comparison toolboxes/FullBNT-1.0.7/graph/triangulate.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 [G, cliques, fill_ins] = triangulate(G, order)
2 % TRIANGULATE Ensure G is triangulated (chordal), i.e., every cycle of length > 3 has a chord.
3 % [G, cliques, fill_ins, cliques_containing_node] = triangulate(G, order)
4 %
5 % cliques{i} is the i'th maximal complete subgraph of the triangulated graph.
6 % fill_ins(i,j) = 1 iff we add a fill-in arc between i and j.
7 %
8 % To find the maximal cliques, we save each induced cluster (created by adding connecting
9 % neighbors) that is not a subset of any previously saved cluster. (A cluster is a complete,
10 % but not necessarily maximal, set of nodes.)
11
12 MG = G;
13 n = length(G);
14 eliminated = zeros(1,n);
15 cliques = {};
16 for i=1:n
17 u = order(i);
18 U = find(~eliminated); % uneliminated
19 nodes = myintersect(neighbors(G,u), U); % look up neighbors in the partially filled-in graph
20 nodes = myunion(nodes, u); % the clique will always contain at least u
21 G(nodes,nodes) = 1; % make them all connected to each other
22 G = setdiag(G,0);
23 eliminated(u) = 1;
24
25 exclude = 0;
26 for c=1:length(cliques)
27 if mysubset(nodes,cliques{c}) % not maximal
28 exclude = 1;
29 break;
30 end
31 end
32 if ~exclude
33 cnum = length(cliques)+1;
34 cliques{cnum} = nodes;
35 end
36 end
37
38 fill_ins = sparse(triu(max(0, G - MG), 1));
39
40 %assert(check_triangulated(G)); % takes 72% of the time!
41
42