Using different distances with evalclusters

Hey everyone!

I would like to use the evalclusters function with linkage method. However, the doc for the function states that specifying 'linkage' in input in evalclusters will use agglorometive clustering with ward's distance. However, I would like to use complete distance as opposed to ward's. I've tried this to no avail:

f = @(X)linkage(X,'complete');
eva = evalclusters(Data,f,'klist',[1:6]);

All this does is return an empty Evaluation object with NaNs as the outputs.

How would I go about specifying distance in these functions?

Any help would be great. Thanks!

Answer 1

[Shraddha Jain]

Hi Manash,

The second input argument clust in the function evalclusters refers to the clustering algorithm that is used.

When clust is specified as 'linkage', it means that clusterdata agglomerative clustering algorithm will be used to cluster the given input data x with the algorithm for computing the distance between clusters'Linkage' pre-defined to 'ward'.

This 'Linkage'algorithm could certainly be changed to something other than 'ward' by speifying it in a function handle using clusterdata and passing that as the clust argument in evalclusters,

f = @(x,k) clusterdata(x,'linkage','complete','maxclust',k);
eva = evalclusters(Data,f,'klist',[1:6]);

The reason that Ward Linkage is used as default in clusterdata as it the minimum variance method, therefore it minimizes the total within-cluster variance.

Hope this helps!