Struct ResidualFactorInput 

pub struct ResidualFactorInput<'a> {
    pub residuals: ArrayBase<ViewRepr<&'a f64>, Dim<[usize; 2]>>,
    pub activity: ArrayBase<ViewRepr<&'a f64>, Dim<[usize; 1]>>,
    pub max_factor_rank: usize,
}

Estimator inputs: the residual matrix and the smooth activity coordinate.

residuals is R ∈ ℝ^{n×p}. activity is z ∈ ℝ^n — the coordinate the scale law c(z) is smooth in (e.g. an assignment-mass or activation-strength summary per row). When no genuine activity coordinate is available, passing a constant z recovers a homoscedastic factor model (c(z) ≡ const).

Fields

residuals: ArrayBase<ViewRepr<&'a f64>, Dim<[usize; 2]>>

Residual matrix R ∈ ℝ^{n×p}.

activity: ArrayBase<ViewRepr<&'a f64>, Dim<[usize; 1]>>

Activity coordinate z ∈ ℝ^n the scale law is smooth in.

max_factor_rank: usize

Maximum factor rank the evidence ladder is allowed to consider. The ladder scores r = 0, 1, …, min(max_factor_rank, p−1) and keeps the penalized-evidence maximizer. 0 forces the pure-diagonal model.