npnn is a purely NumPy-based package for forward-only neural networks, intended for derivative-free population-based optimization such as genetic algorithm (GA) and evolution strategies (ES). Accordingly, its main goal is to include neural network modules that are normally incompatible with backpropagation.