Compositional Pattern Producing Networks (CPPNs) are cool, so I wrote a C# implementation and generated some! This was a just-for-fun project unrelated to any class assignments, games or competitions.
Fitness of a network was based on how well the resulting image didn’t compress via the Deflate algorithm – hence I refer to these networks as Complex Pattern Producing Networks. There’s some nifty Nyquistian nonsense where the evolved networks take advantage of how I’m sampling. GAs are so exploitative.
I was pleased with the results of this experiment because I had no expectations or goals – that’s usually a recipe for happiness. The other cool part is I’ve got all the genomes saved, so it’s trivial to expand one of the 256 x 256 images to N x M and use it as a desktop background. Genome sizes varied between 16 and 2048 bytes.