Write code for NN using the Weight and Bias data retrieved from the NN tool box

Hi,

I have trained a simple neural network and saved all the resultant biases and weights.

Now, I want to use them to build the network manually which I did. I mapped the inputs into -1 and 1 range. multiplied by inputWeights and added the bias. I applied the tansig transfer function.

Then for the second layer, I multiplied the previous output with the second group of weights, added the bias and applied the transfer function. However, I don't get correct results.

To compare: output of net(inputs) and myNN(inputs is different) and mine is wrong.

 MatlabCorrect  Mine(wrong)

 0.0001         -0.8220
 0.9733         -0.9720
 0.0015         -0.9963
 0.0003         -0.9991
 0.0201         -0.9949
 0.0026         -0.9638

What could be the reason?

Here is my code if it helps

% ILW: input layer weight
% HLW: hidden layer weight
% ILB: input layer bias
% HLB: hidden layer bias

ALLSAM = load ('allsamples.mat');
minArr = (min(allsamples'))';
maxArr = (max(allsamples'))';
%random sample
input = ALLSAM (:,15);
%mapping to -1 - +1 range
input = 2*((a1-minArr)./(maxArr-minArr)) -1; 

[R C] = size(input);
L1 = zeros(size(ILW));
for i = 1:C
     L1(:,i) = input(i) .* ILW(:,i);
end
L1 = sum(L1');
L1 = L1' + ILB;
L1 = tansig(L1);
[R C] = size (HLW);
O = zeros(size(HLW));
for i = 1:C
     O(:,i) = L1(i) .* HLW(:,i);
end
O = sum(O');
O = O' + HLB;
O = tansig(O)

Answer 1

[Greg Heath]

Searching the Newsgroup using IW heath

close all, clear all, clc

[ x, t ] = simplefit_dataset;

[ I N ] = size(x)

[ O N ] = size(t)

figure

hold on

plot(x,t,'o','LineWidth',2)

drawnow

H = 10

net = feedforwardnet(H);

net = train(net,x,t);

y = net(x);

plot(x,y,'r','LineWidth',2)

drawnow

IW = net.IW{1,1}

LW = net.LW{2,1}

b1 = net.b{1}

b2 = net.b{2}

xmin = min(x)

xmax=max(x)

xn = -1+ 2*(x-xmin)/(xmax-xmin) ;

h =tansig(IW*xn+b1*ones(1,N));

yn = LW*h+b2;

tmin = min(t)

tmax = max(t)

yhat = tmin +(tmax-tmin)*(yn +1)/2;

plot(x,yhat,'g','LineWidth',2')

mse(y-yhat)

Hope this helps.

Greg

[HARINI R]

Can you help me with the code for NARX network (10 inputs, 1 target, 6 i/p and feedback delays) to manually calculate the output using extracted weights and bias from trained network?