How to store R-value and MSE value in workspace while using Neural Network

Hi, I'm working on developing a neural network with 2 input variables and 1 output. I'm a beginner in Neural Networks, so I tried to use 'nftool' GUI to setup and train the network. Once I was done with one training session, I generated an 'M-file' and was able to successfully execute the same training. Now, the problem that I face is that, I'm not sure how to save the 'R' Value and the 'MSE' value to the workspace. My Matlab version is 2009b. Please advise.

Answer 1

[Greg Heath]

If you use the syntax

[ net tr ] = train(net, x, t);

Most of the important training info is in the structure tr. Type, without the ending smicolon,

tr = tr

Hope this helps.

Thank you for formally accepting my answer.

Greg

[Raghu]

Hi Greg, Thank you very much for your response. I tried saving this structure 'tr' and I couldn't find the values for 'R' and "MSE' inside this structure. Please help!

Thanks again, Raghu

[Greg Heath]

MSEtrn = tr.perf(end)

MSEval = tr.vperf(end)

MSEtst = tr.tperf(end)

The Rs aren't included because they aren't very important. You will have to get those by fitting the ys vs ts to straight lines.

[chaudhry]

Hi Greg, thank you very much for the good answer and how to reduce MSE value and kindly explain which is more important MSEtrn or MSEval or MSEtst. which MSE value is acceptable.

[Greg Heath]

http://www.mathworks.com/matlabcentral/newsreader/view_thread/331993#912331

Naïve Constant Output Model

 y00 =  repmat(mean(t,2),1,N);
 e00 =  t-y00;
 MSE00 = mse(e00)
 MSE00 = mean(var(t',1)) % Surprised? ... Think about it.

Neural Net Model

 [ net tr y e ] = train(net,x,t);
 % y = net(x);
 % e = t-y;
  MEANe = mean(e)          % 0
  MSE   = mse(e)
  MSE   = mean(var(e',1))  % Surprised? ... Think about it.
  NMSE  = MSE/MSE00        % Normalized MSE
  R2    = 1-NMSE       % R^2, coefficient of determination (see Wikipedia)
  R     = sqrt(R2)     % Surprised?

Individual trn/val/tst contributions can be deduced using the indices in tr.

Greg