In SOM, (how) can I extract rows in each cluster?

Question

[english.server]

In SOM clustering, (how) can I extract rows in each cluster?

library(kohonen)
a= matrix (rnorm(4000),500 )
row.names(a)= paste('r', 1:500, sep='')
s= som (a, grid= somgrid ( 3,4,'hexagonal'))
plot (s)

I used either

s$data
s$codes

with no success. Thank you in advance.

Answer 1

[Kevin Blighe]

These are held in the unit.classif object. For example, I re-ran your code and my random data-matrix produced 12 clusters. A quick check of unit.classif reveals 12 categorical levels:

levels(factor(s$unit.classif))
 [1] "1"  "2"  "3"  "4"  "5"  "6"  "7"  "8"  "9"  "10" "11" "12"

You can then create a simple data-frame with the assignments, along with each row's distance to its respective assigned cluster:

res <- data.frame(rownames(a), s$distances, s$unit.classif)
colnames(res) <- c("rowname","distFromClus","Cluster")
res[1:20,]
   rowname distFromClus Cluster
1       r1     6.705452      11
2       r2     3.816565       3
3       r3     3.501137       5
4       r4     4.911634       2
5       r5     6.684077       6
6       r6    10.983671      11
7       r7     8.375258       3
8       r8     1.739472       3
9       r9     3.370915      12
10     r10     3.411521       1
11     r11     4.112711       8
12     r12     3.161650       4
13     r13     3.938754       5
14     r14     3.193206      12
15     r15     5.757975       6
16     r16     4.449785      11
17     r17     4.633186      11
18     r18     5.883848       8
19     r19    10.179297       4
20     r20     3.014707       2

Kevin

[english.server]

Thank you indeed very much. It works.

[Kevin Blighe]

You're welcome, english.server