In neural net function, how can I see normalized input data?

The neural network function uses mapminmax function.

When I use neural network, I want to see normalized input data using mapminmax function.

But, I can't find normalized input data.

How can I see normalized input data?

Please help me.

Answer 1

[Greg Heath]

You have to use MAPMINMAX and calculate them yourself.

The normalization inside of the training function is not accessible. For details you can look up

 help mapminmax

 doc mapminmax

Hope this helps

Thank you for formally accepting my answer

Greg

[kyeonghwan kim]

Thank you for your reply.

I have a question.

I tried to normalize it in the way you gave it.

This is the code.

 % Automatic Normalization
[x,t] = simplefit_dataset;
net = newfit(x,t,10);
net = setwb(net,0.5*(ones(10+1))); 
net.divideFcn = 'divideind';   
net.divideParam.trainInd = 1:60; 
net.divideParam.valInd   = 61:94; 
net.divideParam.testInd  = 94; 
[net, tr] = train(net,x,t);
y = net(4)
 % Manual Normalization
[x,t] = simplefit_dataset;
net = newfit(x,t,10);
net.inputs{1}.processFcns={};
net.outputs{2}.processFcns={};
[x,ps] = mapminmax(x);
[t,pt] = mapminmax(t);
net = setwb(net,0.5*(ones(10+1)));
net.divideFcn = 'divideind';   
net.divideParam.trainInd = 1:60; 
net.divideParam.valInd   = 61:94;
net.divideParam.testInd  = 94; 
[net, tr] = train(net,x,t);
y = net(mapminmax('apply',4,ps));
y = mapminmax('reverse',y,pt)
The first code uses the normalization 
function of the neural network, the second 
code turns off the normalization function 
of the neural network, and the mapminmax 
function is entered manually.

I think the result of this code should be the same. However, the result y was different. Could you tell me why?

[Greg Heath]

I will answer soon. Meanwhile, you may want to take a look at some of these in BOTH Newsreader and Answers.

Hope this is helpful.

Greg

Search using

 greg simplefit_dataset

P.S. NEWFIT is obsolete. Do you have FITNET?

[Greg Heath]

Another quickie that takes advantage of defaults

   tic,    [ x,t ] = simplefit_dataset;
   rng(0), [net tr y e ] = train(fitnet,x,t);
   plot(x,t,'b',x,y,'ro'), NMSE = mse(e)/var(t,1), toc
% NMSE =   1.7558e-05
% Elapsed time is 0.548107 seconds.

[kyeonghwan kim]

Yes, I have a fitnet.

And I mean

Case1 Using mapminmax function provided by fitnet function

 % Automatic Normalization
[x,t] = simplefit_dataset;
net = fitnet(10);
rng(0)
net.divideFcn = 'divideind';   
net.divideParam.trainInd = 1:60; 
net.divideParam.valInd   = 61:94; 
net.divideParam.testInd  = 94; 
[net, tr] = train(net,x,t);
y = net(4) % y = 7.2376

Case2 Using mapminmax directly without using the internal mapminmax function of fitnet

 % Manual Normalization
[x,t] = simplefit_dataset;
net = fitnet(10);
rng(0)
net.inputs{1}.processFcns={}; % off mapminmax
net.outputs{2}.processFcns={}; % off mapminmax
[x,ps] = mapminmax(x); 
[t,pt] = mapminmax(t);
net.divideFcn = 'divideind';   
net.divideParam.trainInd = 1:60; 
net.divideParam.valInd   = 61:94;
net.divideParam.testInd  = 94; 
[net, tr] = train(net,x,t);
y = net(mapminmax('apply',4,ps));
y = mapminmax('reverse',y,pt) % y = 7.1137

The result of 2 cases vary from 7.2376 to 7.1137.

Do these two cases have the same result?

[Greg Heath]

Do not just look at just one output point when you have 94. The best measure I can think of is the normalized mean square difference

         NMSD = mse(y2-y1)/var(t,1)

However, use the default 0.7/0.15/0.15 data division ratio and compute the differences for the trn, val and tst subsets.

Hope this helps.

Thank you for formally accepting my answer

Greg

[Greg Heath]

Thank you for this post. I really learned from it. As I result, see

http://www.mathworks.com/matlabcentral/... newsreader/view_thread/349217#955612

Thanks again,

Greg