Talk:Lloyd's algorithm

Lloyd's method is identical to the k-means algorithm. Oddly, the k-means article says the method "converges very quickly in practice", while this one says "the algorithm converges slowly". — Jeﬀ Erickson ✍ 04:30, 6 April 2007 (UTC)[reply]

The principal difference between Lloyd's method and the k-means algorithm is that k-means applies to a finite set of prescribed discrete entities, whereas the method described on this page applies to a continuum with a prescribed density function. The 'points' referred to on this page correspond to the 'centroids' of k-means clusters, in each case iteratively optimised to model the distribution of the prescribed data.

Mathematically these are indeed equivalent. Computationally, however, they are very different - the difference between a discrete and a continuous problem. This explains the difference in performance, and the fact that of the two methods only k-means can guarantee convergence in finitely many iterations. - Robert Stanforth 14:50, 15 May 2007 (UTC)[reply]

I think the two are different enough to remove the merge tag. Weston.pace 19:50, 12 June 2007 (UTC)[reply]