Mercurial > hg > camir-aes2014
comparison toolboxes/FullBNT-1.0.7/netlab3.3/demgmm5.m @ 0:e9a9cd732c1e tip
first hg version after svn
author | wolffd |
---|---|
date | Tue, 10 Feb 2015 15:05:51 +0000 |
parents | |
children |
comparison
equal
deleted
inserted
replaced
-1:000000000000 | 0:e9a9cd732c1e |
---|---|
1 %DEMGMM5 Demonstrate density modelling with a PPCA mixture model. | |
2 % | |
3 % Description | |
4 % The problem consists of modelling data generated by a mixture of | |
5 % three Gaussians in 2 dimensions with a mixture model using full | |
6 % covariance matrices. The priors are 0.3, 0.5 and 0.2; the centres | |
7 % are (2, 3.5), (0, 0) and (0,2); the variances are (0.16, 0.64) axis | |
8 % aligned, (0.25, 1) rotated by 30 degrees and the identity matrix. The | |
9 % first figure contains a scatter plot of the data. | |
10 % | |
11 % A mixture model with three one-dimensional PPCA components is trained | |
12 % using EM. The parameter vector is printed before training and after | |
13 % training. The parameter vector consists of priors (the column), and | |
14 % centres (given as (x, y) pairs as the next two columns). | |
15 % | |
16 % The second figure is a 3 dimensional view of the density function, | |
17 % while the third shows the axes of the 1-standard deviation ellipses | |
18 % for the three components of the mixture model together with the one | |
19 % standard deviation along the principal component of each mixture | |
20 % model component. | |
21 % | |
22 % See also | |
23 % GMM, GMMINIT, GMMEM, GMMPROB, PPCA | |
24 % | |
25 | |
26 % Copyright (c) Ian T Nabney (1996-2001) | |
27 | |
28 | |
29 ndata = 500; | |
30 data = randn(ndata, 2); | |
31 prior = [0.3 0.5 0.2]; | |
32 % Mixture model swaps clusters 1 and 3 | |
33 datap = [0.2 0.5 0.3]; | |
34 datac = [0 2; 0 0; 2 3.5]; | |
35 datacov = repmat(eye(2), [1 1 3]); | |
36 data1 = data(1:prior(1)*ndata,:); | |
37 data2 = data(prior(1)*ndata+1:(prior(2)+prior(1))*ndata, :); | |
38 data3 = data((prior(1)+prior(2))*ndata +1:ndata, :); | |
39 | |
40 % First cluster has axis aligned variance and centre (2, 3.5) | |
41 data1(:, 1) = data1(:, 1)*0.1 + 2.0; | |
42 data1(:, 2) = data1(:, 2)*0.8 + 3.5; | |
43 datacov(:, :, 3) = [0.1*0.1 0; 0 0.8*0.8]; | |
44 | |
45 % Second cluster has variance axes rotated by 30 degrees and centre (0, 0) | |
46 rotn = [cos(pi/6) -sin(pi/6); sin(pi/6) cos(pi/6)]; | |
47 data2(:,1) = data2(:, 1)*0.2; | |
48 data2 = data2*rotn; | |
49 datacov(:, :, 2) = rotn' * [0.04 0; 0 1] * rotn; | |
50 | |
51 % Third cluster is at (0,2) | |
52 data3(:, 2) = data3(:, 2)*0.1; | |
53 data3 = data3 + repmat([0 2], prior(3)*ndata, 1); | |
54 | |
55 % Put the dataset together again | |
56 data = [data1; data2; data3]; | |
57 | |
58 ndata = 100; % Number of data points. | |
59 noise = 0.2; % Standard deviation of noise distribution. | |
60 x = [0:1/(2*(ndata - 1)):0.5]'; | |
61 randn('state', 1); | |
62 rand('state', 1); | |
63 t = sin(2*pi*x) + noise*randn(ndata, 1); | |
64 | |
65 % Fit three one-dimensional PPCA models | |
66 ncentres = 3; | |
67 ppca_dim = 1; | |
68 | |
69 clc | |
70 disp('This demonstration illustrates the use of a Gaussian mixture model') | |
71 disp('with a probabilistic PCA covariance structure to approximate the') | |
72 disp('unconditional probability density of data in a two-dimensional space.') | |
73 disp('We begin by generating the data from a mixture of three Gaussians and') | |
74 disp('plotting it.') | |
75 disp(' ') | |
76 disp('The first cluster has axis aligned variance and centre (0, 2).') | |
77 disp('The variance parallel to the x-axis is significantly greater') | |
78 disp('than that parallel to the y-axis.') | |
79 disp('The second cluster has variance axes rotated by 30 degrees') | |
80 disp('and centre (0, 0). The third cluster has significant variance') | |
81 disp('parallel to the y-axis and centre (2, 3.5).') | |
82 disp(' ') | |
83 disp('Press any key to continue.') | |
84 pause | |
85 | |
86 fh1 = figure; | |
87 plot(data(:, 1), data(:, 2), 'o') | |
88 set(gca, 'Box', 'on') | |
89 axis equal | |
90 hold on | |
91 | |
92 mix = gmm(2, ncentres, 'ppca', ppca_dim); | |
93 options = foptions; | |
94 options(14) = 10; | |
95 options(1) = -1; % Switch off all warnings | |
96 | |
97 % Just use 10 iterations of k-means in initialisation | |
98 % Initialise the model parameters from the data | |
99 mix = gmminit(mix, data, options); | |
100 disp('The mixture model has three components with 1-dimensional') | |
101 disp('PPCA subspaces. The model parameters after initialisation using') | |
102 disp('the k-means algorithm are as follows') | |
103 disp(' Priors Centres') | |
104 disp([mix.priors' mix.centres]) | |
105 disp(' ') | |
106 disp('Press any key to continue') | |
107 pause | |
108 | |
109 options(1) = 1; % Prints out error values. | |
110 options(14) = 30; % Number of iterations. | |
111 | |
112 disp('We now train the model using the EM algorithm for up to 30 iterations.') | |
113 disp(' ') | |
114 disp('Press any key to continue.') | |
115 pause | |
116 | |
117 [mix, options, errlog] = gmmem(mix, data, options); | |
118 disp('The trained model has priors and centres:') | |
119 disp(' Priors Centres') | |
120 disp([mix.priors' mix.centres]) | |
121 | |
122 % Now plot the result | |
123 for i = 1:ncentres | |
124 % Plot the PC vectors | |
125 v = mix.U(:,:,i); | |
126 start=mix.centres(i,:)-sqrt(mix.lambda(i))*(v'); | |
127 endpt=mix.centres(i,:)+sqrt(mix.lambda(i))*(v'); | |
128 linex = [start(1) endpt(1)]; | |
129 liney = [start(2) endpt(2)]; | |
130 line(linex, liney, 'Color', 'k', 'LineWidth', 3) | |
131 % Plot ellipses of one standard deviation | |
132 theta = 0:0.02:2*pi; | |
133 x = sqrt(mix.lambda(i))*cos(theta); | |
134 y = sqrt(mix.covars(i))*sin(theta); | |
135 % Rotate ellipse axes | |
136 rot_matrix = [v(1) -v(2); v(2) v(1)]; | |
137 ellipse = (rot_matrix*([x; y]))'; | |
138 % Adjust centre | |
139 ellipse = ellipse + ones(length(theta), 1)*mix.centres(i,:); | |
140 plot(ellipse(:,1), ellipse(:,2), 'r-') | |
141 end | |
142 | |
143 disp(' ') | |
144 disp('Press any key to exit') | |
145 pause | |
146 close (fh1); | |
147 clear all; |