Daniel@0: function pot = convert_to_pot(CPD, pot_type, domain, evidence)
Daniel@0: % CONVERT_TO_POT Convert a tabular CPD to one or more potentials
Daniel@0: % pot = convert_to_pot(CPD, pot_type, domain, evidence)
Daniel@0: 
Daniel@0: % This is the same as discrete_CPD/convert_to_pot,
Daniel@0: % except we didn't want to the kernel to inherit methods like sample_node etc.
Daniel@0: 
Daniel@0: sz = CPD.sz;
Daniel@0: ns = zeros(1, max(domain));
Daniel@0: ns(domain) = sz;
Daniel@0: 
Daniel@0: odom = domain(~isemptycell(evidence(domain)));
Daniel@0: T = convert_to_table(CPD, domain, evidence);
Daniel@0: 
Daniel@0: switch pot_type
Daniel@0:  case 'u',
Daniel@0:   pot = upot(domain, sz, T, 0*myones(sz));  
Daniel@0:  case 'd',
Daniel@0:   ns(odom) = 1;
Daniel@0:   pot = dpot(domain, ns(domain), T);          
Daniel@0:  case 'c',
Daniel@0:   % Since we want the output to be a Gaussian, the whole family must be observed.
Daniel@0:   % In other words, the potential is really just a constant.
Daniel@0:   p = T.p;
Daniel@0:   %p = prob_node(CPD, evidence(domain(end)), evidence(domain(1:end-1)));
Daniel@0:   ns(domain) = 0;
Daniel@0:   pot = cpot(domain, ns(domain), log(p));       
Daniel@0:  case 'cg',
Daniel@0:   T = T(:);
Daniel@0:   ns(odom) = 1;
Daniel@0:   can = cell(1, length(T));
Daniel@0:   for i=1:length(T)
Daniel@0:     can{i} = cpot([], [], log(T(i)));
Daniel@0:   end
Daniel@0:   pot = cgpot(domain, [], ns, can);   
Daniel@0: end
Daniel@0: