source: distools/intrdim.m @ 100

%INTRDIM Estimate Intrinsic dimension from dissimilarity data
%
%       K = INTRDIM(D)
%
% INPUT
%   D   NxN Dissimilarity dataset
%
% OUTPUT
%   K   Estimated intrinsic dimension
%
% DESCRIPTION
% It is assumed that the data is generated from a normal distribution
% in K dimensions and that the Eulcidean distance measure has been used
% to compute D. 
% Do NOT supply squared distances! They will definitely generate the wrong
% result.
%
% EXAMPLE
% A = GENDATB(100);
% D = SQRT(DISTM(A));
% K = INTRDIM(D);
%
% A = GENDATD(100,10);
% D = SQRT(DISTM(A*SCALEM(A,'VARIANCE')));
% K = INTRDIM(D);
%

% Copyright: R.P.W. Duin, r.p.w.duin@prtools.org
% Faculty EWI, Delft University of Technology
% P.O. Box 5031, 2600 GA Delft, The Netherlands

function k = intrdim(d)

d = +d.^2;
[m,n] = size(d);

% Assume selfreflecting dissim matrix
if m==n & all(diag(+d) == 0)  
        d(1:n+1:n*n) = [];
end

k = round(2*(mean(d(:)).^2)/var(d(:))); % based on Chi square distribution