|
wolffd@0
|
1 function [B,D,mu] = extract_params_from_gbn(bnet)
|
|
wolffd@0
|
2 % Extract all the local parameters of each Gaussian node, and collect them into global matrices.
|
|
wolffd@0
|
3 % [B,D,mu] = extract_params_from_gbn(bnet)
|
|
wolffd@0
|
4 %
|
|
wolffd@0
|
5 % B(i,j) is a block matrix that contains the transposed weight matrix from node i to node j.
|
|
wolffd@0
|
6 % D(i,i) is a block matrix that contains the noise covariance matrix for node i.
|
|
wolffd@0
|
7 % mu(i) is a block vector that contains the shifted noise mean for node i.
|
|
wolffd@0
|
8
|
|
wolffd@0
|
9 % In Shachter's model, the mean of each node in the global gaussian is
|
|
wolffd@0
|
10 % the same as the node's local unconditional mean.
|
|
wolffd@0
|
11 % In Alag's model (which we use), the global mean gets shifted.
|
|
wolffd@0
|
12
|
|
wolffd@0
|
13
|
|
wolffd@0
|
14 num_nodes = length(bnet.dag);
|
|
wolffd@0
|
15 bs = bnet.node_sizes(:); % bs = block sizes
|
|
wolffd@0
|
16 N = sum(bs); % num scalar nodes
|
|
wolffd@0
|
17
|
|
wolffd@0
|
18 B = zeros(N,N);
|
|
wolffd@0
|
19 D = zeros(N,N);
|
|
wolffd@0
|
20 mu = zeros(N,1);
|
|
wolffd@0
|
21
|
|
wolffd@0
|
22 for i=1:num_nodes % in topological order
|
|
wolffd@0
|
23 ps = parents(bnet.dag, i);
|
|
wolffd@0
|
24 e = bnet.equiv_class(i);
|
|
wolffd@0
|
25 %[m, Sigma, weights] = extract_params_from_CPD(bnet.CPD{e});
|
|
wolffd@0
|
26 s = struct(bnet.CPD{e}); % violate privacy of object
|
|
wolffd@0
|
27 m = s.mean; Sigma = s.cov; weights = s.weights;
|
|
wolffd@0
|
28 if length(ps) == 0
|
|
wolffd@0
|
29 mu(block(i,bs)) = m;
|
|
wolffd@0
|
30 else
|
|
wolffd@0
|
31 mu(block(i,bs)) = m + weights * mu(block(ps,bs));
|
|
wolffd@0
|
32 end
|
|
wolffd@0
|
33 B(block(ps,bs), block(i,bs)) = weights';
|
|
wolffd@0
|
34 D(block(i,bs), block(i,bs)) = Sigma;
|
|
wolffd@0
|
35 end
|
|
wolffd@0
|
36
|
|
wolffd@0
|
37
|
|
wolffd@0
|
38
|
