Struct EncodeObjective 

pub struct EncodeObjective<'a> {
    pub metric_factor: Option<ArrayView2<'a, f64>>,
    pub prior_alpha: Option<&'a [f64]>,
    pub metric_norm_bound: f64,
}

The TRUE per-row encode objective’s non-Euclidean ingredients (F3), so the Kantorovich certificate certifies the SAME functional the fit optimized t against — not a bare Euclidean stand-in.

The fit’s per-row data loss is generalized least squares ½ rᵀ M_n r (r = m(t) − x) under a per-row output metric M_n = U_n U_nᵀ (gam_problem::RowMetric), plus a per-axis latent coordinate prior Σ_a ArdAxisPrior(α_a, t_a) (the ARD Gaussian / von-Mises energy the fit placed on the coordinate). The metric-free encode drops BOTH, so its certified root solves a different problem whenever a non-identity metric or an active prior is present. EncodeObjective threads them back in:

• metric_factor — the per-row factor U_n ∈ ℝ^{p×rank}; None ⇒ M = I. Whitening reduces to M r = U(Uᵀr) and Jᵀ M J = (UᵀJ)ᵀ(UᵀJ).
• prior_alpha — per-axis ARD precision α_a (the caller folds in any row weight); None/0 ⇒ no prior on that axis. The period is read from the atom’s basis kind so a periodic axis uses the von-Mises energy.
• metric_norm_bound — a GLOBAL upper bound max_n ‖M_n‖ used to scale the offline chart Lipschitz (the per-row M_n enters β, η online; the certificate needs a valid L upper bound, and L_data scales by ‖M‖).

EncodeObjective::euclidean (all None, bound 1) reproduces the metric- free field bit-for-bit, so every existing caller is unchanged.

Fields

metric_factor: Option<ArrayView2<'a, f64>>

Per-row output-metric factor U ∈ ℝ^{p×rank} (M = U Uᵀ). None ⇒ I.

prior_alpha: Option<&'a [f64]>

Per-axis ARD precision α_a (row weight folded in). None ⇒ no prior.

metric_norm_bound: f64

Global bound max_n ‖M_n‖ scaling the offline chart Lipschitz. 1.0 for the Euclidean objective.

Implementations

impl<'a> EncodeObjective<'a>

pub fn euclidean() -> Self

The bare Euclidean, prior-free objective — every code path is bit-identical to the metric-free encode under this value.

Trait Implementations

impl<'a> Clone for EncodeObjective<'a>

fn clone(&self) -> EncodeObjective<'a>

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<'a> Copy for EncodeObjective<'a>