Re: Backpropagation: How Many Weights?

nevlis <[EMAIL PROTECTED]> wrote in message 
news:<[EMAIL PROTECTED]>...
> nevlis <[EMAIL PROTECTED]> wrote in
> news:[EMAIL PROTECTED]: 
> 
> > Here is my problem:  I am trying to create a small COM object that
> > will allow me to run a NN from a web-page.  The com object does not
> > have to train the the NN just run the backpropagation algorithm with
> > given weights.  
> > 
> > The group that I am working with provided me with 3288 Weights for a
> > NN that has 4 layers.  Input layer 81 nodes, hidden layer 1 32 nodes,
> > hidden layer 2 16 nodes and output layer 8 nodes.
> > 
> > To get howmany weights, I thought it would calculated this way.
> > 
> > 81 * 32 = 2592
> > 
> > 32 * 16 = 512
> > 
> > 16 * 8  = 128
> > 
> > Total Weights = 2592 + 512 + 128 = 3232
> > 
> > There is a difference of 3288 - 3232 = 56 weights 
> > 
> > Can someone tell me if I am calculating the above correctly or am I 
> > missing something.
> > 
> > Would someone no where I could find a simple algorithm that
> > would allow me to run a NN with the above specifications
> > 
> > I know the Acitvation function the user would like to use
> > is the sigmoid.
> > 
> > 
> > Thanks nevlis
> 
> Ok I know about the bias now and how it works... I read that the Bias 
> node should be set at 1.0 is this correct?

1.0 is standard. However, any reasonable constant value would work.

Hope this helps.

Greg