MYPROXM MyProximity mapping W = MYPROXM(A,TYPE,P,G) Computation of the k*m proximity mapping (or kernel) defined by the m*k dataset A. The proximities are defined by the following possible TYPEs: 'linear' | 'l': a*b' 'polynomial' | 'p': sign(a*b'+1).*(a*b'+1).^p 'exponential' | 'e': exp(-(||a-b||)/p) 'radial_basis'| 'r': exp(-(||a-b||.^2)/(p*p)) 'sigmoid' | 's': sigm((sign(a*b').*(a*b'))/p) 'distance' | 'd': ||a-b||.^p 'minkowski' | 'm': sum(|a-b|^p).^(1/p) 'city-block' | 'c': sum(|a-b|) 'gower' | 'g': gower-dissimilarity (see gower.m) 'kcenter' | 'k': k-center prototype In the polynomial case and p not integer D is computed by D = sign(d)*abs(d).^p in order to avoid problems with negative inner products d. The features of the objects in A may be weighted by the weights in the vector g (default 1). Default is the Euclidean distance: type = 'distance', p = 1
W = MYPROXM(A,TYPE,P,G,NR_PROTO) W = MYPROXM(A,TYPE,P,G,FRAC_PROTO)
For situations where the dataset is (very) large, you can subsample the dataset and randomly choose NR_PROTO prototypes (or a fraction FRAC_PROTO of the dataset). See also: |mappings| |datasets| gower sqeucldistm Copyright: D.M.J. Tax, D.M.J.Tax@prtools.org Faculty EWI, Delft University of Technology P.O. Box 5031, 2600 GA Delft, The Netherlands