camir-aes2014: toolboxes/FullBNT-1.0.7/bnt/general/Old/calc_mpe_given_inf

first hg version after svn

rev	line source
wolffd@0	1 function [mpe, prob] = calc_mpe_given_inf_engine(engine, evidence)
wolffd@0	2 % CALC_MPE_GIVEN_ENGINE Computes the most probable explanation of the evidence
wolffd@0	3 % [mpe, prob] = calc_mpe_given_inf_engine(engine, evidence)
wolffd@0	4 %
wolffd@0	5 % INPUT
wolffd@0	6 % engine must support max-propagation
wolffd@0	7 % evidence{i} is the obsevred value of node i, or [] if hidden
wolffd@0	8 %
wolffd@0	9 % OUTPUT
wolffd@0	10 % mpe(i) is the most likely value of node i
wolffd@0	11 % prob is the likelihood of the globally best assignment
wolffd@0	12 %
wolffd@0	13 % This currently only works when all nodes are discrete
wolffd@0	14
wolffd@0	15 [engine, ll] = enter_evidence(engine, evidence);
wolffd@0	16
wolffd@0	17 observed = ~isemptycell(evidence);
wolffd@0	18 N = length(evidence);
wolffd@0	19 mpe = zeros(1,N);
wolffd@0	20 for i=1:N
wolffd@0	21 m = marginal_nodes(engine, i);
wolffd@0	22 % discrete observed nodes are all set to 1 inside the inference engine, so we must undo this
wolffd@0	23 if observed(i)
wolffd@0	24 mpe(i) = evidence{i};
wolffd@0	25 else
wolffd@0	26 mpe(i) = argmax(m.T);
wolffd@0	27 end
wolffd@0	28 end
wolffd@0	29
wolffd@0	30 bnet = bnet_from_engine(engine);
wolffd@0	31 ll = log_lik_complete(bnet, num2cell(mpe(:)));
wolffd@0	32 prob = exp(ll);

Mercurial > hg > camir-aes2014