This is a hobby project with a goal of creating a simple version of AlphaZero self-learning algorithm for chess.
Current status:
- Self-learning AI for connect 4 is very strong
- Same algorithm used for chess is not yet working, needs debugging
- Fast chess move generation: >200M moves/s (see chess/perft.cpp, perft)
- ML parts are very difficult to compile, requires compatible version combination of tensorflow, cuda, cudnn and gcc/clang.