Weighted k-Nearest Neighbors

This function computes a weighted k-nearest neighbors graph or adjacency matrix from a data matrix. The function takes into account the Euclidean distance between instances and applies a kernel function to convert the distances into similarities.

Usage

weighted_knn(
  X,
  k = 5,
  FUN = heat_kernel,
  type = c("normal", "mutual", "asym"),
  as = c("igraph", "sparse"),
  backend = c("nanoflann", "hnsw"),
  M = 16,
  ef = 200,
  ...
)

Arguments

X

A data matrix where rows are instances and columns are features.

k

An integer specifying the number of nearest neighbors to consider (default: 5).

FUN

A kernel function used to convert Euclidean distances into similarities (default: heat_kernel).

type

A character string indicating the type of k-nearest neighbors graph to compute. One of "normal", "mutual", or "asym" (default: "normal").

as

A character string specifying the format of the output. One of "igraph" or "sparse" (default: "igraph").

backend

Nearest-neighbor backend. `"nanoflann"` uses exact Euclidean search; `"hnsw"` uses approximate search via `RcppHNSW`.

M, ef

HNSW tuning parameters used only when `backend="hnsw"`.

...

Additional arguments passed to the nearest neighbor search function (Rnanoflann::nn).

Value

If 'as' is "igraph", an igraph object representing the weighted k-nearest neighbors graph. If 'as' is "sparse", a sparse adjacency matrix.

Details

Distances passed to `FUN` are Euclidean distances. With `backend="hnsw"`, squared L2 distances from `RcppHNSW` are converted back to Euclidean distances before weighting.

Examples

X <- matrix(rnorm(10 * 10), 10, 10)
w <- weighted_knn(X, k = 5)