The LDR functions all operate in the same fashion.

 X: p*n matrix, p > n.
 The p rows are features, the n columns are samples.
 LDR is intended for the case p >> n.
 Returns: (n-1)*n matrix

 In R, instead of kmeans(Y, ...), use kmeans(t(ldr(t(Y))), ...), Y is n*p, p > n

 Do not use if p <= n.
 If p <= n, returns X itself unmodified.