I do have a problem about mean square error (MSE) in Neural network Toolbox.

I do have a problem about mean square error (MSE) in Neural network Toolbox.

Post by Pearl Peer » Sat, 05 Feb 2005 22:15:22


Would you please help me?
I do have a problem about mean square error (MSE) in Neural network
Toolbox.

When I use
net=newff(minmax(pn),[10,1],{'tansig','tansig'},'traingd'),

to train the training samples, and test with testing samples to get
the optimal network (minimum MSE). The MSE value is a bit different.

How can I get the same value of mean square error for each time
running by the same network?

One of my friend said I should manually set initial weight and bias
as close as zero. After I tried, it turns out to be that the MSE
values of each network (different hidden node and epoch) are not
different at all. I don know what else did I miss.

Please help
 
 
 

I do have a problem about mean square error (MSE) in Neural network Toolbox.

Post by Greg Heat » Sun, 06 Feb 2005 07:52:55


The net automatically uses rand to initialize the weights


Set the random number generator to the same initial state.

help rand

Hope this helps.

Greg