[R] Re: clara(*, stand=) and pam(*, stand=)

Martin Maechler maechler at stat.math.ethz.ch
Mon Nov 17 19:58:04 CET 2003

```>>>>> "ReCH" == optimisation1 stagiaire <optimisation1.stagiaire at lagardere-active.com>
>>>>>     on Mon, 17 Nov 2003 11:23:41 +0100 writes:

ReCH> I need informations about the clara routine. The
ReCH> on-line doc say that the argument stand is a logical,
ReCH> indicating if the measurements in x are standardized
ReCH> before calculating the dissimilarities. Measurements
ReCH> are standardized for each variable (column), by
ReCH> subtracting the variable's mean value and dividing by
ReCH> the variable's mean absolute deviation. If we note
ReCH> STAND = TRUE, I suppose that the data will not be
ReCH> standardized before clustering. On the contrary, STAND
ReCH> = FALSE means that the data will be standardized
ReCH> before clustering.  Each sub-dataset is partitioned
ReCH> into k clusters using the same algorithm as in
ReCH> pam. But the pam routine argument stand is a logical;
ReCH> if true, the measurements in x are standardized before
ReCH> calculating the dissimilarities.  Measurements are
ReCH> standardized for each variable (column), by
ReCH> subtracting the variable's mean value and dividing by
ReCH> the variable's mean absolute deviation. If x is
ReCH> already a dissimilarity matrix, then this argument
ReCH> will be ignored. If we note STAND = TRUE, I suppose
ReCH> that the data will be standardized.

ReCH> There is a big difference as clara and pam use nearly
ReCH> the same algorithm.

There is no difference between  clara() and pam() concerning the
meaning and handling of the argument `stand'.

This is very quickly seen if you look at the two functiondefinitions
which both have almost the same statement

x2 <-
if (stand)
scale(x, scale = apply(x, 2, meanabsdev))
else x

ReCH> I need to use clara because I have a large
ReCH> dataset. Could help me about the argument stand ? May
ReCH> I have to standardize my datas with excel before ? If
ReCH> yes, what I have to write : STAND = ?? ?

ReCH> Best regards,
ReCH> Cordialement,

ReCH> Régis CHARIGNON

```