Struct ExactFullConformalSubstrate 

pub struct ExactFullConformalSubstrate { /* private fields */ }

Persistable substrate for the EXACT Gaussian-identity full-conformal set (#942 Layer 1 + the Layer-3 frozen-ρ self-diagnostic), the analogue of GaussianJackknifePlusStats for the exact set.

Unlike jackknife+, the exact full-conformal set has no test-point-independent factorization: every test covariate x_* enters the augmented normal matrix M = XᵀX + x_*x_*ᵀ + Sλ, so the substrate persists the training design X, response y, and the (frozen) penalty Sλ and rebuilds ExactGaussianFullConformal per test row — one Cholesky per test point, zero refits. Valid for any penalized smooth with an arbitrary Sλ and basis.

Sλ is recovered once at fit time from the converged penalized Hessian M₀ = XᵀX + Sλ (the Gaussian-identity, unit-weight, dispersion-unscaled normal matrix stored in [FitGeometry]) as Sλ = M₀ − XᵀX, so no penalty re-derivation is needed — exactly the seam the jackknife+ substrate uses.

The frozen-ρ self-diagnostic treats the entire frozen penalty as carrying a single global log-smoothing parameter ρ with S(ρ) = eᵖ·Sλ and runs the closed-form GaussianRemlRhoResponse::certified_full_conformal: it re-selects the global ρ̂(z) on the augmented data, bounds the score perturbation freezing ρ̂ could induce, and reports whether freezing is accepted under the stated rho-grid Lipschitz assumption. This is a sound, conservative global-scale check that applies to any penalized smooth (it does not require the model to be single-penalty); per-penalty re-selection is the research-core Layer 3 and is not asserted here.

Unit prior weights are required, as everywhere in this module: a reweighted training row is not exchangeable with the test row.

Implementations

impl ExactFullConformalSubstrate

pub fn from_design_unit_weight_normal_matrix( x: &Array2<f64>, y: &Array1<f64>, prior_weights: &Array1<f64>, m: &Array2<f64>, ) -> Result<Self, String>

Build the substrate from the training design, response, prior weights, and the converged penalized normal matrix M₀ = XᵀX + Sλ. Recovers the frozen penalty Sλ = M₀ − XᵀX once. Rejects non-unit prior weights and shape mismatches, identically to the rest of this module.

pub fn p(&self) -> usize

Coefficient dimension p.

pub fn n(&self) -> usize

Training-row count n.

pub fn interval( &self, x_star: &Array1<f64>, alpha: f64, ) -> Result<ExactFullConformalInterval, String>

The exact full-conformal verdict at one test row x_* and miscoverage alpha: the exact set, its outer envelope, and the frozen-ρ self-diagnostics flag. One Cholesky per call, zero refits.

Trait Implementations

impl Clone for ExactFullConformalSubstrate

fn clone(&self) -> ExactFullConformalSubstrate

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ExactFullConformalSubstrate

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for ExactFullConformalSubstrate

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for ExactFullConformalSubstrate

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.