Mercurial > hg > camir-aes2014
view toolboxes/FullBNT-1.0.7/bnt/inference/static/@gibbs_sampling_inf_engine/private/compute_posterior_dbn.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
function post = compute_posterior_dbn(bnet, state, i, n, strides, families, ... CPT) % COMPUTE_POSTERIOR % % post = compute_posterior(bnet, state, i, n, strides, families, % cpts) % % Compute the posterior distribution on node X_i^n of a DBN, % conditional on evidence in the cell array state % % strides is the cached result of compute_strides(bnet) % families is the cached result of compute_families(bnet) % cpt is the cached result of get_cpts(bnet) % % post is a one-dimensional table % First multiply in the cpt of the node itself post = get_slice_dbn(bnet, state, i, n, i, n, strides, families, CPT); post = post(:); % Then multiply in CPTs of children that are in this slice for j = children(bnet.intra, i) slice = get_slice_dbn(bnet, state, j, n, i, n, strides, families, CPT); post = post.*slice(:); end % Finally, if necessary, multiply in CPTs of children in the next % slice if (n < size(state,2)) for j = children(bnet.inter, i) slice = get_slice_dbn(bnet, state, j, n+1, i, n, strides, families, ... CPT); post = post.*slice(:); end end post = normalise(post);