Skip to contents

K-means recording the full assignment trajectory

Source: R/kmeans.R
kmeans_with_trajectory.Rd

Runs Lloyd's k-means while recording the cluster assignment vector at every iteration (the "trajectory"), which is needed for trajectory-based selective conditioning.

Usage

kmeans_with_trajectory(
  x,
  k = 3,
  max_iter = 50,
  init_centers_idx = NULL,
  seed = NULL,
  init_method = c("random_indices", "kmeans++")
)

Arguments

x

A numeric data matrix with observations in rows.

k

Number of clusters.

max_iter

Maximum number of Lloyd iterations.

init_centers_idx

Optional integer vector of length k giving row indices of x to use as initial centers. If NULL, centers are chosen by init_method.

seed

Optional integer seed applied locally (global RNG restored on exit).

init_method

Initialization scheme when init_centers_idx is NULL: "random_indices" (default, data-independent) or "kmeans++".

Value

A list with elements cluster (final assignments), trajectory (a list of per-iteration assignment vectors), center_history, init_centers_idx, and n_iter.

Details

This function does not silently re-seed the global RNG. Reproducibility is the caller's responsibility: seed once at the call site, or pass an explicit init_centers_idx. If seed is supplied it is applied locally and the global RNG stream is restored on exit, so calling this inside an MCMC loop can never freeze the chain's proposals.