Gllobally optimal clustering in STATA

[Marco Greco]

Hello statalisters,
during my researches I came across several advanced clustering methods that, differently from K-means, would allow finding globally optimal clustering instead of local ones.
Is there any of these methods available on Stata 14?

Thanks,
Marco

[Phil Bromiley]

Look them up - the manual explains the clustering available. Also, enter findit cluster in the command window to see if a user written procedures does what you need.

[Brendan Halpin]

I have just put up an implementation of Partitioning Around Medoids at http://teaching.sociology.ul.ie/statacode. PAM attempts a global maximisation.

Try:

net from http://teaching.sociology.ul.ie/statacode
net describe clutils
net install clutils

If you use the code, please let me know how you get on. It hasn't been widely tested.

Working example:

set matsize 1000
sysuse citytemp
keep if !missing(tempjan)
gen id = _n
sort id
matrix dissim dd = temp*, L2squared
clpam k4, dist(dd) id(id) medoids(4) many
separate tempjan, by(k4)
scatter tempjan1-tempjan4 tempjul, legend(off)