evo

Artificial intelligence for the game of Go using neural networks and genetic algorithms

This is an experiment to see if genetic algorithms can be used to evolve a neural network for playing the game of Go. I don't expect this to be a good program anytime soon (or ever), I just want to play around.

Requirements

• GnuGo
• GoGui
• brown
• AmiGoGtp
• gcc (or compatible, like clang)
• make
• ruby

Installation

1. Clone the repository
2. Set up the submodules using git submodule init && git submodule update
3. Build everything using make
4. Run the tests using make test

Running the evolution of the neural net

1. Ensure that you have the Ruby version installed as specified in .ruby-version (or use rvm or similar to do it automatically).
2. Execute ./runner EXPERIMENT_NAME and answer the setup questions
3. If you quit you can just restart the experiment with the same command

Running brown against itself

BLACK="./engine/evo"
WHITE="./engine/evo"
REFEREE="gnugo --mode gtp"
TWOGTP="gogui-twogtp -black \"$BLACK\" -white \"$WHITE\" -referee \"$REFEREE\" -games 10 -size 9 -alternate -sgffile evo"
gogui -size 9 -program "$TWOGTP" -computer-both -auto