Re: Netlab on Octave

[Wilson de Oliveira]

Tatsuro, great!  Thanks for your effort! Much appreciated!

 

After your help with the Octave right version and discovering that the real problem was with the uicontroI function which creates user interface control, I  went google´ing and found out that the subject of importing gui related MATLAB functions is a trick one; particularly those defining an user interface, making wide use of commands like uicontrol and uimenu.  I found the sourceforge project  Octawrap which I am about to try which aims at resolving this incompatibility.

 

I am also going to try later the Octave extra package Java Graphic Handles at  sourceforge.

 

 

Many thanks for help and patient.

Best regards

Wilson de Oliveira
UFRPE-DEInfo
http://www.cin.ufpe.br/~wrdo

2009/9/10 Tatsuro MATSUOKA <address@hidden>

Hello Wilson de Oliveira

--- Wilson de Oliveira wrote:

> Ok. Sorry.
>  For a start, Netlab is a Matlab toolbox for Pattern Recognition which
> includes amongst others models Neural Networks. It is freely available at:
> http://www.ncrg.aston.ac.uk/netlab/index.php
> demnlab.m is a demonstration program which runs several models in gui
> interface.
>
> As regard the version of the octave, thanks for your help! The octave 3.0.0
> came with qtoctave - a gui interface for octave; Now that I´ve installed the
> 3.2.2 version available at sourceforge.net I got fewer errors:
>
>  > demnlab
> error: `uicontrol' undefined near line 33 column 3
> error: called from:
> error:   C:\Octave\Netlab\demnlab.m at line 33, column 3
> octave-3.2.2.exe:6:C:\Octave\Netlab
> >
>
> Thanks for your help
> Wilson de Oliveira
> UFRPE-DEInfo
> http://www.cin.ufpe.br/~wrdo
>
>

Unfortunately Octave does not support GUI graphic control.
You should control graphics with graphic handle functions.
Please  see section 15 "Plotting" in octave manual.
You should read care fully "15.1 Plotting Basics" band "15.2 Advanced Plotting".
Perhaps these types functions are compatible to MATLAB.


I have tried a simple example in web page.

ndata=20
% Generate the matrix of inputs x and targets t.
x = [0:1/19:1]';
t = sin(2*pi*x) + 0.2*randn(ndata, 1);

% Set up network parameters.
net = mlp(1, 3, 1, 'linear')

net =
{
 type = mlp
 nin =  1
 nhidden =  3
 nout =  1
 nwts =  10
 outfn = linear
 w1 =

   -1.148633   1.104356   0.049478

 b1 =

   -0.070054   0.902782   1.349561

 w2 =

    0.17895
    0.27658
    0.36492

 b2 =  0.66142
}
****************
% Set up vector of options for the optimiser.
options = zeros(1,18);
options(1) = 1;
options(9) = 1;
options(14) = 100;

% Train using scaled conjugate gradients.
[net, options] = netopt(net, options, x, t, 'scg');
error: `fcnchk' undefined near line 68 column 5
error: called from:
error:   D:\usr\Tatsu\program\Octave\netlab\scg.m at line 68, column 3
error:   D:\usr\Tatsu\program\Octave\netlab\netopt.m at line 36, column 17

Octave does not support `fcnchk'
But I found
http://en.wikibooks.org/wiki/MATLAB_Programming/Differences_between_Octave_and_MATLAB

********
If something (like Netlab) need a function called fcnchk you can just put the following into a file
called fcnchk.m and put it somewhere Octave can find it:
function f=fcnchk(x, n)
 f = x;
end
********
I placed fcnchk.m under the same folder of netlab

SO I get

Checking gradient ...

  analytic   diffs     delta

 1.1904e+000  1.1904e+000  -1.3890e-009
 4.8963e-001  4.8963e-001  8.2556e-010
 1.1238e+000  1.1238e+000  -5.8918e-010
 2.3292e+000  2.3292e+000  -1.3039e-010
 1.1362e+000  1.1362e+000  -1.2848e-009
 1.8523e+000  1.8523e+000  6.3412e-010
 -1.3883e+001  -1.3883e+001  -1.9210e-010
 2.0438e+001  2.0438e+001  5.2320e-010
 1.9960e+001  1.9960e+001  -9.0928e-011
 2.2659e+001  2.2659e+001  -8.4377e-010
Cycle    1  Error    4.629483  Scale 1.000000e+000
Cycle    2  Error    4.629483  Scale 5.000000e-001
Cycle    3  Error    4.629483  Scale 2.000000e+000
Cycle    4  Error    4.259136  Scale 8.000000e+000
Cycle    5  Error    4.050134  Scale 4.000000e+000
Cycle    6  Error    3.775366  Scale 2.000000e+000
Cycle    7  Error    3.390233  Scale 2.000000e+000
Cycle    8  Error    3.288238  Scale 1.000000e+000
Cycle    9  Error    3.286530  Scale 5.000000e-001
Cycle   10  Error    3.235898  Scale 2.500000e-001
Cycle   11  Error    3.212532  Scale 1.250000e-001
Cycle   12  Error    3.192542  Scale 6.250000e-002
Cycle   13  Error    3.187294  Scale 3.125000e-002
 :
 :
*****************************************
% Plot the trained network predictions.
plotvals = [0:0.01:1]';
y = mlpfwd(net, plotvals);
plot(plotvals, y, 'ob')

gives a plot like attatchment files.

********
Honestly I am not familiar with this type of work.
In addition, I have no experience of MATLAB

Hopefully users who use both MATLAB and octave give you much better suggestions.

The aboves are all what I can do for you.

Good luck!!

Regards

Tatsuro

--------------------------------------
Thanks 10 years!  Yahoo! Shopping and Yahoo! Auctions
http://pr.mail.yahoo.co.jp/ec10years/