Struct FindKOptimalOutput 

pub struct FindKOptimalOutput {
    pub k_star: usize,
    pub selector: String,
    pub cv_scores: Option<BTreeMap<usize, f64>>,
    pub cv_scores_se: Option<BTreeMap<usize, f64>>,
    pub bic_scores: Option<BTreeMap<usize, f64>>,
    pub pvalues: Option<Col<f64>>,
    pub diagnostic: Option<String>,
    pub seed: u64,
    pub n_eff: f64,
    pub stable_rank: Option<f64>,
}

Result of pls1_find_k_optimal.

Fields

k_star: usize

The selected number of components. 0 when the full-data fit cannot extract a first component (y constant, or orthogonal to X up to rounding: the k_used = 0 zero model of pls1_fit); the score maps are then empty and the diagnostic, if requested, tests nothing.

selector: String

Selection criterion used ("r2_se", "r2_max", or "bic").

cv_scores: Option<BTreeMap<usize, f64>>

CV R² per K (present for R2Se and R2Max selectors).

cv_scores_se: Option<BTreeMap<usize, f64>>

SE of CV R² per K (present only for R2Se selector).

bic_scores: Option<BTreeMap<usize, f64>>

BIC scores per K (present only for Bic selector).

pvalues: Option<Col<f64>>

Per-component p-values from the same-sample sequential diagnostic up to K* (present when diagnostic.is_some()). Same-sample → diagnostic only, not honest inference.

diagnostic: Option<String>

Name of the diagnostic method that actually ran (present when diagnostic.is_some()). A split_nb request the sequence-level auto-gate flagged reads "split_exact".

seed: u64

RNG seed actually used.

n_eff: f64

Kish’s effective sample size. Equals n_samples for uniform/absent weights.

stable_rank: Option<f64>

Stable rank of the standardized X, as the sequence-level auto-gate saw it. Some whenever diagnostic == Some(SplitNb) — whether the gate fired or not, and also under force. None otherwise, including when no diagnostic was requested. NaN when the Gram matrix of the standardized X is not finite; the gate then fires.

Implementations

impl FindKOptimalOutput

pub fn k_to_fit(&self) -> PlsKitResult<usize>

The component count a fit at the selected K uses: k_star.

This is the policy of pls1_fit(k = "optimal") in every wrapper, kept here so they share it: selection and fit are separate calls, and a wrapper that composes them asks this rather than reading k_star itself.

Errors

PlsKitError::OptimalNoComponent when k_star == 0 (no first component can be extracted). An explicit k still fits the k_used = 0 zero model; this refuses only to present that model as the selected one.

Trait Implementations

impl Clone for FindKOptimalOutput

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for FindKOptimalOutput

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

Formats the value using the given formatter.