Struct CumulantArrays 

pub struct CumulantArrays {
    pub q: usize,
    pub info: Vec<f64>,
    pub nu3: Vec<f64>,
    pub nu4: Vec<f64>,
}

The exact cumulant arrays the Bartlett/Skovgaard expansions consume, over a tested coefficient block Z (the n × q design columns of the term under test). For a GLM-type log-likelihood ℓ = Σᵢ ℓᵢ(ηᵢ) with ηᵢ = xᵢᵀβ, the derivatives w.r.t. the block coefficients factor through ηᵢ by the chain rule, so every cumulant array is a row sum of the per-row η-derivative times an outer product of Z-rows:

info_{ab}     =  −Σᵢ ℓ''ᵢ · Z_{ia} Z_{ib}          (observed/expected Fisher info)
nu3_{abc}     =   Σᵢ ℓ'''ᵢ · Z_{ia} Z_{ib} Z_{ic}
nu4_{abcd}    =   Σᵢ ℓ''''ᵢ · Z_{ia} Z_{ib} Z_{ic} Z_{id}

These are exact (the per-row ℓ⁽ᵏ⁾ come from the #932 tower) and fully symmetric in their indices by construction. They are stored flattened in row-major order (nu3 length q³, nu4 length q⁴) so the consuming contraction can stride them without re-deriving the symmetry.

Fields

q: usize

Block dimension q.

info: Vec<f64>

Fisher information block info_{ab} (q × q, row-major).

nu3: Vec<f64>

Third cumulant array nu3_{abc} (q³, row-major).

nu4: Vec<f64>

Fourth cumulant array nu4_{abcd} (q⁴, row-major).

Implementations

impl CumulantArrays

pub fn info(&self, a: usize, b: usize) -> f64

pub fn nu3(&self, a: usize, b: usize, c: usize) -> f64

pub fn nu4(&self, a: usize, b: usize, c: usize, d: usize) -> f64

Trait Implementations

impl Clone for CumulantArrays

fn clone(&self) -> CumulantArrays

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for CumulantArrays

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

Formats the value using the given formatter.