Skip to content

k- means cluster analysis

3 messages · Ngayee J Law, Sundar Dorai-Raj, Brian Ripley

Original
Ngayee J Law
# 
Hi all,

I am trying to run the k-means cluster analysis using the function kmeans
in the package cluster.
The data are:
x = c(-0.26, -0.23, -0.05, -0.20,  0.30, -0.84, -0.10, -0.12,  0.10, -0.31,
-0.19,  0.18, -0.26,
      -0.23, -0.37, -0.23)
I've got two different solutions when I ran this function over a few times:
kmeans(x, centers=2)

The first solution gives the following:
$cluster
 [1] 2 2 1 2 1 2 2 2 1 2 2 1 2 2 2 2
$centers
        [,1]
1  0.1325000
2 -0.2783333
$withinss
[1] 0.0646750 0.4033667
$size
[1]  4 12

The second solution gives the following:
$cluster
 [1] 1 1 1 1 1 2 1 1 1 1 1 1 1 1 1 1
$centers
        [,1]
1 -0.1313333
2 -0.8400000
$withinss
[1] 0.5035733 0.0000000
$size
[1] 15  1

I don't understand why this is happening, and how do I choose between the
two solutions. Also, how can I ensure
consistent solution over times? Thanks a lot!

- Jacqueline
Sundar Dorai-Raj
#
Ngayee J Law wrote:
From the help page for `kmeans':

  centers: Either the number of clusters or a set of initial cluster
           centers. If the first, a random set of rows in `x' are chosen
           as the initial centers.

If you want the same results try supplying an initial center, as in:

kmeans(x, centers = c(0.1, -0.2))

However, choosing bad starting values could cause kmeans to crash, as in:

kmeans(x, centers = c(0, 0))

Regards,
Sundar
Brian Ripley
#
On Wed, 12 Feb 2003, Sundar Dorai-Raj wrote:

            

      
      
      
    

        
I think it's the one in package mva.

[...]

            

      
      
      
    

        
Really?  That does not crash here, but correctly reports an error message:

            

      
      
      
    

        
Error in switch(Z$ifault, stop("empty cluster: try a better set of initial 
centers"),  : 
        empty cluster: try a better set of initial centers

If you actually have a crash, please report to R-bugs.

There is an increasing trend for people to describe informative error 
messages describing their errors as `crashes', but that is confusing at 
best.