Performs hierarchical clustering of data using specified method and
seraches for optimal cutoff empoying VIF criterion suggested in "Okada Y. et al - Detection of Cluster Boundary in Microarray Data by Reference to MIPS Functional Catalogue Database (2001)".
Namely, it searches cutoff where groups are independent. The techinque uses an econometric approach of verifying that variables in
multiple regression are linearly independent: if all the diagonal
elements of inverse correlation matrix of data are less than VIF (as
rule of thumb VIF=10).
Searching procedure is the variaition of bisection method, so it's
complexity is log(n) at most. At each iteration it chooses one item
from every clusters, constructs correlation matrix of these items and
look at diagonal element of its inverse.
Denis, it looks good, but explanation are not very helpful for beginners like me.
So I used the function as "ClusterData(R, 'ward', 10, 1)" , what and where is the output of your function, how should I apply this to my clustering then?
and an explanation that compare the VIF criterion (somehow set as 10) to other methods could be nice as well.
Denis, How do you interpret the return cutoff variable? Also, how can I find the exact cuttoff value (the optimal cuttoff).