Looking for an algorithm to deal with purely categorical data. It was suggested to me to look into the K-medoids algorithm. Anyone know if there is a K-medoids algorithm R library(package)?
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
1
K-medoids of course works for purely categorical data.
But there also is k-modes, easy to find with a little bit of research... It's much cheaper to run, it is O(n), not O(n²)
It doesn't solve the problem that categorical data usually will not cluster in a meaningful way...
answered Nov 23, 2018 at 18:04
-
$\begingroup$ Categorical data without a (partial) order usually does not cluster in a meaningful way. In fact introducing partial order, i.e. a structure, is adding messing meaning to categories that otherwise would be just labels. $\endgroup$– maptoCommented Nov 24, 2018 at 6:44