%DEMGMM3 Demonstrate density modelling with a Gaussian mixture model.%%	Description%	 The problem consists of modelling data generated by a mixture of%	three Gaussians in 2 dimensions with a mixture model using diagonal%	covariance matrices.  The priors are 0.3, 0.5 and 0.2; the centres%	are (2, 3.5), (0, 0) and (0,2); the covariances are all axis aligned%	(0.16, 0.64), (0.25, 1) and the identity matrix. The first figure%	contains a scatter plot of the data.%%	A Gaussian mixture model with three components is trained using EM.%	The parameter vector is printed before training and after training.%	The user should press any key to continue at these points.  The%	parameter vector consists of priors (the column), and centres (given%	as (x, y) pairs as the next two columns).  The diagonal entries of%	the covariance matrices are printed separately.%%	The second figure is a 3 dimensional view of the density function,%	while the third shows the axes of the 1-standard deviation circles%	for the three components of the mixture model.%%	See also%	GMM, GMMINIT, GMMEM, GMMPROB, GMMUNPAK%%	Copyright (c) Ian T Nabney (1996-2001)% Generate the datandata = 500;% Fix the seeds for reproducible resultsrandn('state', 42);rand('state', 42);data = randn(ndata, 2);prior = [0.3 0.5 0.2];% Mixture model swaps clusters 1 and 3datap = [0.2 0.5 0.3];datac = [0 2; 0 0; 2 3.5];datacov = [1 1;1 0.25; 0.4*0.4 0.8*0.8];data1 = data(1:prior(1)*ndata,:);data2 = data(prior(1)*ndata+1:(prior(2)+prior(1))*ndata, :);data3 = data((prior(1)+prior(2))*ndata +1:ndata, :);% First cluster has axis aligned variance and centre (2, 3.5)data1(:, 1) = data1(:, 1)*0.4 + 2.0;data1(:, 2) = data1(:, 2)*0.8 + 3.5;% Second cluster has axis aligned variance and centre (0, 0)data2(:,2) = data2(:, 2)*0.5;% Third cluster is at (0,2) with identity matrix for covariancedata3 = data3 + repmat([0 2], prior(3)*ndata, 1);% Put the dataset together againdata = [data1; data2; data3];clcdisp('This demonstration illustrates the use of a Gaussian mixture model')disp('with diagonal covariance matrices to approximate the unconditional')disp('probability density of data in a two-dimensional space.')disp('We begin by generating the data from a mixture of three Gaussians')disp('with axis aligned covariance structure and plotting it.')disp(' ')disp('The first cluster has centre (0, 2).')disp('The second cluster has centre (0, 0).')disp('The third cluster has centre (2, 3.5).')disp(' ')disp('Press any key to continue')pausefh1 = figure;plot(data(:, 1), data(:, 2), 'o')set(gca, 'Box', 'on')% Set up mixture modelncentres = 3;input_dim = 2;mix = gmm(input_dim, ncentres, 'diag');options = foptions;options(14) = 5;	% Just use 5 iterations of k-means in initialisation% Initialise the model parameters from the datamix = gmminit(mix, data, options);% Print out modeldisp('The mixture model has three components and diagonal covariance')disp('matrices.  The model parameters after initialisation using the')disp('k-means algorithm are as follows')disp('    Priors        Centres')disp([mix.priors' mix.centres])disp('Covariance diagonals are')disp(mix.covars)disp('Press any key to continue.')pause% Set up vector of options for EM traineroptions = zeros(1, 18);options(1)  = 1;		% Prints out error values.options(14) = 20;		% Number of iterations.disp('We now train the model using the EM algorithm for 20 iterations.')disp(' ')disp('Press any key to continue.')pause[mix, options, errlog] = gmmem(mix, data, options);% Print out modeldisp(' ')disp('The trained model has priors and centres:')disp('    Priors        Centres')disp([mix.priors' mix.centres])disp('The data generator has priors and centres')disp('    Priors        Centres')disp([datap' datac])disp('Model covariance diagonals are')disp(mix.covars)disp('Data generator covariance diagonals are')disp(datacov)disp('Note the close correspondence between these parameters and those')disp('of the distribution used to generate the data.')disp(' ')disp('Press any key to continue.')pauseclcdisp('We now plot the density given by the mixture model as a surface plot.')disp(' ')disp('Press any key to continue.')pause% Plot the resultx = -4.0:0.2:5.0;y = -4.0:0.2:5.0;[X, Y] = meshgrid(x,y);X = X(:);Y = Y(:);grid = [X Y];Z = gmmprob(mix, grid);Z = reshape(Z, length(x), length(y));c = mesh(x, y, Z);hold ontitle('Surface plot of probability density')hold offdrawnowclcdisp('The final plot shows the centres and widths, given by one standard')disp('deviation, of the three components of the mixture model.  The axes')disp('of the ellipses of constant density are shown.')disp(' ')disp('Press any key to continue.')pause% Try to calculate a sensible position for the second figure, below the firstfig1_pos = get(fh1, 'Position');fig2_pos = fig1_pos;fig2_pos(2) = fig2_pos(2) - fig1_pos(4);fh2 = figure('Position', fig2_pos);h = plot(data(:, 1), data(:, 2), 'bo');hold onaxis('equal');title('Plot of data and covariances')for i = 1:ncentres  v = [1 0];  for j = 1:2    start=mix.centres(i,:)-sqrt(mix.covars(i,:).*v);    endpt=mix.centres(i,:)+sqrt(mix.covars(i,:).*v);    linex = [start(1) endpt(1)];    liney = [start(2) endpt(2)];    line(linex, liney, 'Color', 'k', 'LineWidth', 3)    v = [0 1];  end  % Plot ellipses of one standard deviation  theta = 0:0.02:2*pi;  x = sqrt(mix.covars(i,1))*cos(theta) + mix.centres(i,1);  y = sqrt(mix.covars(i,2))*sin(theta) + mix.centres(i,2);  plot(x, y, 'r-');endhold offdisp('Note how the data cluster positions and widths are captured by')disp('the mixture model.')disp(' ')disp('Press any key to end.')pauseclose(fh1);close(fh2);clear all;