Mercurial > hg > camir-ismir2012

annotate toolboxes/FullBNT-1.0.7/bnt/learning/bic_score_family.m @ 0:cc4b1211e677 tip

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
initial commit to HG from Changeset: 646 (e263d8a21543) added further path and more save "camirversion.m"
author Daniel Wolff
date Fri, 19 Aug 2016 13:07:06 +0200
parents
children
rev   line source
Daniel@0 1 function [S, LL] = bic_score(counts, CPT, ncases)
Daniel@0 2 % BIC_SCORE Bayesian Information Criterion score for a single family
Daniel@0 3 % [S, LL] = bic_score(counts, CPT, ncases)
Daniel@0 4 %
Daniel@0 5 % S is a large sample approximation to the log marginal likelihood,
Daniel@0 6 % which can be computed using dirichlet_score.
Daniel@0 7 %
Daniel@0 8 % S = \log [ prod_j _prod_k theta_ijk ^ N_ijk ] - 0.5*d*log(ncases)
Daniel@0 9 % where counts encode N_ijk, theta_ijk is the MLE comptued from counts,
Daniel@0 10 % and d is the num of free parameters.
Daniel@0 11
Daniel@0 12 %CPT = mk_stochastic(counts);
Daniel@0 13 tiny = exp(-700);
Daniel@0 14 LL = sum(log(CPT(:) + tiny) .* counts(:));
Daniel@0 15 % CPT(i) = 0 iff counts(i) = 0 so it is okay to add tiny
Daniel@0 16
Daniel@0 17 ns = mysize(counts);
Daniel@0 18 ns_ps = ns(1:end-1);
Daniel@0 19 ns_self = ns(end);
Daniel@0 20 nparams = prod([ns_ps (ns_self-1)]);
Daniel@0 21 % sum-to-1 constraint reduces the effective num. vals of the node by 1
Daniel@0 22
Daniel@0 23 S = LL - 0.5*nparams*log(ncases);