Python Neural Network Programming

Originally published as Make Your Own Neural Network, this is an introductory book that explains the principles of neural networks clearly. It starts from the simplest ideas, first explaining how neural networks actually work: beginning with a simple predictor and a classifier, it gradually introduces concepts such as training, error, multiple nodes, backpropagation, and gradient descent, then uses matrix multiplication to connect the computation and error backpropagation of a three-layer neural network, breaking seemingly advanced mathematics into steps that can be followed by hand. Part Two moves into hands-on practice: it introduces the basics of Python and IPython, guides readers line by line in writing their own neural network code, and then trains and tests it on the MNIST handwritten digit dataset to observe how well the network recognizes handwritten digits. Part Three is a playful extension: readers input their own handwritten digits, inspect the internals of the trained network, augment training data by rotating images, and learn how to run the program on a Raspberry Pi. The appendices add a primer on calculus and instructions for setting up a Raspberry Pi environment. The book is short, heavily illustrated, and conversational in tone, making it suitable as a first book on neural networks.