|
wolffd@0
|
1 function pot = convert_to_pot(CPD, pot_type, domain, evidence)
|
|
wolffd@0
|
2 % CONVERT_TO_POT Convert a gmux CPD to a Gaussian potential
|
|
wolffd@0
|
3 % pot = convert_to_pot(CPD, pot_type, domain, evidence)
|
|
wolffd@0
|
4
|
|
wolffd@0
|
5 switch pot_type
|
|
wolffd@0
|
6 case {'d', 'u', 'cg', 'scg'},
|
|
wolffd@0
|
7 error(['can''t convert gmux to potential of type ' pot_type])
|
|
wolffd@0
|
8
|
|
wolffd@0
|
9 case {'c','g'},
|
|
wolffd@0
|
10 % We create a large weight matrix with zeros in all blocks corresponding
|
|
wolffd@0
|
11 % to the non-chosen parents, since they are effectively disconnected.
|
|
wolffd@0
|
12 % The chosen parent is determined by the value, m, of the discrete parent.
|
|
wolffd@0
|
13 % Thus the potential is as large as the whole family.
|
|
wolffd@0
|
14 ps = domain(1:end-1);
|
|
wolffd@0
|
15 dps = ps(CPD.dps); % CPD.dps is an index, not a node number (because of param tying)
|
|
wolffd@0
|
16 cps = ps(CPD.cps);
|
|
wolffd@0
|
17 m = evidence{dps};
|
|
wolffd@0
|
18 if isempty(m)
|
|
wolffd@0
|
19 error('gmux node must have observed discrete parent')
|
|
wolffd@0
|
20 end
|
|
wolffd@0
|
21 bs = CPD.sizes(CPD.cps);
|
|
wolffd@0
|
22 b = block(m, bs);
|
|
wolffd@0
|
23 sum_cpsz = sum(CPD.sizes(CPD.cps));
|
|
wolffd@0
|
24 selfsz = CPD.sizes(end);
|
|
wolffd@0
|
25 W = zeros(selfsz, sum_cpsz);
|
|
wolffd@0
|
26 W(:,b) = CPD.weights(:,:,m);
|
|
wolffd@0
|
27
|
|
wolffd@0
|
28 ns = zeros(1, max(domain));
|
|
wolffd@0
|
29 ns(domain) = CPD.sizes;
|
|
wolffd@0
|
30 self = domain(end);
|
|
wolffd@0
|
31 cdom = [cps(:)' self];
|
|
wolffd@0
|
32 pot = linear_gaussian_to_cpot(CPD.mean(:,m), CPD.cov(:,:,m), W, domain, ns, cdom, evidence);
|
|
wolffd@0
|
33
|
|
wolffd@0
|
34 otherwise,
|
|
wolffd@0
|
35 error(['unrecognized pot_type' pot_type])
|
|
wolffd@0
|
36 end
|
|
wolffd@0
|
37
|
