|
wolffd@0
|
1 function [bnet, vals] = mk_minimal_qmr_bnet(G, inhibit, leak, prior, pos, neg, pos_only)
|
|
wolffd@0
|
2 % MK_MINIMAL_QMR_BNET Make a QMR model which only contains the observed findings
|
|
wolffd@0
|
3 % [bnet, vals] = mk_minimal_qmr_bnet(G, inhibit, prior, leak, pos, neg)
|
|
wolffd@0
|
4 %
|
|
wolffd@0
|
5 % Input:
|
|
wolffd@0
|
6 % G(i,j) = 1 iff there is an arc from disease i to finding j
|
|
wolffd@0
|
7 % inhibit(i,j) = inhibition probability on i->j arc
|
|
wolffd@0
|
8 % leak(j) = inhibition prob. on leak->j arc
|
|
wolffd@0
|
9 % prior(i) = prob. disease i is on
|
|
wolffd@0
|
10 % pos = list of leaves that have positive observations
|
|
wolffd@0
|
11 % neg = list of leaves that have negative observations
|
|
wolffd@0
|
12 % pos_only = 1 means only include positively observed leaves in the model - the negative
|
|
wolffd@0
|
13 % ones are absorbed into the prior terms
|
|
wolffd@0
|
14 %
|
|
wolffd@0
|
15 % Output:
|
|
wolffd@0
|
16 % bnet
|
|
wolffd@0
|
17 % vals is their value
|
|
wolffd@0
|
18
|
|
wolffd@0
|
19 if pos_only
|
|
wolffd@0
|
20 obs = pos;
|
|
wolffd@0
|
21 else
|
|
wolffd@0
|
22 obs = myunion(pos, neg);
|
|
wolffd@0
|
23 end
|
|
wolffd@0
|
24 Nfindings = length(obs);
|
|
wolffd@0
|
25 [Ndiseases maxNfindings] = size(inhibit);
|
|
wolffd@0
|
26 N = Ndiseases + Nfindings;
|
|
wolffd@0
|
27 finding_node = Ndiseases+1:N;
|
|
wolffd@0
|
28
|
|
wolffd@0
|
29 % j = finding_node(i) means the i'th finding node is the j'th node in the bnet
|
|
wolffd@0
|
30 % k = obs(i) means the i'th observed (positive) finding is the k'th finding overall
|
|
wolffd@0
|
31 % If all findings are observed, and posonly = 0, we have i = obs(i) for all i.
|
|
wolffd@0
|
32
|
|
wolffd@0
|
33 %dag = sparse(N, N);
|
|
wolffd@0
|
34 dag = zeros(N, N);
|
|
wolffd@0
|
35 dag(1:Ndiseases, Ndiseases+1:N) = G(:,obs);
|
|
wolffd@0
|
36
|
|
wolffd@0
|
37 ns = 2*ones(1,N);
|
|
wolffd@0
|
38 bnet = mk_bnet(dag, ns, 'observed', finding_node);
|
|
wolffd@0
|
39
|
|
wolffd@0
|
40 CPT = cell(1, Ndiseases);
|
|
wolffd@0
|
41 for d=1:Ndiseases
|
|
wolffd@0
|
42 CPT{d} = [1-prior(d) prior(d)];
|
|
wolffd@0
|
43 end
|
|
wolffd@0
|
44
|
|
wolffd@0
|
45 if pos_only
|
|
wolffd@0
|
46 % Fold in the negative evidence into the prior
|
|
wolffd@0
|
47 for i=1:length(neg)
|
|
wolffd@0
|
48 n = neg(i);
|
|
wolffd@0
|
49 ps = parents(G,n);
|
|
wolffd@0
|
50 for pi=1:length(ps)
|
|
wolffd@0
|
51 p = ps(pi);
|
|
wolffd@0
|
52 q = inhibit(p,n);
|
|
wolffd@0
|
53 CPT{p} = CPT{p} .* [1 q];
|
|
wolffd@0
|
54 end
|
|
wolffd@0
|
55 % Arbitrarily attach the leak term to the first parent
|
|
wolffd@0
|
56 p = ps(1);
|
|
wolffd@0
|
57 q = leak(n);
|
|
wolffd@0
|
58 CPT{p} = CPT{p} .* [q q];
|
|
wolffd@0
|
59 end
|
|
wolffd@0
|
60 end
|
|
wolffd@0
|
61
|
|
wolffd@0
|
62 for d=1:Ndiseases
|
|
wolffd@0
|
63 bnet.CPD{d} = tabular_CPD(bnet, d, CPT{d}');
|
|
wolffd@0
|
64 end
|
|
wolffd@0
|
65
|
|
wolffd@0
|
66 for i=1:Nfindings
|
|
wolffd@0
|
67 fnode = finding_node(i);
|
|
wolffd@0
|
68 fid = obs(i);
|
|
wolffd@0
|
69 ps = parents(G, fid);
|
|
wolffd@0
|
70 bnet.CPD{fnode} = noisyor_CPD(bnet, fnode, leak(fid), inhibit(ps, fid));
|
|
wolffd@0
|
71 end
|
|
wolffd@0
|
72
|
|
wolffd@0
|
73 obs_nodes = finding_node;
|
|
wolffd@0
|
74 vals = sparse(1, maxNfindings);
|
|
wolffd@0
|
75 vals(pos) = 2;
|
|
wolffd@0
|
76 vals(neg) = 1;
|
|
wolffd@0
|
77 vals = full(vals(obs));
|
|
wolffd@0
|
78
|
|
wolffd@0
|
79
|
|
wolffd@0
|
80
|
|
wolffd@0
|
81
|
|
wolffd@0
|
82
|
