wolffd@0: function risultati = fhme(net, nodes_info, data, n)
wolffd@0: %HMEFWD	Forward propagation through an HME model
wolffd@0: %
wolffd@0: % Each row of the (n x class_num) matrix 'risultati' containes the estimated class posterior prob.
wolffd@0: %
wolffd@0: % ----------------------------------------------------------------------------------------------------
wolffd@0: % -> pierpaolo_b@hotmail.com   or   -> pampo@interfree.it
wolffd@0: % ----------------------------------------------------------------------------------------------------
wolffd@0: %
wolffd@0: ns=net.node_sizes;
wolffd@0: if nargin==3
wolffd@0:     ndata=n;
wolffd@0: else
wolffd@0:     ndata=size(data, 1);
wolffd@0: end
wolffd@0: altezza=size(ns,2);
wolffd@0: coeff=cell(altezza-1,1);
wolffd@0: for m=1:ndata
wolffd@0:     %- i=2 --------------------------------------------------------------------------------------
wolffd@0:     s=struct(net.CPD{2});    
wolffd@0:     if nodes_info(1,2)==0,
wolffd@0:         mu=[]; W=[]; predict=[];
wolffd@0:         mu=s.mean(:,:);
wolffd@0:         W=s.weights(:,:,:);
wolffd@0:         predict=mu(:,:)+W(:,:,:)*data(m,:)';            
wolffd@0:         coeff{1,1}=predict';            
wolffd@0:     elseif nodes_info(1,2)==1,
wolffd@0:         coeff{1,1}=fglm(s.glim{1}, data(m,:));
wolffd@0:     else,
wolffd@0:         coeff{1,1}=fmlp(s.mlp{1}, data(m,:));
wolffd@0:     end
wolffd@0:     %----------------------------------------------------------------------------------------------
wolffd@0:     if altezza>3,
wolffd@0:         for i=3:altezza-1,
wolffd@0:             s=[]; f=[]; dpsz=[];
wolffd@0:             f=family(net.dag,i); f=f(2:end-1); dpsz=prod(ns(f));
wolffd@0:             s=struct(net.CPD{i});
wolffd@0:             for j=1:dpsz,
wolffd@0:                 if nodes_info(1,i)==1,
wolffd@0:                     coeff{i-1,1}(j,:)=coeff{i-2,1}(1,j)*fglm(s.glim{j}, data(m,:));
wolffd@0:                 else
wolffd@0:                     coeff{i-1,1}(j,:)=coeff{i-2,1}(1,j)*fmlp(s.mlp{j}, data(m,:));
wolffd@0:                 end
wolffd@0:             end       
wolffd@0:             app=cat(2, coeff{i-1,1}(:)); coeff{i-1,1}=app'; clear app;
wolffd@0:         end
wolffd@0:     end
wolffd@0:     %- i=altezza ----------------------------------------------------------------------------------
wolffd@0:     if altezza>2,
wolffd@0:         i=altezza;
wolffd@0:         s=[]; f=[]; dpsz=[];
wolffd@0:         f=family(net.dag,i); f=f(2:end-1); dpsz=prod(ns(f));
wolffd@0:         s=struct(net.CPD{i});
wolffd@0:         if nodes_info(1,i)==0,            
wolffd@0:             mu=[]; W=[];
wolffd@0:             mu=s.mean(:,:);
wolffd@0:             W=s.weights(:,:,:);
wolffd@0:         end
wolffd@0:         for j=1:dpsz,
wolffd@0:             if nodes_info(1,i)==0,            
wolffd@0:                 predict=[];
wolffd@0:                 predict=mu(:,j)+W(:,:,j)*data(m,:)';            
wolffd@0:                 coeff{i-1,1}(j,:)=coeff{i-2,1}(1,j)*predict';            
wolffd@0:             elseif nodes_info(1,i)==1,
wolffd@0:                 coeff{i-1,1}(j,:)=coeff{i-2,1}(1,j)*fglm(s.glim{j}, data(m,:));
wolffd@0:             else
wolffd@0:                 coeff{i-1,1}(j,:)=coeff{i-2,1}(1,j)*fmlp(s.mlp{j}, data(m,:));
wolffd@0:             end
wolffd@0:         end
wolffd@0:     end
wolffd@0:     %----------------------------------------------------------------------------------------------
wolffd@0:     risultati(m,:)=sum(coeff{altezza-1,1},1);
wolffd@0:     clear coeff; coeff=cell(altezza-1,1);
wolffd@0: end
wolffd@0: return
wolffd@0: 
wolffd@0: %-------------------------------------------------------------------
wolffd@0: 
wolffd@0: function [y, a] = fglm(net, x)
wolffd@0: %GLMFWD	Forward propagation through 1-layer net->GLM statistical model
wolffd@0: 
wolffd@0: ndata = size(x, 1);
wolffd@0: 
wolffd@0: a = x*net.w1 + ones(ndata, 1)*net.b1;
wolffd@0: 
wolffd@0: nout = size(a,2);
wolffd@0: % Ensure that sum(exp(a), 2) does not overflow
wolffd@0: maxcut = log(realmax) - log(nout);
wolffd@0: % Ensure that exp(a) > 0
wolffd@0: mincut = log(realmin);
wolffd@0: a = min(a, maxcut);
wolffd@0: a = max(a, mincut);
wolffd@0: temp = exp(a);
wolffd@0: y = temp./(sum(temp, 2)*ones(1,nout));
wolffd@0: 
wolffd@0: %-------------------------------------------------------------------
wolffd@0: 
wolffd@0: function [y, z, a] = fmlp(net, x)
wolffd@0: %MLPFWD	Forward propagation through 2-layer network.
wolffd@0: 
wolffd@0: ndata = size(x, 1);
wolffd@0: 
wolffd@0: z = tanh(x*net.w1 + ones(ndata, 1)*net.b1);
wolffd@0: a = z*net.w2 + ones(ndata, 1)*net.b2;  
wolffd@0: temp = exp(a);
wolffd@0: nout = size(a,2);
wolffd@0: y = temp./(sum(temp,2)*ones(1,nout));
wolffd@0: 
wolffd@0: %-------------------------------------------------------------------