comparison toolboxes/FullBNT-1.0.7/bnt/inference/dynamic/@ff_inf_engine/marginal_family.m @ 0:e9a9cd732c1e tip

first hg version after svn

+function marginal = marginal_family(engine, i, t)
+% MARGINAL_FAMILY Compute the marginal on the specified family (ff)
+% marginal = marginal_family(engine, i, t)
+
+
+if engine.filter
+  error('can''t currently use marginal_family when filtering with ff');
+end
+
+if nargin < 3, t = 1; end
+
+% The method is similar to the following HMM equation:
+% xi(i,j,t) = normalise( alpha(i,t) * transmat(i,j) * obsmat(j,t+1) * beta(j,t+1) )
+% where xi(i,j,t) = Pr(Q(t)=i, Q(t+1)=j | y(1:T))
+
+bnet = bnet_from_engine(engine);
+ss = length(bnet.intra);
+
+if myismember(i, engine.onodes)
+  ps = parents(bnet.dag, i);
+  p = ps(1);
+  marginal = pot_to_marginal(engine.marginals{ps(1),t});
+  fam = ([ps i]) + (t-1)*ss;
+elseif t==1
+  marginal = pot_to_marginal(engine.marginals{i,t});
+  fam = i + (t-1)*ss;
+else
+  pot = engine.CPDpot{i,t};
+  c = engine.obschild(i);
+  if c>0
+    pot = multiply_by_pot(pot, engine.CPDpot{c,t});
+  end
+  pot = multiply_by_pot(pot, engine.back{i,t});
+  ps = parents(bnet.dag, i+ss);
+  for p=ps(:)'
+    pot = multiply_by_pot(pot, engine.fwd{p,t-1});
+  end
+  marginal = pot_to_marginal(normalize_pot(pot));
+  fam = ([ps i+ss]) + (t-2)*ss;
+end
+
+% we convert the domain to the unrolled numbering system
+% so that update_ess extracts the right evidence.
+marginal.domain = fam;