Struct TermCollectionSpec 

pub struct TermCollectionSpec {
    pub linear_terms: Vec<LinearTermSpec>,
    pub random_effect_terms: Vec<RandomEffectTermSpec>,
    pub smooth_terms: Vec<SmoothTermSpec>,
}

Fields

linear_terms: Vec<LinearTermSpec>

random_effect_terms: Vec<RandomEffectTermSpec>

smooth_terms: Vec<SmoothTermSpec>

Implementations

impl TermCollectionSpec

pub fn write_structural_shape_hash(&self, h: &mut Fingerprinter)

Write this collection’s topology identity into a warm-start cache fingerprint (#869).

The persistent warm-start cache_key hashes only family + raw input dimensions, so two fits on the same data that differ only in their smooth topology (the s(..., type=AUTO) candidate enumeration: sphere vs torus vs euclidean vs duchon) collide on one key and seed each other with geometrically incompatible β/ρ. Folding the per-term structural kind + feature columns + linear/random-effect counts into the shape hash gives each candidate its own key, so the screen→full-refit reuse of one candidate is preserved while cross-candidate contamination is removed. Only the structural identity is hashed (not fitted coefficients or frozen knot values), so a refit of the same topology still hits.

pub fn validate_frozen(&self, label: &str) -> Result<(), String>

Validate that a term collection spec represents a fully frozen model (i.e. all knots/centers are pre-computed, identifiability transforms are baked in, and random-effect levels are fixed).

pub fn remap_feature_columns<E, F>( &self, remap: F, ) -> Result<TermCollectionSpec, E>

where F: FnMut(usize) -> Result<usize, E>,

Re-resolve every stored feature-column index through remap, returning a spec that addresses a different column layout.

A frozen TermCollectionSpec stores feature columns as absolute indices into the training table. To replay it on a fresh dataset whose columns sit at different positions — the common case at prediction time, where the response column is unknown and may be absent entirely — every index must be re-resolved against the new layout. remap receives each training-table index and returns its position in the runtime table; callers typically implement it as “look the name up in the training headers, then resolve that name against the prediction dataset”.

This is the single authority on which fields carry a column index across every basis variant (linear, random-effect, the by= column of ByVariable/FactorSumToZero/BySmooth, the continuous and group columns of a FactorSmooth, and the multi-axis feature_cols of every spatial/tensor basis), so a predict-time realignment cannot silently miss one and dereference a stale training index.

Trait Implementations

impl Clone for TermCollectionSpec

fn clone(&self) -> TermCollectionSpec

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for TermCollectionSpec

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for TermCollectionSpec

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

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for TermCollectionSpec

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

where __S: Serializer,

Serialize this value into the given Serde serializer.