wolffd@0: function engine = belprop_fg_inf_engine(fg, varargin) 
wolffd@0: % BELPROP_FG_INF_ENGINE Make a belief propagation inference engine for factor graphs
wolffd@0: % engine = belprop_fg_inf_engine(factor_graph, ...)
wolffd@0: %
wolffd@0: % The following optional arguments can be specified in the form of name/value pairs:
wolffd@0: % [default in brackets]
wolffd@0: % e.g., engine = belprop_inf_engine(fg, 'tol', 1e-2, 'max_iter', 10)
wolffd@0: %
wolffd@0: % max_iter - max. num. iterations [ 2*num_nodes ]
wolffd@0: % momentum - weight assigned to old message in convex combination (useful for damping oscillations) [0]
wolffd@0: % tol - tolerance used to assess convergence [1e-3]
wolffd@0: % maximize - 1 means use max-product, 0 means use sum-product [0]
wolffd@0: %
wolffd@0: % This uses potential objects, like belprop_inf_engine, and hence is quite slow.
wolffd@0: 
wolffd@0: engine = init_fields;
wolffd@0: engine = class(engine, 'belprop_fg_inf_engine');
wolffd@0: 
wolffd@0: % set params to default values
wolffd@0: N = length(fg.G);
wolffd@0: engine.max_iter = 2*N;
wolffd@0: engine.momentum = 0;
wolffd@0: engine.tol = 1e-3;
wolffd@0: engine.maximize = 0;
wolffd@0: 
wolffd@0: % parse optional arguments
wolffd@0: engine = set_params(engine, varargin);
wolffd@0: 
wolffd@0: engine.fgraph = fg;
wolffd@0: 
wolffd@0: % store results computed by enter_evidence here
wolffd@0: engine.marginal_nodes = cell(1, fg.nvars);
wolffd@0: engine.evidence = [];
wolffd@0: 
wolffd@0: 
wolffd@0: %%%%%%%%%%%%
wolffd@0: 
wolffd@0: function engine = init_fields()
wolffd@0: 
wolffd@0: engine.fgraph = [];
wolffd@0: engine.max_iter = [];
wolffd@0: engine.momentum = [];
wolffd@0: engine.tol = [];
wolffd@0: engine.maximize = [];
wolffd@0: engine.marginal_nodes = [];
wolffd@0: engine.evidence = [];
wolffd@0: engine.niter = [];