gppdc

Generic Projection Pursuit Divisive Clustering
[IDX,T] = GPPDC(X, K, PPHANDLE, VARARGIN)

  [IDX, T] = GPPDC(X, K, PPHANDLE) produces a divisive hierarchical clustering of the 
  N-by-D data matrix (X) into (K) clusters, using binary partitions produced by a generic
  projection pursuit function defined in the function handle (PPHANDLE).

  [IDX, T] = GPPDC(X, K, PPHANDLE, 'PARAM1',val1, 'PARAM2',val2, ...) specifies optional parameters
  in the form of Name,Value pairs. 

  Inputs:
     (X): N-by-D data matrix
    (K): Number of clusters
    (PPHANDLE): A handle to a function of the type:
        [V,FVAL,IDX] = PPHANDLE(X,PARAMS), where,
            (X): X is the data matrix
            (PARAMS): Is a structure containining all the parameters of the projection pursuit algorithm
            Returns:
            (V): Optimal projection matrix/ vector
            (FVAL): Value of projection index for (v) [default projection index]
                (If PP algorithm aims to minimise the projection index then -fval must
                be returned, since at each step cluster with maximum 'split_index' is partitioned)
            (IDX): Cluster assignment in {1,2}

  Optional Parameters:

  'param' - Structure containing parameter settings employed by PPHANDLE
    The contents of this structure are provided as input to (PPHANDLE, see above)

  'split_index' - Criterion determining which cluster to split
    Function Handle: index = split_index(v, X, pars)
            (v: projection vector, X:data matrix, pars: parameters structure)
    Cluster with MAXIMUM INDEX is split at each step of the algorithm
    Two standard choices of split index can be enabled by settgin 'split_index' to 
    one of the strings below:
        + 'fval':    Split cluster whose hyperplane achieves the lowest density integral
        + 'size':    Split largest cluster
    (default: split_index = 'fval' estimated in PPHANDLE)

  'labels' - true cluster labels. Used only to evaluate quality of binary partitions at the end