Struct LinearTermSpec 

pub struct LinearTermSpec {
    pub name: String,
    pub feature_col: usize,
    pub feature_cols: Vec<usize>,
    pub categorical_levels: Vec<(usize, u64)>,
    pub double_penalty: bool,
    pub coefficient_geometry: LinearCoefficientGeometry,
    pub coefficient_min: Option<f64>,
    pub coefficient_max: Option<f64>,
}

Fields

name: String

feature_col: usize

Primary feature column index. For Wilkinson-Rogers : interaction terms (a:b[:c...]) this is the first column in feature_cols; the realized design column is the elementwise product across every entry of feature_cols. Plain (non-interaction) linear terms set feature_cols == vec![feature_col].

feature_cols: Vec<usize>

Full list of columns whose elementwise product yields this term’s design column. len() >= 1; len() == 1 is a plain linear effect.

categorical_levels: Vec<(usize, u64)>

Categorical-level gates for a factor-aware : interaction.

Each (col, level_bits) multiplies the realized design column by the indicator 1[data[row, col].to_bits() == level_bits]. This is how a factor:x (or factor:factor) interaction is expanded: build_termspec emits one LinearTermSpec per surviving cell of the categorical operand(s) (treatment-coded, first level dropped per factor), each carrying the numeric operands in feature_cols and the cell’s level gate(s) here. Empty for a plain numeric : interaction or main effect, in which case the realized column is exactly the numeric product.

double_penalty: bool

Optional ridge (S = I, REML-selected λ) on this linear coefficient. A parametric linear term carries no wiggliness, so it is unpenalized by default — gam reports the MLE, matching mgcv/glm/survreg/VGAM (which penalize parametric terms only under an explicit paraPen). Set true to opt into an explicit shrinkage ridge (a zero-mean Gaussian prior β ~ N(0, λ⁻¹)); doing so adds one outer REML smoothing coordinate.

coefficient_geometry: LinearCoefficientGeometry

coefficient_min: Option<f64>

coefficient_max: Option<f64>

Implementations

impl LinearTermSpec

pub fn effective_feature_cols(&self) -> Vec<usize>

Return the effective list of feature columns. Backfills from feature_col for legacy specs that predate the multi-column field.

pub fn is_interaction(&self) -> bool

True when this term is a Wilkinson-Rogers : interaction (multi-col).

pub fn realized_design_column( &self, data: ArrayView2<'_, f64>, ) -> Result<Array1<f64>, String>

Realize this linear term’s (n,) design column from data.

The column is the elementwise product of every numeric feature column (effective_feature_cols) gated by the categorical-level indicators in categorical_levels: each (col, level_bits) multiplies the running column by 1[data[row, col].to_bits() == level_bits]. A plain numeric term (no categorical_levels) reduces to the bare product, matching the historical behaviour. A pure categorical interaction (empty feature_cols, non-empty categorical_levels) reduces to the cell indicator. Bounds are validated here; the returned column has length data.nrows().

Trait Implementations

impl Clone for LinearTermSpec

fn clone(&self) -> LinearTermSpec

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for LinearTermSpec

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for LinearTermSpec

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

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for LinearTermSpec

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

where __S: Serializer,

Serialize this value into the given Serde serializer.