Thread Subject: how to specify initial weight & bias

I am a student who work on prediction of protein concentration using neural network. When i'm using mlp/feedforward neural network, the result vary every time I run the simulation. Someone said that it happen because in the training, the initial value for the weight and bias are set randomly (default) by the Matlab. I can specify it if i want. Does anyone know how to specify the weight & bias?


AR=[11]; % [hidden node]
PR=[minmax(P)];
net=newff(P,T,AR,{'tansig' 'purelin'}, 'trainlm');
net.iw{1,1}
net.b{1}

On Nov 4, 3:40 am, "aurikel Radzali" <suriar...@gmail.com> wrote:
> I am a student who work on prediction of protein concentration using neural network. When i'm using mlp/feedforward neural network, the result vary every time I run the simulation. Someone said that it happen because in the training, the initial value for the weight and bias are set randomly (default) by the Matlab. I can specify it if i want. Does anyone know how to specify the weight & bias?

Yes. However stick to the MATLAB default. To reproduce runs
reset the random generator to the same state.

state0 = 2009
rand('state',state0)
> AR=[11];   % [hidden node]
> PR=[minmax(P)];
> net=newff(P,T,AR,{'tansig' 'purelin'}, 'trainlm');

net = newff(P,T,AR); % Using defaults

> net.iw{1,1}
> net.b{1}

net.lw{2,1}

Hope this helps.

Greg

Greg Heath <heath@alumni.brown.edu> wrote in message <6a37872f-758d-4926-a45d-6b3ff5b7fc50@h2g2000vbd.googlegroups.com>...
> On Nov 4, 3:40?am, "aurikel Radzali" <suriar...@gmail.com> wrote:
> > I am a student who work on prediction of protein concentration using neural network. When i'm using mlp/feedforward neural network, the result vary every time I run the simulation. Someone said that it happen because in the training, the initial value for the weight and bias are set randomly (default) by the Matlab. I can specify it if i want. Does anyone know how to specify the weight & bias?
>
> Yes. However stick to the MATLAB default. To reproduce runs
> reset the random generator to the same state.
>
> state0 = 2009
> rand('state',state0)
> > AR=[11]; ? % [hidden node]
> > PR=[minmax(P)];
> > net=newff(P,T,AR,{'tansig' 'purelin'}, 'trainlm');
>
> net = newff(P,T,AR); % Using defaults
>
> > net.iw{1,1}


Thanks, it really help..
> > net.b{1}
>
> net.lw{2,1}
>
> Hope this helps.
>
> Greg