function [Criterion,MSE,Comp,Model,y,total_out,w]=evaluate_model(InpPat,TarPat,Model)

% [Criterion,MSE,Comp]=evaluate_model(InpPat,TarPat,Model)
%
% Estimates the parameters of the current model using a pseudo-inverse, and computes the BIC
%
% Input parameters:	InpPat - Input Matrix
%							TarPat - Desired output vector
%							Model - model to be evaluated
%
% Output parameters:	Criterion - Bayesian Information Criterion - Criterion=n_pat*ln(MSE)+Comp*ln(n_pat)
%							MSE - Mean-Square-Error
%							Comp - Complexity of the model (total number of basis functions
%
% Written by A. Ruano 16.6.99
%
[n_pat,n_inp]=size(InpPat);

n_s_b=Model.n_s_m;

total_out=[];
for i=1:n_s_b
   Sub_model=getfield(Model,'Sub_Model',{i});
   [y,out,w]=train_B_splines(InpPat(:,Sub_model.var),TarPat,Sub_model.x_min,Sub_model.x_max,...
      Sub_model.order,Sub_model.Knots,Sub_model.n_Knots,Sub_model.n_fun,0);
   [n_pat,m]=size(out);
   a_w(1:m,i)=ones(m,1);
   for j=1:m
      if sum(zeros(n_pat,1)-out(:,j))==0
         % null column
         a_w(j,i)=0;
      else
         total_out=[total_out out(:,j)];
      end   
   end   
end

w=pinv(total_out)*TarPat;
y=total_out*w;

pos=1;
Comp=0;
for i=1:n_s_b
   ww=zeros(length(Model.Sub_Model(i).w),1);
   m=length(ww);
   for j=1:m
      if a_w(j,i)==1
         ww(j)=w(pos);
         pos=pos+1;
      else
         ww(j)=0;
      end
   end   
   Model.Sub_Model(i).w=ww;
   Model.Sub_Model(i).a_w=a_w(1:m,i);;

   Comp=Comp+sum(a_w(1:m,i));
end

MSE=(norm(TarPat-y)^2)/n_pat;
Criterion=n_pat*log(MSE)+Comp*log(n_pat);

Model.crit=Criterion;