I am currently using this site as a primer - not python, but a good feed-forward network example and pretty straightforward to follow.
At the same time I have been reading The Essence of Neural Networks by Robert Callan (ISBN 0-13-908732-X) which has a wide range of network architectures and applications and is an easy read.
Since you mention python I should direct you to this IBM site, which I found very useful and the underlying code is in python here. Citation should go to Neil Schemenauer.
I should also mention that I took the python code and ported it to numpy because it ran very slowly. I was unsuccessful, but before I rubbish numpy I have to say that I suspect my implementation was not very good and I'm sure there is a vectorised way of doing forward passes and backpropagations, I just didn't find it.
What I have ended up doing is implementing in Java by a simple port of the python code. This only took an hour or so and it runs about 100 times faster. I think this is more proof that I don't know what I'm doing with numpy, but if you are starting from scratch I would question whether raw python is the right language for you. You may be better coding this sort of thing in C or C++ if you have to use python.
Best of luck.
