wolffd@0
|
1 function L = log_marg_prob_node(CPD, self_ev, pev, usecell)
|
wolffd@0
|
2 % LOG_MARG_PROB_NODE Compute sum_m log P(x(i,m)| x(pi_i,m)) for node i (tabular)
|
wolffd@0
|
3 % L = log_marg_prob_node(CPD, self_ev, pev)
|
wolffd@0
|
4 %
|
wolffd@0
|
5 % This differs from log_prob_node because we integrate out the parameters.
|
wolffd@0
|
6 % self_ev(m) is the evidence on this node in case m.
|
wolffd@0
|
7 % pev(i,m) is the evidence on the i'th parent in case m (if there are any parents).
|
wolffd@0
|
8 % (These may also be cell arrays.)
|
wolffd@0
|
9
|
wolffd@0
|
10 ncases = length(self_ev);
|
wolffd@0
|
11 sz = CPD.sizes;
|
wolffd@0
|
12 nparents = length(sz)-1;
|
wolffd@0
|
13 assert(ncases == size(pev, 2));
|
wolffd@0
|
14
|
wolffd@0
|
15 if nargin < 4
|
wolffd@0
|
16 %usecell = 0;
|
wolffd@0
|
17 if iscell(self_ev)
|
wolffd@0
|
18 usecell = 1;
|
wolffd@0
|
19 else
|
wolffd@0
|
20 usecell = 0;
|
wolffd@0
|
21 end
|
wolffd@0
|
22 end
|
wolffd@0
|
23
|
wolffd@0
|
24
|
wolffd@0
|
25 if ncases==0
|
wolffd@0
|
26 L = 0;
|
wolffd@0
|
27 return;
|
wolffd@0
|
28 elseif ncases==1 % speedup the sequential learning case
|
wolffd@0
|
29 CPT = CPD.CPT;
|
wolffd@0
|
30 % We assume the CPTs are already set to the mean of the posterior (due to bayes_update_params)
|
wolffd@0
|
31 if usecell
|
wolffd@0
|
32 x = cat(1, pev{:})';
|
wolffd@0
|
33 y = self_ev{1};
|
wolffd@0
|
34 else
|
wolffd@0
|
35 %x = pev(:)';
|
wolffd@0
|
36 x = pev;
|
wolffd@0
|
37 y = self_ev;
|
wolffd@0
|
38 end
|
wolffd@0
|
39 switch nparents
|
wolffd@0
|
40 case 0, p = CPT(y);
|
wolffd@0
|
41 case 1, p = CPT(x(1), y);
|
wolffd@0
|
42 case 2, p = CPT(x(1), x(2), y);
|
wolffd@0
|
43 case 3, p = CPT(x(1), x(2), x(3), y);
|
wolffd@0
|
44 otherwise,
|
wolffd@0
|
45 ind = subv2ind(sz, [x y]);
|
wolffd@0
|
46 p = CPT(ind);
|
wolffd@0
|
47 end
|
wolffd@0
|
48 L = log(p);
|
wolffd@0
|
49 else
|
wolffd@0
|
50 % We ignore the CPTs here and assume the prior has not been changed
|
wolffd@0
|
51
|
wolffd@0
|
52 % We arrange the data as in the following example.
|
wolffd@0
|
53 % Let there be 2 parents and 3 cases. Let p(i,m) be parent i in case m,
|
wolffd@0
|
54 % and y(m) be the child in case m. Then we create the data matrix
|
wolffd@0
|
55 %
|
wolffd@0
|
56 % p(1,1) p(1,2) p(1,3)
|
wolffd@0
|
57 % p(2,1) p(2,2) p(2,3)
|
wolffd@0
|
58 % y(1) y(2) y(3)
|
wolffd@0
|
59 if usecell
|
wolffd@0
|
60 data = [cell2num(pev); cell2num(self_ev)];
|
wolffd@0
|
61 else
|
wolffd@0
|
62 data = [pev; self_ev];
|
wolffd@0
|
63 end
|
wolffd@0
|
64 %S = struct(CPD); fprintf('log marg prob node %d, ps\n', S.self); disp(S.parents)
|
wolffd@0
|
65 counts = compute_counts(data, sz);
|
wolffd@0
|
66 L = dirichlet_score_family(counts, CPD.dirichlet);
|
wolffd@0
|
67 end
|
wolffd@0
|
68
|
wolffd@0
|
69
|