Mercurial > hg > camir-aes2014

annotate toolboxes/FullBNT-1.0.7/bnt/examples/dynamic/jtree_clq_test.m @ 0:e9a9cd732c1e tip

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
first hg version after svn
author wolffd
date Tue, 10 Feb 2015 15:05:51 +0000
parents
children
rev   line source
wolffd@0 1 % Construct various DBNs and examine their clique structure.
wolffd@0 2 % This was used to generate various figures in chap 3-4 of my thesis.
wolffd@0 3
wolffd@0 4 % Examine the cliques in the unrolled mildew net
wolffd@0 5
wolffd@0 6 %dbn = mk_mildew_dbn;
wolffd@0 7 dbn = mk_chmm(4);
wolffd@0 8 ss = dbn.nnodes_per_slice;
wolffd@0 9 T = 7;
wolffd@0 10 N = ss*T;
wolffd@0 11 bnet = dbn_to_bnet(dbn, T);
wolffd@0 12
wolffd@0 13 constrained = 0;
wolffd@0 14 if constrained
wolffd@0 15 stages = num2cell(unroll_set(1:ss, ss, T), 1);
wolffd@0 16 else
wolffd@0 17 stages = { 1:N; };
wolffd@0 18 end
wolffd@0 19 clusters = {};
wolffd@0 20 %[jtree, root, cliques, B, w, elim_order, moral_edges, fill_in_edges] = ...
wolffd@0 21 % dag_to_jtree(bnet, bnet.observed, stages, clusters);
wolffd@0 22 [jtree, root, cliques] = graph_to_jtree(moralize(bnet.dag), ones(1,N), stages, clusters);
wolffd@0 23
wolffd@0 24 flip=1;
wolffd@0 25 clf;[dummyx, dummyy, h] = draw_dbn(dbn.intra, dbn.inter, flip, T, -1);
wolffd@0 26 dir = '/home/eecs/murphyk/WP/Thesis/Figures/Inf/MildewUnrolled';
wolffd@0 27 mk_ps_from_clqs(dbn, T, cliques, [])
wolffd@0 28 %mk_collage_from_clqs(dir, cliques)
wolffd@0 29
wolffd@0 30
wolffd@0 31 % Examine the cliques in the cascade DBN
wolffd@0 32
wolffd@0 33 % A-A
wolffd@0 34 % \
wolffd@0 35 % B B
wolffd@0 36 % \
wolffd@0 37 % C C
wolffd@0 38 % \
wolffd@0 39 % D D
wolffd@0 40 ss = 4;
wolffd@0 41 intra = zeros(ss);
wolffd@0 42 inter = zeros(ss);
wolffd@0 43 inter(1, [1 2])=1;
wolffd@0 44 for i=2:ss-1
wolffd@0 45 inter(i,i+1)=1;
wolffd@0 46 end
wolffd@0 47
wolffd@0 48
wolffd@0 49 % 2 coupled HMMs 1,3 and 2,4
wolffd@0 50 ss = 4;
wolffd@0 51 intra = zeros(ss);
wolffd@0 52 inter = zeros(ss); % no persistent edges
wolffd@0 53 %inter = diag(ones(ss,1)); % persitence edges
wolffd@0 54 inter(1,3)=1; inter(3,1)=1;
wolffd@0 55 inter(2,4)=1; inter(4,2)=1;
wolffd@0 56
wolffd@0 57 %bnet = mk_fhmm(3);
wolffd@0 58 bnet = mk_chmm(4);
wolffd@0 59 intra = bnet.intra;
wolffd@0 60 inter = bnet.inter;
wolffd@0 61
wolffd@0 62 clqs = compute_minimal_interface(intra, inter);
wolffd@0 63 celldisp(clqs)
wolffd@0 64
wolffd@0 65
wolffd@0 66
wolffd@0 67
wolffd@0 68 % A A
wolffd@0 69 % \
wolffd@0 70 % B B
wolffd@0 71 % \
wolffd@0 72 % C C
wolffd@0 73 % \
wolffd@0 74 % D-D
wolffd@0 75 ss = 4;
wolffd@0 76 intra = zeros(ss);
wolffd@0 77 inter = zeros(ss);
wolffd@0 78 for i=1:ss-1
wolffd@0 79 inter(i,i+1)=1;
wolffd@0 80 end
wolffd@0 81 inter(4,4)=1;
wolffd@0 82
wolffd@0 83
wolffd@0 84
wolffd@0 85 ns = 2*ones(1,ss);
wolffd@0 86 dbn = mk_dbn(intra, inter, ns);
wolffd@0 87 for i=2*ss
wolffd@0 88 dbn.CPD{i} = tabular_CPD(bnet, i);
wolffd@0 89 end
wolffd@0 90
wolffd@0 91 T = 4;
wolffd@0 92 N = ss*T;
wolffd@0 93 bnet = dbn_to_bnet(dbn, T);
wolffd@0 94
wolffd@0 95 constrained = 1;
wolffd@0 96 if constrained
wolffd@0 97 % elim first 3 slices first in any order
wolffd@0 98 stages = {1:12, 13:16};
wolffd@0 99 %stages = num2cell(unroll_set(1:ss, ss, T), 1);
wolffd@0 100 else
wolffd@0 101 stages = { 1:N; };
wolffd@0 102 end
wolffd@0 103 clusters = {};
wolffd@0 104 %[jtree, root, cliques, B, w, elim_order, moral_edges, fill_in_edges] = ...
wolffd@0 105 % dag_to_jtree(bnet, bnet.observed, stages, clusters);
wolffd@0 106 [jtree, root, cliques] = graph_to_jtree(moralize(bnet.dag), ones(1,N), stages, clusters);
wolffd@0 107
wolffd@0 108
wolffd@0 109
wolffd@0 110
wolffd@0 111
wolffd@0 112 % Examine the cliques in the 1.5 slice DBN
wolffd@0 113
wolffd@0 114 %dbn = mk_mildew_dbn;
wolffd@0 115 dbn = mk_water_dbn;
wolffd@0 116 %dbn = mk_bat_dbn;
wolffd@0 117 ss = dbn.nnodes_per_slice;
wolffd@0 118 int = compute_fwd_interface(dbn);
wolffd@0 119 bnet15 = mk_slice_and_half_dbn(dbn, int);
wolffd@0 120 N = length(bnet15.dag);
wolffd@0 121 stages = {1:N};
wolffd@0 122
wolffd@0 123 % bat
wolffd@0 124 %cl1 = [16 17 19 7 14];
wolffd@0 125 %cl2 = [27 25 21 23 20];
wolffd@0 126 %clusters = {cl1, cl2, cl1+ss, cl2+ss};
wolffd@0 127
wolffd@0 128 % water
wolffd@0 129 %cl1 = 1:2; cl2 = 3:6; cl3 = 7:8;
wolffd@0 130 %clusters = {cl1, cl2, cl3, cl1+ss, cl2+ss, cl3+ss};
wolffd@0 131
wolffd@0 132 %clusters = {};
wolffd@0 133 clusters = {int, int+ss};
wolffd@0 134 %[jtree, root, cliques, B, w, elim_order, moral_edges, fill_in_edges] = ...
wolffd@0 135 % dag_to_jtree(bnet15, bnet.observed, stages, clusters);
wolffd@0 136 [jtree, root, cliques] = graph_to_jtree(moralize(bnet15.dag), ones(1,N), stages, clusters);
wolffd@0 137
wolffd@0 138 clq_len = [];
wolffd@0 139 for c=1:length(cliques)
wolffd@0 140 clq_len(c) = length(cliques{c});
wolffd@0 141 end
wolffd@0 142 hist(clq_len, 1:max(clq_len));
wolffd@0 143 h=hist(clq_len, 1:max(clq_len));
wolffd@0 144 axis([1 max(clq_len)+1 0 max(h)+1])
wolffd@0 145 xlabel('clique size','fontsize',16)
wolffd@0 146 ylabel('number','fontsize',16)
wolffd@0 147
wolffd@0 148
wolffd@0 149
wolffd@0 150