Skip to content

Jaccard dissimilarity matrix for PCA

5 messages · Marcelo Laia, David L Lorenz, Christian Hennig +1 more

Flabbergaster
# 
Hi
I have a large dataset, containing a wide range of binary variables.
I would like first of all to compute a jaccard matrix, then do a PCA on this
matrix, so that I finally can do a hierarchical clustering on the principal
components. 
My problem is, that I don't know how to compute the jaccard dissimilarity
matrix in R? Which package to use, and so on...
Can anybody help me?
Alternatively I'm search for another way to explore the clusters present in
my data.
Another problem is, that I have cases with missing values on different
variables.

Jacob
Christian Hennig
# 
jaccard in package prabclus computes a Jaccard matrix for you.

By the way, if you want to do hierarchical clustering, it doesn't seem to 
be a good idea to me to run PCA first. Why 
not cluster the dissimilarity matrix directly without information loss by 
PCA? (I should not make too general statements on this because generally 
how to cluster data always depends on the aim of clustering, the cluster 
concept you are interested in etc.)

prabclus also contains clustering methods for such data; have a 
look at the functions prabclust and hprabclust (however, they are 
documented as functions for clustering species distribution ranges, so if 
your application is different, you may have to think about whether and how 
to adapt them).

Hope this helps,
Christian
On Tue, 28 Dec 2010, Flabbergaster wrote:

            

      
      
      
    

        

      
      
    

        
*** --- ***
Christian Hennig
University College London, Department of Statistical Science
Gower St., London WC1E 6BT, phone +44 207 679 1698
chrish at stats.ucl.ac.uk, www.homepages.ucl.ac.uk/~ucakche

  
  


  


      

    

    

      
      

        


  

        

      Flabbergaster
    

    

      
      #
    

  


  

      
This sounds like something I could use..
I'm kind of new with R, meaning I've having some minor troubles all the
time...
Say I have a range of binary(0,1) variables X1 to Xn, with missing data for
different cases.
At the moment my data is a binary indicator matrix; rows representing the i
individuals or subjects, columns representing presence(1)/absence(0) of
various characteristics. 
Actually I have 5 groups of variables (102 variables in total), describing
different aspects of the subject(s) I'm studying (people; i.e. refugees).
O -> O1 to O43 
A -> A1 to A38
R -> R1 to R6
AP -> AP1 to AP8
PT -> PT1 to PT7

Can someone help me with the programming of a jaccard matrix in prabclus (or
in any other package). I'm having troubles defining the input-object to the
function, I think?
I get error messages like:
'x' must be an array of at least two dimensions
ERROR:  argument is not a matrix

Jacob

        
Christian Hennig wrote: