how to calculate the consistency of different clusterings
Hello, I've been waiting to see if anyone else would answer this. I've previously used random reallocation of objects to groups (clusters) as a monte-carlo test of the informativeness of groups, as described here: http://lastresortsoftware.blogspot.com/2010/09/monte-carlo-testing-of-classification.html However, in your case it sounds like you want to investigate the influence of particular attributes (traits) or groups of attributes on the classification - is that correct ? If so, I can probably help with some R code but I'd need to know the clustering method you are using (e.g. hclust). Michael
On 14 January 2011 02:36, Mao Jianfeng <jianfeng.mao at gmail.com> wrote:
Dear R-listers,
I do clustering on tens of individuals by thousands of traits. I have
known the assignment of each individual. I want to classify the
individuals by randomly resampling different subsets of the traits,
for example, randomly resampling 100 traits for 100 times, then 200
traits for 100 times, then 300 traits for 100 times, ,,,,,,. By each
subset of traits, I do clustering of the same individuals.
In the end, I want to get the consistency (in percentage) of each of
these clusterings (as examples, here "cluster.1", "cluster.2" and
"cluster.3" in the dummy data) with the assignment which is already
known ("populations" in the dummy data). I want to know how such work
can be implemented, maybe by using R.
#dummy data,
clus.data <- data.frame(individual = paste("ind", 1:12, sep = ""),
populations = c(rep("popA", 5), rep("popB", 7)), cluster.1 = c(rep(1,
5), rep(2, 7)), cluster.2 = c(rep(2, 4), rep(1, 8)), cluster.3 =
c(rep(4, 7), rep(5, 5)))
clus.data
Thanks.
--
Jian-Feng, Mao
the Institute of Botany,
Chinese Academy of Botany,
______________________________________________ R-help at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.