Re: [computer-go] Neural Network DB Serialization?

[Peter Drake <drake@xxxxxxxxxx>]

If you're working in Java, I believe you can just declare all of the 
classes "serializable" and the system will take care of the details.  
See chapter 12 of "Core Java 2, volume 1" by Horstmann and Cornell for 
details.

If you have to do it manually, and you have a strictly feed-forward, 
layered network, I think all you have to save is one weight matrix per 
layer (not including the input layer).  This is inefficient if each 
layer is not fully connected to the next layer.

On Wednesday, January 28, 2004, at 11:48  PM, Thomas Johnson wrote:

> Does anyone want to share their methods for serializing (i.e., saving
> for later use) neural network topologies and weights, specifically to
> databases? Is there an elegant way to do this? Right now I'm thinking
> one table that contains the network's inputs, outputs, and a primary
> key; one table that contains the number of neurons in each layer for 
> the
> network; and one table that contains the weights for each neuron in 
> each
> layer. However, that seems messy to me and it feels like there ought to
> be a more elegant way of doing this. Any suggestions?
>
> Thomas
>
> --
> Thomas Johnson <thomas_johnson@xxxxxxxxxxxxxxxxx>
> _______________________________________________
> computer-go mailing list
> computer-go@xxxxxxxxxxxxxxxxx
> http://computer-go.org/mailman/listinfo/computer-go
Peter Drake
Assistant Professor of Computer Science
Lewis & Clark College
http://www.lclark.edu/~drake/

_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://computer-go.org/mailman/listinfo/computer-go