Mercurial > hg > camir-aes2014
view toolboxes/FullBNT-1.0.7/bnt/examples/static/Belprop/belprop_loop1_gauss.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 source
% Compare different loopy belief propagation algorithms on a graph with a single loop. % LBP should give exact results if it converges. N = 4; dag = zeros(N,N); C = 1; S = 2; R = 3; W = 4; dag(C,[R S]) = 1; dag(R,W) = 1; dag(S,W)=1; ns = 2*ones(1,N); bnet = mk_bnet(dag, ns, 'discrete', []); for i=1:N bnet.CPD{i} = gaussian_CPD(bnet, i); end engines = {}; engines{end+1} = jtree_inf_engine(bnet); engines{end+1} = pearl_inf_engine(bnet, 'protocol', 'parallel', 'max_iter', 20); %engines{end+1} = pearl_inf_engine(bnet, 'protocol', 'parallel', 'max_iter', 20, 'filename', ... % '/home/eecs/murphyk/matlab/gausspearl.txt', 'tol', 1e-5); % pearl gaussian does not compute loglik [time, engines] = cmp_inference_static(bnet, engines, 'maximize', 0, 'exact', 1, 'observed', [2], ... 'check_ll', 0, 'singletons_only', 0, 'check_converged', [2]);