Mercurial > hg > camir-aes2014
comparison toolboxes/FullBNT-1.0.7/bnt/examples/static/mfa1.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 % Factor analysis | |
2 % Z -> X, Z in R^k, X in R^D, k << D (high dimensional observations explained by small source) | |
3 % Z ~ N(0,I), X|Z ~ N(L z, Psi), where Psi is diagonal. | |
4 % | |
5 % Mixtures of FA | |
6 % Now X|Z,W=i ~ N(mu(i) + L(i) Z, Psi(i)) | |
7 % | |
8 % We compare to Zoubin Ghahramani's code. | |
9 | |
10 randn('state', 0); | |
11 max_iter = 3; | |
12 M = 2; | |
13 k = 3; | |
14 D = 5; | |
15 | |
16 n = 5; | |
17 X1 = randn(n, D); | |
18 X2 = randn(n, D) + 2; % move the mean to (2,2,2...) | |
19 X = [X1; X2]; | |
20 N = size(X, 1); | |
21 | |
22 % initialise as in mfa | |
23 tiny=exp(-700); | |
24 mX = mean(X); | |
25 cX=cov(X); | |
26 scale=det(cX)^(1/D); | |
27 randn('state',0); % must reset seed here so initial params are identical to mfa | |
28 L0=randn(D*M,k)*sqrt(scale/k); | |
29 W0 = permute(reshape(L0, [D M k]), [1 3 2]); % use D,K,M | |
30 Psi0=diag(cX)+tiny; | |
31 Pi0=ones(M,1)/M; | |
32 Mu0=randn(M,D)*sqrtm(cX)+ones(M,1)*mX; | |
33 | |
34 [Lh1, Ph1, Mu1, Pi1, LL1] = mfa(X,M,k,max_iter); | |
35 Lh1 = permute(reshape(Lh1, [D M k]), [1 3 2]); % use D,K,M | |
36 | |
37 | |
38 ns = [M k D]; | |
39 dag = zeros(3); | |
40 dag(1,3) = 1; | |
41 dag(2,3) = 1; | |
42 dnodes = 1; | |
43 onodes = 3; | |
44 | |
45 bnet = mk_bnet(dag, ns, 'discrete', dnodes, 'observed', onodes); | |
46 bnet.CPD{1} = tabular_CPD(bnet, 1, Pi0); | |
47 | |
48 %bnet.CPD{2} = gaussian_CPD(bnet, 2, zeros(k, 1), eye(k), [], 'diag', 'untied', 'clamp_mean', 'clamp_cov'); | |
49 | |
50 bnet.CPD{2} = gaussian_CPD(bnet, 2, 'mean', zeros(k, 1), 'cov', eye(k), 'cov_type', 'diag', ... | |
51 'cov_prior_weight', 0, 'clamp_mean', 1, 'clamp_cov', 1); | |
52 | |
53 %bnet.CPD{3} = gaussian_CPD(bnet, 3, Mu0', repmat(diag(Psi0), [1 1 M]), W0, 'diag', 'tied'); | |
54 | |
55 bnet.CPD{3} = gaussian_CPD(bnet, 3, 'mean', Mu0', 'cov', repmat(diag(Psi0), [1 1 M]), ... | |
56 'weights', W0, 'cov_type', 'diag', 'cov_prior_weight', 0, 'tied_cov', 1); | |
57 | |
58 engine = jtree_inf_engine(bnet); | |
59 evidence = cell(3, N); | |
60 evidence(3,:) = num2cell(X', 1); | |
61 | |
62 [bnet2, LL2, engine2] = learn_params_em(engine, evidence, max_iter); | |
63 | |
64 s = struct(bnet2.CPD{1}); | |
65 Pi2 = s.CPT(:); | |
66 s = struct(bnet2.CPD{3}); | |
67 Mu2 = s.mean; | |
68 W2 = s.weights; | |
69 Sigma2 = s.cov; | |
70 | |
71 | |
72 % Compare to Zoubin's code | |
73 assert(approxeq(LL1,LL2)); | |
74 for i=1:M | |
75 assert(approxeq(W2(:,:,i), Lh1(:,:,i))); | |
76 assert(approxeq(Sigma2(:,:,i), diag(Ph1))); | |
77 assert(approxeq(Mu2(:,i), Mu1(i,:))); | |
78 assert(approxeq(Pi2(:), Pi1(:))); | |
79 end | |
80 |