function [Y,G,E,Y2]=TwoLayers2(Inp,Tar,W,NNP)


W=W(:)';

[np,ni]=size(Inp);

%	PWeigths --- pointer to the weights
%	PBias --- pointer to the current bias

PWeigths=1:ni;
PBias=ni*NNP(1)+NNP(1)*NNP(2)+1;

%	Obtain the weigths from the input to the first layer in
%	W1 and the bias for the first hidden layer in T1

for i=1:NNP(1)
  W1(:,i)=W(PWeigths)';
  PWeigths=PWeigths+ni;
end;
T1=W(PBias:PBias-1+NNP(1));

%	Obtain the weigths from the first to the second layer in
%	W2 and the bias for the first hidden layer in T2

PWeigths=(1:NNP(1))+NNP(1)*ni;
PBias=PBias+NNP(1);

for i=1:NNP(2)
  W2(:,i)=W(PWeigths)';
  PWeigths=PWeigths+NNP(1);
end;
T2=W(PBias:PBias-1+NNP(2));

%	Xi is the net imput to the neurons in the i-th layer; Yi is the 
%	output of the neurons in the i-th layer; Deri is the derivate of
%	Yi w.r.t. Xi

X1=Inp*W1+ones(np,1)*T1;
Y1=ones(np,NNP(1))./(1+exp(-X1));
Der1=Y1.*(1-Y1);

X2=Y1*W2+ones(np,1)*T2;
Y2=ones(np,NNP(2))./(1+exp(-X2));
Der2=Y2.*(1-Y2);

teta=[ones(np,1) Y2]\Tar;
keyboard
Y=[ones(np,1) Y2]*teta;


for i=1:NNP(2)
  Der2(:,i)=Der2(:,i)*teta(i+1);
end;

l=1;
PBias=(ni*NNP(1))+NNP(1)*NNP(2)+1;
for i=1:NNP(1)
  delta=(Der2*W2(i,:)').*Der1(:,i);
  for j=1:ni
    G(:,l)=delta.*Inp(:,j);
    l=l+1;
  end;
  G(:,PBias)=delta;
  PBias=PBias+1;
end;

for i=1:NNP(2)
  for j=1:NNP(1);
    G(:,l)=Der2(:,i).*Y1(:,j);
    l=l+1;
  end;
  G(:,PBias)=Der2(:,i);
  PBias=PBias+1;
end;

E=G'*(Y-Tar);