Right now I’ve got a LSTM cloud-of-neurons RNN that’s just about ready to go, along with a generic DNA setup for genetic mixing of whatever. My testing program will be to attempt to evolve (via crossbreeding!) many populations of these networks to play Go. Neural network traditionally do pretty badly at Go so I’m curious to see how far my particular design can get. (C++11)
Downside: I don’t yet have the ability to actually score the Go board, which is pretty important for population ranking. I’ve looked up some algorithms and the general consensus is “it’s hard”. At least I can get all eyes, liberties, and groups, and used the “draw a Go board” part of the program to learn the basics of wxWidgets.