Several PDP++ questions

[Lucas Meijer <mach82@mach8.nl>]

Hi List,

  I've got several things I'd like to find out. I'm learning to work with
PDP, and I've ran into some questions:

- my network looks like this: 3 input nodes, 2 hidden nodes, 2 output
nodes. In all the patterns I feed the network, the output merely depends
on wether input node #1 is positive or negative. I'd like the network to
recognise this.

I think I've succeeded in this. My network is now trained. When I click
either of the hidden nodes, and have 'view' selected, and 'r.wt' it shows
that the irrelevant input nodes have neglectible weights, and that the
relevant input node has a heavy weight. Hooray!

What I don't understand is this: when I select one of the irrelevant input
nodes, and have 's.wt' selected, it shows that these units have non
neglectible weights going to the hidden layer.

I thought r.wt and s.wt display the same thing. Is this incorrect? What
does s.wt reflect?



Question 2:

I'd like to use the network I've trained with PDP++ in a game I'm
developing. (eventually, now I'm just learning). Is there a way for PDP++
to output the code it is using to evaluate a test pattern?

Is there a way I can export the weights from the network, so I can use
them in my own application? I've looked at the ouput of a saved network,
but I was hoping on something with only the network layout and the
weights.


Thanks for reading this far :) Any insights or links to
tutorials/examples/anything that would make my learning go faster would be
greatly appriciated.

Thanks, Lucas