comparison toolboxes/FullBNT-1.0.7/graph/cliques_to_jtree.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 function [jtree, root, B, w] = cliques_to_jtree(cliques, ns)
2 % MK_JTREE Make an optimal junction tree.
3 % [jtree, root, B, w] = mk_jtree(cliques, ns)
4 %
5 % A junction tree is a tree that satisfies the jtree property, which says:
6 % for each pair of cliques U,V with intersection S, all cliques on the path between U and V
7 % contain S. (This ensures that local propagation leads to global consistency.)
8 %
9 % We can create a junction tree by computing the maximal spanning tree of the junction graph.
10 % (The junction graph connects all cliques, and the weight of an edge (i,j) is
11 % |C(i) intersect C(j)|, where C(i) is the i'th clique.)
12 %
13 % The best jtree is the maximal spanning tree which minimizes the sum of the costs on each edge,
14 % where cost(i,j) = w(C(i)) + w(C(j)), and w(C) is the weight of clique C,
15 % which is the total number of values C can take on.
16 %
17 % For details, see
18 % - Jensen and Jensen, "Optimal Junction Trees", UAI 94.
19 %
20 % Input:
21 % cliques{i} = nodes in clique i
22 % ns(i) = number of values node i can take on
23 % Output:
24 % jtree(i,j) = 1 iff cliques i and j aer connected
25 % root = the clique that should be used as root
26 % B(i,j) = 1 iff node j occurs in clique i
27 % w(i) = weight of clique i
28
29
30
31 num_cliques = length(cliques);
32 w = zeros(num_cliques, 1);
33 B = sparse(num_cliques, 1);
34 for i=1:num_cliques
35 B(i, cliques{i}) = 1;
36 w(i) = prod(ns(cliques{i}));
37 end
38
39
40 % C1(i,j) = length(intersect(cliques{i}, cliques{j}));
41 % The length of the intersection of two sets is the dot product of their bit vector representation.
42 C1 = B*B';
43 C1 = setdiag(C1, 0);
44
45 % C2(i,j) = w(i) + w(j)
46 num_cliques = length(w);
47 W = repmat(w, 1, num_cliques);
48 C2 = W + W';
49 C2 = setdiag(C2, 0);
50
51 jtree = sparse(minimum_spanning_tree(-C1, C2)); % Using -C1 gives *maximum* spanning tree
52
53 % The root is arbitrary, but since the first pass is towards the root,
54 % we would like this to correspond to going forward in time in a DBN.
55 root = num_cliques;
56
57