Mercurial > hg > camir-aes2014
comparison toolboxes/FullBNT-1.0.7/bnt/learning/score_family.m @ 0:e9a9cd732c1e tip
first hg version after svn
author | wolffd |
---|---|
date | Tue, 10 Feb 2015 15:05:51 +0000 |
parents | |
children |
comparison
equal
deleted
inserted
replaced
-1:000000000000 | 0:e9a9cd732c1e |
---|---|
1 function score = score_family(j, ps, node_type, scoring_fn, ns, discrete, data, args) | |
2 % SCORE_FAMILY_COMPLETE Compute the score of a node and its parents given completely observed data | |
3 % score = score_family(j, ps, node_type, scoring_fn, ns, discrete, data, args) | |
4 % | |
5 % data(i,m) is the value of node i in case m (can be a cell array) | |
6 % args is a cell array containing optional arguments passed to the constructor, | |
7 % or is [] if none | |
8 % | |
9 % We create a whole Bayes net which only connects parents to node, | |
10 % where node has a CPD of the specified type (with default parameters). | |
11 % We then evaluate its score ('bic' or 'bayesian') | |
12 | |
13 % We should use a cache to avoid unnecessary computation. | |
14 % In particular, log_marginal_prob_node for tabular CPDs calls gammaln | |
15 % and compute_counts, both of which are slow. | |
16 | |
17 [n ncases] = size(data); | |
18 dag = zeros(n,n); | |
19 % SML added to sort ps b/c mk_bnet, learn_params use sorted ps to make | |
20 % CPTs | |
21 % Kevin had: if ~isempty(ps), dag(ps, j) = 1; end | |
22 if ~isempty(ps), dag(ps, j) = 1;, ps = sort(ps);, end | |
23 | |
24 bnet = mk_bnet(dag, ns, 'discrete', discrete); | |
25 %bnet.CPD{j} = xxx_CPD(bnet, j); | |
26 %eval(sprintf('bnet.CPD{j} = %s_CPD(bnet, j);', node_type)); | |
27 fname = sprintf('%s_CPD', node_type); | |
28 %fprintf('score CPD %d\n', j); | |
29 if isempty(args) | |
30 bnet.CPD{j} = feval(fname, bnet, j); | |
31 else | |
32 bnet.CPD{j} = feval(fname, bnet, j, args{:}); | |
33 end | |
34 switch scoring_fn | |
35 case 'bic', | |
36 fam = [ps j]; | |
37 %score = BIC_score_CPD(bnet.CPD{j}, fam, data, ns, bnet.cnodes); | |
38 %bnet.CPD{j} = learn_params(bnet.CPD{j}, fam, data, ns, bnet.cnodes); | |
39 | |
40 % SML 03/16/04 had to special case gaussian b/c generic_CPD/learn_params | |
41 % no longer supported because of simple interface to learn_params | |
42 % introduced by KPM for tabular nodes below: | |
43 % KPM 9 June 04 - tabular nodes have changed back! | |
44 if 1 % (isempty(find(j==discrete))) | |
45 bnet.CPD{j} = learn_params(bnet.CPD{j}, fam, data, ns, bnet.cnodes); | |
46 else | |
47 bnet.CPD{j} = learn_params(bnet.CPD{j}, data(fam, :)); | |
48 end | |
49 L = log_prob_node(bnet.CPD{j}, data(j,:), data(ps,:)); | |
50 S = struct(bnet.CPD{j}); % violate object privacy | |
51 score = L - 0.5*S.nparams*log(ncases); | |
52 case 'bayesian', | |
53 %score = bayesian_score_CPD(bnet.CPD{j}, data(fam, :)); | |
54 score = log_marg_prob_node(bnet.CPD{j}, data(j,:), data(ps,:)); | |
55 otherwise, | |
56 error(['unrecognized scoring fn ' scoring_fn]); | |
57 end |