Skip to contents

Compute weight matrix W via an arbitrary ranking + weight mapping

Source: R/api-weights.R
weights_matrix.Rd

This is the general, ranker-agnostic top-layer. It consumes a ranking provider (distance, custom function, or precomputed ranks/orders/distances) and a weighting rule (rank_weight), then returns the weight matrix W (n x m).

Usage

weights_matrix(
  X,
  X0 = NULL,
  x0 = NULL,
  ranking,
  weight = NULL,
  ctx = NULL,
  return = c("W")
)

Arguments

X

numeric matrix (n x p), reference data.

X0

numeric matrix (m x p), target points (rows). If missing, use x0.

x0

optional numeric length-p single target (convenience for m=1).

ranking

one of:

  • a dnn_distance spec, a registered distance name (e.g., "l2","l1"),

  • a function (X, X0, ctx) -> list with $ranks/$ord/$dx, or an n x m ranks matrix,

  • precomputed object: list with $ranks, $ord, or $dx, or a ranks matrix.

weight

a rank_weight spec (or zero-arg factory returning one).

ctx

optional shared context passed to ranking and weight evaluators. On the ranking side, if ctx is NULL and a distance spec defines setup(X), that setup path is used automatically.

return

which extras to include besides W: any of c("ord","ranks","dx").

Value

An object of class weights_result with components:

  • W : numeric matrix n x m (columns typically sum to 1),

  • optional: ord, ranks, dx,

  • meta: list with n, m, p, weight_kind, weight_name.

See also

weights_dnn(), rank_weight_eval_batch(), distance_eval_batch()

Examples

set.seed(1)
x_mat <- matrix(rnorm(20), 10, 2)
x0_mat <- matrix(rnorm(6), 3, 2)
rw <- as_rank_weight(function(r) {
  w <- 1 / as.numeric(r)
  w / sum(w)
}, name = "inv_rank")
out <- weights_matrix(x_mat, x0_mat, ranking = "l2", weight = rw, return = c("W", "ranks"))
dim(out$W)
#> [1] 10  3