|
wolffd@0
|
1 function bnet = mk_cancer_bnet(CPD_type, p)
|
|
wolffd@0
|
2 % MK_CANCER_BNET Make the 'Cancer' Bayes net.
|
|
wolffd@0
|
3 %
|
|
wolffd@0
|
4 % BNET = MK_CANCER_BNET uses the noisy-or parameters specified in Fig 4a of the UAI98 paper by
|
|
wolffd@0
|
5 % Friedman, Murphy and Russell, "Learning the Structure of DPNs", p145.
|
|
wolffd@0
|
6 %
|
|
wolffd@0
|
7 % BNET = MK_CANCER_BNET('noisyor', p) makes each CPD a noisy-or, with probability p of
|
|
wolffd@0
|
8 % suppression for each parent; leaks are turned off.
|
|
wolffd@0
|
9 %
|
|
wolffd@0
|
10 % BNET = MK_CANCER_BNET('cpt', p) uses random CPT parameters drawn from a Dirichlet(p,p,...)
|
|
wolffd@0
|
11 % distribution. If p << 1, this is near deterministic; if p >> 1, this is near 1/k.
|
|
wolffd@0
|
12 % p defaults to 1.0 (uniform distribution).
|
|
wolffd@0
|
13 %
|
|
wolffd@0
|
14 % BNET = MK_CANCER_BNET('bool') makes each CPT a random boolean function.
|
|
wolffd@0
|
15 %
|
|
wolffd@0
|
16 % In all cases, the root is set to a uniform distribution.
|
|
wolffd@0
|
17
|
|
wolffd@0
|
18 if nargin == 0
|
|
wolffd@0
|
19 rnd = 0;
|
|
wolffd@0
|
20 else
|
|
wolffd@0
|
21 rnd = 1;
|
|
wolffd@0
|
22 end
|
|
wolffd@0
|
23
|
|
wolffd@0
|
24 n = 5;
|
|
wolffd@0
|
25 dag = zeros(n);
|
|
wolffd@0
|
26 dag(1,[2 3]) = 1;
|
|
wolffd@0
|
27 dag(2,4) = 1;
|
|
wolffd@0
|
28 dag(3,4) = 1;
|
|
wolffd@0
|
29 dag(4,5) = 1;
|
|
wolffd@0
|
30
|
|
wolffd@0
|
31 ns = 2*ones(1,n);
|
|
wolffd@0
|
32 bnet = mk_bnet(dag, ns);
|
|
wolffd@0
|
33
|
|
wolffd@0
|
34 if ~rnd
|
|
wolffd@0
|
35 bnet.CPD{1} = tabular_CPD(bnet, 1, [0.5 0.5]);
|
|
wolffd@0
|
36 bnet.CPD{2} = noisyor_CPD(bnet, 2, 1.0, 1-0.9);
|
|
wolffd@0
|
37 bnet.CPD{3} = noisyor_CPD(bnet, 3, 1.0, 1-0.2);
|
|
wolffd@0
|
38 bnet.CPD{4} = noisyor_CPD(bnet, 4, 1.0, 1-[0.7 0.6]);
|
|
wolffd@0
|
39 bnet.CPD{5} = noisyor_CPD(bnet, 5, 1.0, 1-0.5);
|
|
wolffd@0
|
40 else
|
|
wolffd@0
|
41 switch CPD_type
|
|
wolffd@0
|
42 case 'noisyor',
|
|
wolffd@0
|
43 for i=1:n
|
|
wolffd@0
|
44 ps = parents(dag, i);
|
|
wolffd@0
|
45 bnet.CPD{i} = noisyor_CPD(bnet, i, 1.0, p*ones(1,length(ps)));
|
|
wolffd@0
|
46 end
|
|
wolffd@0
|
47 case 'bool',
|
|
wolffd@0
|
48 for i=1:n
|
|
wolffd@0
|
49 bnet.CPD{i} = boolean_CPD(bnet, i, 'rnd');
|
|
wolffd@0
|
50 end
|
|
wolffd@0
|
51 case 'cpt',
|
|
wolffd@0
|
52 for i=1:n
|
|
wolffd@0
|
53 bnet.CPD{i} = tabular_CPD(bnet, i, p);
|
|
wolffd@0
|
54 end
|
|
wolffd@0
|
55 otherwise
|
|
wolffd@0
|
56 error(['bad CPD type ' CPD_type]);
|
|
wolffd@0
|
57 end
|
|
wolffd@0
|
58 end
|
|
wolffd@0
|
59
|
|
wolffd@0
|
60
|
|
wolffd@0
|
61
|
