function engine = belprop_inf_engine(fg, max_iter, momentum, tol, maximize) if nargin < 2, max_iter = length(fg.G); end if nargin < 3, momentum = 0; end if nargin < 4, tol = 1e-3; end if nargin < 5, maximize = 0; end engine.fgraph = fg; engine.max_iter = max_iter; engine.momentum = momentum; engine.tol = tol; engine.maximize = maximize; % store results computed by enter_evidence here ndoms = length(fg.doms); nvars = length(fg.vars); engine.marginal_domains = cell(1, ndoms); % to compute the marginal on each variable, we need to know which domain to marginalize % so we represent each domain as a bit vector, and compute its (pre-evidence) weight engine.dom_weight = []; % engine.dom_bitv = sparse(ndoms, nvars); % ns = fg.node_sizes; % for i=1:ndoms % engine.dom_bitv(i, fg.doms{i}) = 1; % engine.dom_weight(i) = prod(ns(fg.doms{i})); % end engine = class(engine, 'belprop_inf_engine');
