On Mon, Jun 15, 2009 at 5:00 PM, Trin Trin
<trin.cz@gmail.com> wrote:
Hi Alp,
- even with correctly programmed back-propagation, it is usually hard to make the net converge.
Yeah, I know, that's why we're training it until the quadratic error goes under 0.1.
- usually you initialize neuron weights with somewhat random values, when working with back-propagation.
Yeah, that'll be done too, once the algorithm will be ready. I'll provide fancy and easy functions to create a neural net just giving the numbers of layers and their sizes.
- do some debug prints of the net error while training to see how it is going
Good idea, yeah.
- xor function cannot be trained with a single layer neural net !!!
That's why there are 2 layers there, one hidden and the output one. I consider the "inputs" as ... inputs, not as a first layer of the NN.