diff toolboxes/FullBNT-1.0.7/bnt/examples/static/qmr1.m @ 0:e9a9cd732c1e tip

first hg version after svn
author wolffd
date Tue, 10 Feb 2015 15:05:51 +0000
parents
children
line wrap: on
line diff
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/toolboxes/FullBNT-1.0.7/bnt/examples/static/qmr1.m	Tue Feb 10 15:05:51 2015 +0000
@@ -0,0 +1,112 @@
+% Make a QMR-like network 
+% This is a bipartite graph, where the top layer contains hidden disease nodes,
+% and the bottom later contains observed finding nodes.
+% The diseases have Bernoulli CPDs, the findings noisy-or CPDs.
+% See quickscore_inf_engine for references.
+
+pMax = 0.01;
+Nfindings = 10;
+Ndiseases = 5;
+%Nfindings = 20;
+%Ndiseases = 10;
+
+N=Nfindings+Ndiseases;
+findings = Ndiseases+1:N;
+diseases = 1:Ndiseases;
+
+G = zeros(Ndiseases, Nfindings);
+for i=1:Nfindings
+  v= rand(1,Ndiseases);
+  rents = find(v<0.8);
+  if (length(rents)==0)
+    rents=ceil(rand(1)*Ndiseases);
+  end
+  G(rents,i)=1;
+end       
+
+prior = pMax*rand(1,Ndiseases);
+leak = 0.5*rand(1,Nfindings); % in real QMR, leak approx exp(-0.02) = 0.98     
+%leak = ones(1,Nfindings); % turns off leaks, which makes inference much harder
+inhibit = rand(Ndiseases, Nfindings);
+inhibit(not(G)) = 1;
+
+
+% first half of findings are +ve, second half -ve
+% The very first and last findings are hidden
+pos = 2:floor(Nfindings/2);
+neg = (pos(end)+1):(Nfindings-1);
+
+% Make the bnet in the straightforward way
+tabular_leaves = 0;
+obs_nodes = myunion(pos, neg) + Ndiseases;
+big_bnet = mk_qmr_bnet(G, inhibit, leak, prior, tabular_leaves, obs_nodes);
+big_evidence = cell(1, N);
+big_evidence(findings(pos)) = num2cell(repmat(2, 1, length(pos)));
+big_evidence(findings(neg)) = num2cell(repmat(1, 1, length(neg)));
+
+%clf;draw_layout(big_bnet.dag);
+%filename = '../public_html/Bayes/Figures/qmr.rnd.jpg';
+%% 3x3 inches
+%set(gcf,'units','inches');
+%set(gcf,'PaperPosition',[0 0 3 3])  
+%print(gcf,'-djpeg','-r100',filename);
+
+
+% Marginalize out hidden leaves apriori
+positive_leaves_only = 1;
+[bnet, vals] = mk_minimal_qmr_bnet(G, inhibit, leak, prior, pos, neg, positive_leaves_only);
+obs_nodes = bnet.observed;
+evidence = cell(1, Ndiseases + length(obs_nodes));
+evidence(obs_nodes) = num2cell(vals);
+
+
+clear engine;
+engine{1} = quickscore_inf_engine(inhibit, leak, prior);
+engine{2} = jtree_inf_engine(big_bnet);
+engine{3} = jtree_inf_engine(bnet);
+
+%fname = '/home/cs/murphyk/matlab/Misc/loopybel.txt';
+global BNT_HOME
+fname = sprintf('%s/loopybel.txt', BNT_HOME);
+
+
+max_iter = 6;
+engine{4} = pearl_inf_engine(bnet, 'protocol', 'parallel', 'max_iter', max_iter);
+%engine{5} = belprop_inf_engine(bnet, 'max_iter', max_iter, 'filename', fname);
+engine{5} = belprop_inf_engine(bnet, 'max_iter', max_iter);
+
+E = length(engine);
+exact = 1:3;
+loopy = [4 5];
+
+ll = zeros(1,E);
+tic; engine{1} = enter_evidence(engine{1}, pos, neg); toc
+tic; [engine{2}, ll(2)] = enter_evidence(engine{2}, big_evidence); toc
+tic; [engine{3}, ll(3)] = enter_evidence(engine{3}, evidence); toc
+tic; [engine{4}, ll(4), niter(4)] = enter_evidence(engine{4}, evidence); toc
+tic; [engine{5}, niter(5)] = enter_evidence(engine{5}, evidence); toc
+
+ll
+
+post = zeros(E, Ndiseases);
+for e=1:E
+  for i=diseases(:)'
+    m = marginal_nodes(engine{e}, i);
+    post(e, i) = m.T(2);
+  end
+end
+
+for e=exact(:)'
+  for i=diseases(:)'
+    assert(approxeq(post(1, i), post(e, i)));
+  end
+end
+
+a = zeros(Ndiseases, 2);
+for ei=1:length(loopy)
+  for i=diseases(:)'
+    a(i,ei) = approxeq(post(1, i), post(loopy(ei), i));
+  end
+end
+disp('is the loopy posterior correct?');
+disp(a)