Struct SpatialLengthScaleOptimizationOptions 

pub struct SpatialLengthScaleOptimizationOptions {
    pub enabled: bool,
    pub max_outer_iter: usize,
    pub rel_tol: f64,
    pub log_step: f64,
    pub min_length_scale: f64,
    pub max_length_scale: f64,
    pub pilot_subsample_threshold: usize,
}

Fields

enabled: bool

Enable outer-loop optimization over spatial κ (= 1 / length_scale) for supported radial-kernel smooths. This applies to ThinPlate, Matérn, and Duchon terms.

max_outer_iter: usize

Maximum number of outer iterations in the exact joint [rho, psi] solve.

rel_tol: f64

Relative improvement threshold for terminating the outer solve.

log_step: f64

Initial log(length_scale) perturbation used for seed construction.

min_length_scale: f64

Minimum allowed length_scale during κ search.

max_length_scale: f64

Maximum allowed length_scale during κ search.

pilot_subsample_threshold: usize

Automatic geometry-initializer threshold for biobank-scale spatial fits.

When n exceeds twice this value, the fitter uses a spatially stratified subsample only to seed κ/anisotropy geometry: centers are resolved, axis contrasts are initialized from center/data spread, and one or two cheap ψ reseeding updates are applied. It never runs PIRLS, REML, ARC, BFGS, or any recursive optimizer on the pilot.

The final coefficients, smoothing parameters, and spatial geometry are always optimized on the full dataset.

Set to 0 to skip the pilot geometry initializer.

Implementations

impl SpatialLengthScaleOptimizationOptions

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

Validate the struct’s invariants. Callers that construct these options from external input (CLI, config, Python API) should call this before passing the options into the fitter. Returns Err with a descriptive message when an invariant is violated; the fitter then panics or returns EstimationError at its own boundary.

Invariants:

• min_length_scale > 0, finite
• max_length_scale > 0, finite
• min_length_scale < max_length_scale
• rel_tol > 0, finite
• log_step > 0, finite

These invariants are what the downstream κ-bound and ψ-window code assumes (-log(max_ls) must be finite, (min,max) must not be inverted, etc.). Without validation, invalid options produce silent NaN-propagation inside the outer optimizer.

Trait Implementations

impl Clone for SpatialLengthScaleOptimizationOptions

fn clone(&self) -> SpatialLengthScaleOptimizationOptions

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for SpatialLengthScaleOptimizationOptions

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

Formats the value using the given formatter.

impl Default for SpatialLengthScaleOptimizationOptions

fn default() -> Self

Returns the “default value” for a type.