how to initialize a network using nlarx

Hi, i'm using matlab 2010a and this is what i want to do:

======

    nnett=newff(zeros(0,10),rand(1,10),[2]);
    %other properties
    for i=1:maxiter
    m{i}=nlarx(data,[2 2 0],neuralnet(nnett));
    end
    %choose the best model

=====

The problem is that at each iteration the network has always the same initial weights (and obvioulsy they all give the same result). I've tried to set

=====

nnett.Network.layers{2}.initFcn='rands'

=====

but it gives this error:

=====

Error in ==> nlarx at 370 ??? Error using ==> rand

Inputs must be numeric.

=====

Any ideas?

Thanks

Answer 1

[Walter Roberson]

    %other properties
    for i=1:maxiter
    nnett=newff(zeros(0,10),rand(1,10),[2]);
    m{i}=nlarx(data,[2 2 0],neuralnet(nnett));
    end
    %choose the best model

[Reu]

Thanks.

But doing that, nnett looses all its properties.

This should work:

for i=1:maxiter

nnett=newff(zeros(0,10),rand(1,10),[2]);

%other properties

m{i}=nlarx(data,[2 2 0],neuralnet(nnett));

end

%choose the best model

but it's not very clean. There isn't something like init(nnett) or someway to let nlarx initialize the network randomly?

Thx again.