[R] classification algorithms with distance matrix

[Peter Langfelder]

On Mon, Jun 7, 2010 at 9:05 AM, sidahmed BENABDERRAHMANE
<sidahmed.benabderrahmane at loria.fr> wrote:
> Dear all,
>
> I have a problem when using some classification functions (Kmeans, PAM,
> FANNY...)  with a distance matrix, and i would to understand how it proceeds
> for the positioning of centroids after one execution step.
>
> In fact, in the classical formulation of the algorithm, after each step, to
> re-position the center, it calculates the distance between any elements of
> the old cluster and its center, or in my case I used  a distance matrix , so
> I do not understand how the algorithm can recalculates the distance between
> the newly generated center and other elements of the cluster knowing that
> the new center does not appear in the distance matrix and there is no way to
> calculate the distance between him and the elements of a cluster. ?
>

In my understading PAM stands for Partitioning Around Medoids. The
medoid of a cluster is the object with the smallest mean distance to
all other objects in the cluster. In each iteration, all objects are
assigned to their nearest medoid, and the medoids are updated if
necessary. The distance matrix is never recalculated. Someone please
correct me if I'm wrong.

Peter