wolffd@0
|
1 function score = score_family(j, ps, node_type, scoring_fn, ns, discrete, data, args)
|
wolffd@0
|
2 % SCORE_FAMILY_COMPLETE Compute the score of a node and its parents given completely observed data
|
wolffd@0
|
3 % score = score_family(j, ps, node_type, scoring_fn, ns, discrete, data, args)
|
wolffd@0
|
4 %
|
wolffd@0
|
5 % data(i,m) is the value of node i in case m (can be a cell array)
|
wolffd@0
|
6 % args is a cell array containing optional arguments passed to the constructor,
|
wolffd@0
|
7 % or is [] if none
|
wolffd@0
|
8 %
|
wolffd@0
|
9 % We create a whole Bayes net which only connects parents to node,
|
wolffd@0
|
10 % where node has a CPD of the specified type (with default parameters).
|
wolffd@0
|
11 % We then evaluate its score ('bic' or 'bayesian')
|
wolffd@0
|
12
|
wolffd@0
|
13 % We should use a cache to avoid unnecessary computation.
|
wolffd@0
|
14 % In particular, log_marginal_prob_node for tabular CPDs calls gammaln
|
wolffd@0
|
15 % and compute_counts, both of which are slow.
|
wolffd@0
|
16
|
wolffd@0
|
17 [n ncases] = size(data);
|
wolffd@0
|
18 dag = zeros(n,n);
|
wolffd@0
|
19 % SML added to sort ps b/c mk_bnet, learn_params use sorted ps to make
|
wolffd@0
|
20 % CPTs
|
wolffd@0
|
21 % Kevin had: if ~isempty(ps), dag(ps, j) = 1; end
|
wolffd@0
|
22 if ~isempty(ps), dag(ps, j) = 1;, ps = sort(ps);, end
|
wolffd@0
|
23
|
wolffd@0
|
24 bnet = mk_bnet(dag, ns, 'discrete', discrete);
|
wolffd@0
|
25 %bnet.CPD{j} = xxx_CPD(bnet, j);
|
wolffd@0
|
26 %eval(sprintf('bnet.CPD{j} = %s_CPD(bnet, j);', node_type));
|
wolffd@0
|
27 fname = sprintf('%s_CPD', node_type);
|
wolffd@0
|
28 %fprintf('score CPD %d\n', j);
|
wolffd@0
|
29 if isempty(args)
|
wolffd@0
|
30 bnet.CPD{j} = feval(fname, bnet, j);
|
wolffd@0
|
31 else
|
wolffd@0
|
32 bnet.CPD{j} = feval(fname, bnet, j, args{:});
|
wolffd@0
|
33 end
|
wolffd@0
|
34 switch scoring_fn
|
wolffd@0
|
35 case 'bic',
|
wolffd@0
|
36 fam = [ps j];
|
wolffd@0
|
37 %score = BIC_score_CPD(bnet.CPD{j}, fam, data, ns, bnet.cnodes);
|
wolffd@0
|
38 %bnet.CPD{j} = learn_params(bnet.CPD{j}, fam, data, ns, bnet.cnodes);
|
wolffd@0
|
39
|
wolffd@0
|
40 % SML 03/16/04 had to special case gaussian b/c generic_CPD/learn_params
|
wolffd@0
|
41 % no longer supported because of simple interface to learn_params
|
wolffd@0
|
42 % introduced by KPM for tabular nodes below:
|
wolffd@0
|
43 % KPM 9 June 04 - tabular nodes have changed back!
|
wolffd@0
|
44 if 1 % (isempty(find(j==discrete)))
|
wolffd@0
|
45 bnet.CPD{j} = learn_params(bnet.CPD{j}, fam, data, ns, bnet.cnodes);
|
wolffd@0
|
46 else
|
wolffd@0
|
47 bnet.CPD{j} = learn_params(bnet.CPD{j}, data(fam, :));
|
wolffd@0
|
48 end
|
wolffd@0
|
49 L = log_prob_node(bnet.CPD{j}, data(j,:), data(ps,:));
|
wolffd@0
|
50 S = struct(bnet.CPD{j}); % violate object privacy
|
wolffd@0
|
51 score = L - 0.5*S.nparams*log(ncases);
|
wolffd@0
|
52 case 'bayesian',
|
wolffd@0
|
53 %score = bayesian_score_CPD(bnet.CPD{j}, data(fam, :));
|
wolffd@0
|
54 score = log_marg_prob_node(bnet.CPD{j}, data(j,:), data(ps,:));
|
wolffd@0
|
55 otherwise,
|
wolffd@0
|
56 error(['unrecognized scoring fn ' scoring_fn]);
|
wolffd@0
|
57 end
|