Simple (<100 loc) implementation of a feed forward neural network in python
Requirements: numpy and pyplot (only for unit tests and examples)
It is possible to specify the number of layers and the number of neurons of each layer upon creation as well as an activation function together with its first derivative (which is only needed when learning).
The code is fully vectorized and takes advantage of numpy's arrays to perform both forward-propagation and back-propagation, but information exchange with the outside world is done through normal python lists.
A simple unit test is provided; it builds a neural network capable of computing the exclusive-or between its arguments. The network has one hidden layer composed of three neurons (including the bias term) and both forward and back propagation are tested. The activation function used is the logistic function, also known as sigmoid. The learning test plots a graph using pyplot.