% demo script for Gaussian mixtures (CS 274A, Winter 2009)



% DATA SET 3: a real-world biomedical data setw with 2 real
% groups (Christine McLaren red blood cell/anemia data set)
[params,memberships] =  gaussian_mixture(dataset3,3,2,123,200);
[params,memberships] =  gaussian_mixture(dataset3,3,2,123,200);


% DATA SET 1: two simulated Gaussians in 2d
% bad local maximum
[params,memberships] =  gaussian_mixture(dataset1,2,2,123,200);

% much faster when we initialize with k-means.....
[params,memberships] =  gaussian_mixture(dataset1,2,1,123,200);

% try more Gaussians for fun....
[params,memberships] =  gaussian_mixture(dataset1,5,1,12345,200);



% DATA SET 2: three simulated Gaussians in 2d  ("butterfly configuration")

% for this data, random initialization is actually better than k-means
[params,memberships] =  gaussian_mixture(dataset2,3,2,123,200);
% now with kmeans...
>> [params,memberships] =  gaussian_mixture(dataset2,3,1,1234,100);



% DATA SET 4: another biomedical data set
[params,memberships] =  gaussian_mixture(dataset4(:,2:3),3,2,123,200);


% Ripley's 4 Gaussian simulated data set.....
[params,memberships] =  gaussian_mixture(ripdata,4,1,123,200);
% then with random initialization...
[params,memberships] =  gaussian_mixture(ripdata,4,2,123,200);


% non-Gaussian data - note narrow Gaussians on discrete-valued x-axis
[params,memberships] =  gaussian_mixture(newdata2,8,1,123,1000);