mahito-sugiyama

kmeansp2 <- function(x, k, iter.max = 10, nstart = 1, ...) {
  n <- nrow(x) # number of data points
  centers <- numeric(k) # IDs of centers
  distances <- matrix(numeric(n * (k - 1)), ncol = k - 1) # distances[i, j]: The distance between x[i,] and x[centers[j],]
  res.best <- list(tot.withinss = Inf) # the best result among <nstart> iterations
  for (rep in 1:nstart) {
    pr <- rep(1, n) # probability for sampling centers
    for (i in 1:(k - 1)) {
      centers[i] <- sample.int(n, 1, prob = pr) # Pick up the ith center
      distances[, i] <- colSums((t(x) - x[centers[i], ])^2) # Compute (the square of) distances to the center