- Portable 100% ANSI C++ code

- Developed and tested in commercial environment

- Python bindings with pickleable Genomes available through Boost.Python

- Heavily commented



Neural Networks


- Cache-friendly fast neural network class

- Real-Time Recurrent Learning

- Leaky integrators

- Hebbian learning

- Activation functions: Sigmoid, Tanh, Cubed Tanh, Gaussian, Sine, Step, Absolute value, Linear

- Visualization in Python through OpenCV (2.0 or later)





- Generational and Realtime evolution (rtNEAT)

- Novelty Search

- Subtractive (deleting) mutations

- Complexifying, Simplifying, Blended or Phased (alternating between complexifying and simplifying)

- Mutation undo mechanism which prevents defective genomes to enter the population

- Parametric multi-dimensional HyperNEAT substrates

- Genomes can have disconnected inputs for Feature Selective NEAT

- Resizable populations and varying mutation rates per population or species

- Large set of parameters covering every aspect of the algorithm