Skip to contents

Fast delete-one jackknife for rank-weighted functionals (auto C++ when available, R fallback otherwise)

Source: R/jackknife-rank.R
jackknife_fast_rank.Rd

Computes the baseline aggregate theta0 and all delete-one aggregates theta_loo for generic rank-weighted L-stat rules with finite support s.

Usage

jackknife_fast_rank(ranks, a, phi, normalize = TRUE)

Arguments

ranks

Integer matrix (T x (s+1)) of positive neighbor indices (1-based). Each row t lists the first s+1 neighbors of target t.

a

Finite numeric length-s vector of L-stat coefficients a_1..a_s.

phi

either:

  • finite numeric matrix (n_ref x q): row j is the q-vector phi(j), or

  • function function(j, t) returning a finite numeric length-q vector (allows target dependence). For a scalar \(\phi\), use cbind(y) as the matrix case.

normalize

Scalar logical; if TRUE (default) divide by sum(a).

Value

list with:

  • theta0: numeric length-q baseline aggregate,

  • theta_loo: numeric matrix (n_ref x q) of delete-one aggregates,

  • info: list(s, T, q, normalize).

Details

This utility is for rules that are fully characterized by:

  • the first s + 1 ranked observations for each target, and

  • a finite coefficient vector a_1, ..., a_s.

It is therefore appropriate for compact-support L-statistics. The user-level DNN wrappers (np_dnn(), cate_dnn(), plqr_dnn()) do not call this helper directly for exact DNN weights. Those wrappers keep the full rank matrix and reevaluate the DNN rank-weight rule after delete-one updates.

  • If phi is a numeric matrix (n_ref x q) with rows phi[j, ] = phi(j) and the package was compiled with C++ support, this calls the C++ fast path.

  • Otherwise, it uses a general R fallback that also supports target-dependent per-neighbor contributions via a function phi(j, t).

Examples

set.seed(1)
n <- 20
p <- 2
X <- matrix(rnorm(n * p), n, p)
ranks <- cbind(matrix(sample.int(n, n * 3, replace = TRUE), n, 3), sample.int(n, n, replace = TRUE))
a <- c(1, 1, 1)
phi <- cbind(rnorm(n))
out <- jackknife_fast_rank(ranks, a, phi, normalize = TRUE)
str(out$theta0)
#>  num -0.311