-data approximation.

pub fn nrows(&self) -> usize

pub fn ncols(&self) -> usize

pub fn values(&self, beta: ArrayView1<'_, f64>) -> Result<Array1<f64>, String>

Raw (un-normalized) constraint values A β for the full row set, laid out slot-major (r = s·n + i).

Cost: one n × p_cov · p_cov product per coupled row — never the nrows × ncols dense system.

pub fn row_norm(&self, row: usize) -> Result<f64, String>

Unit-normalization denominator of one row (‖ψ_i‖, shared across coupled slots). Zero rows are vacuous (0ᵀβ ≥ 0 always holds) exactly like the canonicalized dense system keeps them inert.

pub fn row_column_support(&self, row: usize) -> Result<Vec<usize>, String>

The coefficient columns row row acts on, ascending.

Row (slot, i) has normal e_k ⊗ ψ_i with k = coupled_rows[slot], and Self::values reads exactly the block β[k·p_cov .. (k+1)·p_cov], so the support is k·p_cov + j over the columns j where ψ_{i,j} ≠ 0. Every other coefficient has a structurally zero coefficient in this row.

pub fn bound(&self, row: usize) -> Result<f64, String>

Per-row right-hand side (0 for the homogeneous cone, shifted values after ConstraintSet::shifted_to_delta).

pub fn gather_rows( &self, rows: &[usize], ) -> Result<LinearInequalityConstraints, String>

Materialize the requested rows as a dense system (active-set KKT use; the id order of rows is preserved). Rows come out RAW (un-normalized), matching the raw dense construction path; callers that need geometric tolerances canonicalize the gathered system.

pub fn to_dense(&self) -> Result<LinearInequalityConstraints, String>

Exact dense equivalent of the ENTIRE cone. Test/oracle use only — this is the materialization the carrier exists to avoid.

Trait Implementations

impl Clone for KhatriRaoConeConstraints

fn clone(&self) -> KhatriRaoConeConstraints

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for KhatriRaoConeConstraints

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

Formats the value using the given formatter.