wolffd@0
|
1 function L = log_prob_node(CPD, self_ev, pev)
|
wolffd@0
|
2 % LOG_PROB_NODE Compute prod_m log P(x(i,m)| x(pi_i,m), theta_i) for node i (gaussian)
|
wolffd@0
|
3 % L = log_prob_node(CPD, self_ev, pev)
|
wolffd@0
|
4 %
|
wolffd@0
|
5 % self_ev(m) is the evidence on this node in case m.
|
wolffd@0
|
6 % pev(i,m) is the evidence on the i'th parent in case m (if there are any parents).
|
wolffd@0
|
7 % (These may also be cell arrays.)
|
wolffd@0
|
8
|
wolffd@0
|
9 if iscell(self_ev), usecell = 1; else usecell = 0; end
|
wolffd@0
|
10
|
wolffd@0
|
11 use_log = 1;
|
wolffd@0
|
12 ncases = length(self_ev);
|
wolffd@0
|
13 nparents = length(CPD.sizes)-1;
|
wolffd@0
|
14 assert(ncases == size(pev, 2));
|
wolffd@0
|
15
|
wolffd@0
|
16 if ncases == 0
|
wolffd@0
|
17 L = 0;
|
wolffd@0
|
18 return;
|
wolffd@0
|
19 end
|
wolffd@0
|
20
|
wolffd@0
|
21 if length(CPD.dps)==0 % no discrete parents, so we can vectorize
|
wolffd@0
|
22 i = 1;
|
wolffd@0
|
23 if usecell
|
wolffd@0
|
24 Y = cell2num(self_ev);
|
wolffd@0
|
25 else
|
wolffd@0
|
26 Y = self_ev;
|
wolffd@0
|
27 end
|
wolffd@0
|
28 if length(CPD.cps) == 0
|
wolffd@0
|
29 L = gaussian_prob(Y, CPD.mean(:,i), CPD.cov(:,:,i), use_log);
|
wolffd@0
|
30 else
|
wolffd@0
|
31 if usecell
|
wolffd@0
|
32 X = cell2num(pev);
|
wolffd@0
|
33 else
|
wolffd@0
|
34 X = pev;
|
wolffd@0
|
35 end
|
wolffd@0
|
36 L = gaussian_prob(Y, CPD.mean(:,i) + CPD.weights(:,:,i)*X, CPD.cov(:,:,i), use_log);
|
wolffd@0
|
37 end
|
wolffd@0
|
38 else % each case uses a (potentially) different set of parameters
|
wolffd@0
|
39 L = 0;
|
wolffd@0
|
40 for m=1:ncases
|
wolffd@0
|
41 if usecell
|
wolffd@0
|
42 dpvals = cat(1, pev{CPD.dps, m});
|
wolffd@0
|
43 else
|
wolffd@0
|
44 dpvals = pev(CPD.dps, m);
|
wolffd@0
|
45 end
|
wolffd@0
|
46 i = subv2ind(CPD.sizes(CPD.dps), dpvals(:)');
|
wolffd@0
|
47 y = self_ev{m};
|
wolffd@0
|
48 if length(CPD.cps) == 0
|
wolffd@0
|
49 L = L + gaussian_prob(y, CPD.mean(:,i), CPD.cov(:,:,i), use_log);
|
wolffd@0
|
50 else
|
wolffd@0
|
51 if usecell
|
wolffd@0
|
52 x = cat(1, pev{CPD.cps, m});
|
wolffd@0
|
53 else
|
wolffd@0
|
54 x = pev(CPD.cps, m);
|
wolffd@0
|
55 end
|
wolffd@0
|
56 L = L + gaussian_prob(y, CPD.mean(:,i) + CPD.weights(:,:,i)*x, CPD.cov(:,:,i), use_log);
|
wolffd@0
|
57 end
|
wolffd@0
|
58 end
|
wolffd@0
|
59 end
|