Re: clustering

Art Kendall wrote:

< ginormous snip >

> Today one seldom takes a single number of clusters as stable until
> different algorithms and starts of the same algorithm for k-means show
> consistent results.


That is definitely NOT true!

When it comes to "clustering", today is no different from 40 years ago
when I was first interested in the problem:

1. No one defined what constitutes a "cluster"!
2. No one knows what clusters they found are "real" (any 4 yr old
when shown will know it's a cluster) or not.

A picture is worth a thousand clusters of words.

(a) How does your statement apply to these clusters?
(b) Will k-means EVER find the two 4-yr-old-proof clusters?

x x x x x x x
x x x x
x x
x x x x
x x x x
x x x x
x x x x
x x x x x x x


-- Bob.

.

Quantcast