SDEACCLUST Evidence Accumulation Clustering

    A2=SDEACLUST(A)
    [A2,W,CA]=SDEACLUST(A,N,options);

    B2=SDEACLUST(W,B)  % apply to new data B

 INPUT
    A       Data set to be clustered
    N       Number of clusterings to accumulate (opt, default: 100)

 OUTPUT
    A2      Data set with cluster labels (and old labels in A)
    W       Structure needed to apply clustering to new data
    CA      Co-association matrix (probability)

 OPTIONS
    'k'     k for k-means, [kmin kmax], default: [3 20]
    'link'  type of linkage used in automatic threshold-setting (def: 'c')
    'feat'  use only specific subset of features in each step. Provide
            a logical matrix (features x N) where 1 indicates that feature
            should be used.

 DESCRIPTION
 SDEACLUST implements evidence accumulation clustering. A simple
 clustering procedure such as k-means is executed N times with randomized
 settings (K). The result is co-association matrix CA estimating a
 probability that two samples belong to the same cluster. CA is clustered
 using other hierarchical clustering. The solution with maximum lifetime
 in complete linkage is returned. The clustering may be applied to new
 data using the structure W.  SDEACLUST may run k-means algorithm in
 user-specified subspaces using the 'feat' option.

 REFERENCE
 Data Clustering Using Evidence Accumulation, A.L.N.Fred and A.K.Jain,
 Proc of ICPR 2002.