dbn_predict.m

Bayes网络工具箱

function [engine] = dbn_predict(engine, bel, T, ev)
% DBN_PREDICT K-step-ahead prediction (hmm)
% [engine] = dbn_predict(engine, bel, K, ev, pot_type)
%
% bel = Pr(X(t-1)|Y(t-1)) is the prior
% ev{i} is the evidence on slice t-1
% We essentially unroll the network to create slices t-1, t, t+1, ..., t+K-1
% so we can extract the predicted value of node i k steps into the future
% using marginal_nodes(i, k+1)

error('unfinished');

onodes = find(~isemptycell(ev));
hnodes = find(isemptycell(ev));
bnet = bnet_from_engine(engine);
ns = bnet.node_sizes_slice;
assert(isequal(onodes, bnet.cnodes_slice)); % we currently assume observed leaves are Gaussians
assert(isequal(hnodes, bnet.dnodes_slice)); 
ns(onodes) = 1;

alpha = zeros(prod(ns(hnodes)), T);
M = engine.transmat';
t = 1;
alpha(:,t) = bel;
for t=2:T
  alpha(:,t) = M * alpha(:,t-1);
end