hmm_inf_engine.m

来自「Bayes网络工具箱」· M 代码 · 共 44 行

function engine = hmm_inf_engine(bnet, onodes, oneslice)
% HMM_INF_ENGINE Inference engine for discrete DBNs which uses the forwards-backwards algorithm.
% engine = hmm_inf_engine(bnet, onodes, oneslice)
%
% 'onodes' specifies which nodes are observed; these must be leaves, and can be discrete or continuous.
% The remaining nodes are all hidden, and must be discrete.
% The hidden nodes must be persistent, i.e., they must have children in
% the next time slice. In addition, they may not have any children within the current slice,
% except to the observed leaves. In other words, the topology must be isomorphic to a standard HMM.
%
% If 'oneslice' = 1, we only compute the 1-slice marginals. (Default: compute 1 and 2-slice marginals.)
% The 2-slice marginals take T*Q*Q space to store, where Q is the number of hidden states,
% so if you are just interested in marginals on single nodes, use oneslice=1.
%
% For details on Hidden Markov Models, see
% - "A tutorial on Hidden Markov Models and selected applications in speech recognition",
%          L. Rabiner, Proc. IEEE 77(2):257--286, 1989.

if nargin < 3, oneslice = 0; end

[persistent_nodes, transient_nodes] = partition_dbn_nodes(bnet.intra, bnet.inter);
assert(isequal(sort(onodes), transient_nodes));
[engine.prior, engine.transmat] = dbn_to_hmm(bnet, onodes);

engine.onodes = onodes;

ss = length(bnet.intra);
hnodes = mysetdiff(1:ss, onodes);
engine.hnodes = hnodes;
ns = bnet.node_sizes;
onodes2 = [onodes onodes+ss];
ns(onodes2) = 1;
engine.node_sizes = ns;
engine.bel = [];

% This is where we will store the results between enter_evidence and marginal_nodes
engine.one_slice_marginal = [];
engine.two_slice_marginal = [];

engine.oneslice = oneslice;

engine = class(engine, 'hmm_inf_engine', inf_engine(bnet));