Making a Neural Network

Hello everyone and thanks for your time!

I'm trying to build a neural network in order to get a relation between some physical properties of a material. The code I've been using is the following:

load ./DataFinalModel/data.txt

% 1st column is (D) % 2nd column is density % 3rd column is rest compressio % 4th column is rest flexio % 5th column is mod young

Nsamples = length(data);

%% Structure the data

m1 = data(1:Nsamples,2); m2 = data(1:Nsamples,3); m3 = data(1:Nsamples,4); m4 = data(1:Nsamples,5); r1 = data(1:Nsamples,1); %% Build the network

p = [m1';m2';m3';m4'];

t = [r1'];

neuronstotal = 10;

transferFunction = 'purelin';

hiddenLayers = 1;

net = newff(p,t,neuronstotal);

net.trainFcn = 'trainscg';

net.trainParam.show = 10;

net.trainParam.lr = 0.1;

net.trainParam.epochs = 100;

net.trainParam.goal = 1e-10;

net.trainParam.min_grad = 1e-15;

net.layers{1}.transferFcn = transferFunction;

net.performFcn = 'mse';

net.outputs{2}.processFcns = {'mapminmax'};

net.inputs{1}.processFcns = {'mapminmax'};

% Test data

index = 2:6:116;

m1t = data(index,2);

m2t = data(index,3);

m3t = data(index,4);

m4t = data(index,5);

r1t = data(index,1);

ptest = [m1t';m2t';m3t';m4t'];

ttest = [r1t'];

test = sim(net,ptest);

rel_err1 = (r1t' - test)./r1t' * 100;

rms_relerror1 = sqrt(sum(1/20*rel_err1.^2))

mean1 = mean(rel_err1)

std1 = std(rel_err1)

xaxis = 1:1:20;

I need the weights and the bias matrix, but I cannot get'em. Also, I need the mapminmax output.

Again, thank you for your time.

Answer 1

[Greg Heath]

1. NEWFF is obsolete . What version of MATLAB and NNTBX are you using? Type the command

ver

2. You have made the solution too complicated. See the examples in the documentation

   help newff;    doc newff

In particular,

a. Understand what properties are defaults
b. Practice on a MATLAB nndataset with dimensions similar to yours  
   before using your own data set

     help nndatasets

 c. Start by using all of the defaults. Then change if needed.

3. Too much transposing. Only have to transpose once:

 data = data';
 t = data(1,:);
 p = data(2:5,:);
 whos                                   % check

4. You never trained your net.

5. Find code examples in the NEWSGROUP or ANSWERS by searching on

 greg newff

Hope this helps.

Thank you for formally accepting my answer

Greg

Greg

Answer 2

[Sergio]

Thank you very much!