wolffd@0: function mpe = find_mpe(engine, evidence, varargin)
wolffd@0: % FIND_MPE Find the most probable explanation of the data  (belprop)
wolffd@0: % function mpe = find_mpe(engine, evidence,...)
wolffd@0: %
wolffd@0: % evidence{i} = [] if X(i) is hidden, and otherwise contains its observed value (scalar or column vector).
wolffd@0: %
wolffd@0: % This finds the marginally most likely value for each hidden node,
wolffd@0: % and may give the wrong results even if the graph is acyclic,
wolffd@0: % unless you set break_ties = 1.
wolffd@0: %
wolffd@0: % The following optional arguments can be specified in the form of name/value pairs:
wolffd@0: % [default value in brackets]
wolffd@0: %
wolffd@0: % break_ties is optional. If 1, we will force ties to be broken consistently
wolffd@0: %  by calling enter_evidence N times. (see Jensen96, p106) Default = 1.
wolffd@0: 
wolffd@0: break_ties = 1;
wolffd@0: 
wolffd@0: % parse optional params
wolffd@0: args = varargin;
wolffd@0: nargs = length(args);
wolffd@0: for i=1:2:nargs
wolffd@0:   switch args{i},
wolffd@0:    case 'break_ties',    break_ties = args{i+1}; 
wolffd@0:    otherwise,  
wolffd@0:     error(['invalid argument name ' args{i}]);       
wolffd@0:   end
wolffd@0: end
wolffd@0: 
wolffd@0: engine = enter_evidence(engine, evidence, 'maximize', 1);
wolffd@0: 
wolffd@0: observed = ~isemptycell(evidence);
wolffd@0: evidence = evidence(:); % hack to handle unrolled DBNs
wolffd@0: N = length(evidence);
wolffd@0: mpe = cell(1,N);
wolffd@0: for i=1:N
wolffd@0:   m = marginal_nodes(engine, i);
wolffd@0:   % observed nodes are all set to 1 inside the inference engine, so we must undo this
wolffd@0:   if observed(i)
wolffd@0:     mpe{i} = evidence{i};
wolffd@0:   else
wolffd@0:     mpe{i} = argmax(m.T);
wolffd@0:     if break_ties
wolffd@0:       evidence{i} = mpe{i};                             
wolffd@0:       [engine, ll] = enter_evidence(engine, evidence, 'maximize', 1);  
wolffd@0:     end
wolffd@0:   end
wolffd@0: end
wolffd@0: