Problem in neural training, regression plot

I need to optimize the parameters of pid controller using neural network. I am using the following code for my work.

load inp
load out

file1=inp./20;
file2=out;

p=file1';
t=file2';

H = 15; % trial No. of hidden nodes for I-H-O node topology

net = newff(minmax(p),[ H 1 ], {'tansig' 'purelin'}, 'trainlm' ); 

net.trainParam.epochs=1000;
net.trainParam.goal=0;
net.trainParam.min_grad=1e-10;
net.trainParam.mu=0.001;
net.trainParam.mu_max=1e10;
net.trainParam.show=100;
net.trainParam.showCommandLine=0;
net.trainParam.lr=0.02;

net.divideFcn='';
[ net tr Y E ] = train(net,p,t);
sim(net,p)

where "inp" and "out" data are collected from the model as input and output of pid block. inp and out are a matrix of 775 X 1 elements.

But in the regression plot obtained after running the code, I only get a cluster of data at the bottom left corner, while it should be uniformly distributed along the fit line

https://docs.google.com/file/d/0Bwn7IPXsZggeU0p5NXVHX3pHblk/edit

Kindly help..

Answer 1

[Greg Heath]

1. plot t vs p before calling newff

2. [ I N ] = size(p) = ?

3. ] O N ] = size(t) = ?

4. Neq = N*O % How many training equations?

5. Why H = 15 ?? It should be as small as possible to try to make Neq >> Nw. Otherwise you need to use a validation set or trainbr.

6. Nw = (I+1)*H+(H+1)*O % How many unknown weights?

7. Is Neq >> Nw ?

8. net = newff(minmax(p),[ H 1 ], {'tansig' 'purelin'}, 'trainlm' )

Warning: NEWFF used in an obsolete way.

 In obs_use at 18
 In newff>create_network at 127
 In newff at 102
 In Untitled0 at 19 
 See help for NEWFF to update calls to the new argument list.

9. What version of MATLAB and NNTBX do you have? (Type the command ver )

10. Why did you clutter up your code by expicitly specifying default parameters? Try net = newff(minmax(p),[H 1]) without the semicolon to see the defaults.

11. net.trainParam.lr is NOT a property of newff wnen using trainlm

12. Y is sim(net,p). Therefore don't need the last command

Hope this helps.

Thank you for formally accepting my answer

Greg