dbn_update_bel.m

来自「Bayes网络工具箱」· M 代码 · 共 23 行

function [newbel, loglik] = dbn_update_bel(engine, oldbel, evidence)
% DBN_UPDATE_BEL Update the belief state (hmm)
% [newbel, loglik] = dbn_update(engine, oldbel, evidence)
%
% oldbel is Pr(X(t-1)|Y(1:t-1)) and newbel is Pr(X(t)|Y(1:t)),
% where X are the hidden nodes, Y the observed nodes.
% If t>1
%  evidence{i,1} contains the evidence on node i in slice t-1
%  evidence{i,2} contains the evidence on node i in slice t
% else
%  evidence{i} contains the evidence on node i in slice 1
%  and oldbel is just the prior (computed using dbn_init_bel)

if size(evidence,2)==2
  obslik = mk_obs_lik(engine, evidence, 2);
  [newbel, lik] = normalise((engine.transmat' * oldbel) .* obslik);
else
  obslik = mk_obs_lik(engine, evidence, 1);
  [newbel, lik] = normalise(oldbel .* obslik);
end

loglik = log(lik);