comparison toolboxes/FullBNT-1.0.7/bnt/examples/dynamic/reveal1.m @ 0:e9a9cd732c1e tip

first hg version after svn

+% Make a DBN with the following inter-connectivity matrix
+%    1
+%   /  \
+%  2   3
+%   \ /
+%    4 
+%    |
+%    5
+% where all arcs point down. In addition, there are persistence arcs from each node to itself.
+% There are no intra-slice connections.
+% Nodes have noisy-or CPDs.
+% Node 1 turns on spontaneously due to its leaky source.
+% This effect trickles down to the other nodes in the order shown.
+% All the other nodes inhibit their leaks.
+% None of the nodes inhibit the connection from themselves, so that once they are on, they remain
+% on (persistence).
+%
+% This model was used in the experiments reported in
+% - "Learning the structure of DBNs", Friedman, Murphy and Russell, UAI 1998.
+% where the structure was learned even in the presence of missing data.
+% In that paper, we used the structural EM algorithm.
+% Here, we assume full observability and tabular CPDs for the learner, so we can use a much
+% simpler learning algorithm.
+
+ss = 5;
+
+inter = eye(ss);
+inter(1,[2 3]) = 1;
+inter(2,4)=1;
+inter(3,4)=1;
+inter(4,5)=1;
+
+intra = zeros(ss);
+ns = 2*ones(1,ss);
+
+bnet = mk_dbn(intra, inter, ns);
+
+% All nodes start out off
+for i=1:ss
+  bnet.CPD{i} = tabular_CPD(bnet, i, [1.0 0.0]');
+end
+
+% The following params correspond to Fig 4a in the UAI 98 paper
+% The first arg is the leak inhibition prob.
+% The vector contains the inhib probs from the parents in the previous slice;
+% the last element is self, which is never inhibited.
+bnet.CPD{1+ss} = noisyor_CPD(bnet, 1+ss, 0.8, 0);
+bnet.CPD{2+ss} = noisyor_CPD(bnet, 2+ss, 1, [0.9 0]);
+bnet.CPD{3+ss} = noisyor_CPD(bnet, 3+ss, 1, [0.8 0]);
+bnet.CPD{4+ss} = noisyor_CPD(bnet, 4+ss, 1, [0.7 0.6 0]);
+bnet.CPD{5+ss} = noisyor_CPD(bnet, 5+ss, 1, [0.5 0]);
+
+
+% Generate some training data
+
+nseqs = 20;
+seqs = cell(1,nseqs);
+T = 30;
+for i=1:nseqs
+  seqs{i} = sample_dbn(bnet, T);
+end
+
+max_fan_in = 3; % let's cheat a little here
+
+% computing num. incorrect edges as a fn of the size of the training set
+%sz = [5 10 15 20];    
+sz = [5 10];    
+h = zeros(1, length(sz));
+for i=1:length(sz)
+  inter2 = learn_struct_dbn_reveal(seqs(1:sz(i)), ns, max_fan_in);
+  h(i) = sum(abs(inter(:)-inter2(:))); % hamming distance
+end
+h