This package contains code to do NeuroEvolution of Augmenting Topologies, or
NEAT.  NEAT is a direct-encoded graph representation meant to encode neural
networks and other graph structures.

The ECJ NEAT code is in the first of two planned stages.  In the first stage,
the code is meant to work more or less exactly the same as the original NEAT
C++ code here:

	https://github.com/FernandoTorres/NEAT/

In this stage, the NEAT implementation is not ECJ-ish at all: it has a
custom breeder, no breeding pipelines, a custom evaluator and initializer,
and so on.  It's single-threaded breeding only.

In the planned second stage, we will carefully make NEAT more ECJ-like.  This
will take work and time as NEAT has a lot of custom and unusual features and
warts.  

You can run NEAT with two applications, both in ec/app:  xor and cartpole

The code here is:

NEATDefaults.java		Package defaults
NEATIndividual.java		Subclass of IntegerVectorIndividual with an
				additional list of "node" genes.
NEATSpecies.java		Special species for NEATIndividual, does a lot
				of the custom heavy-lifting in the package
NEATNode.java			Node genes
NEATGene.java			Edge ("Link") genes
NEATSubspecies.java		NEAT "species" speciation groups.  Called
				Subspecies to differentiate them from ECJ
				Species objects
NEATInnovation.java		Handles NEAT-style "innovation" information
NEATNetwork.java		Phenotype arising from the Node/Gene genotype
NEATInitializer.java		Custom initializer, loads the population from
				a single seed individual in a file.
NEATBreeder.java		Custom breeder, handles speciation and special
				breeding mechanisms.


Parameter files:

neat.params			Abstract top-level parameters for NEAT
