c4ai

A reinforcement learning artificial intelligence system used to play the game of connect-4.

How good it is

Good enough to smash all my friends, myself included pretty much all the time.

Bad enough to lose to a perfect player sometimes.

schwi.py might not work on every system. This problem is a huge pain and I'm not interested in fixing it at the moment.
Specifying training arguments in do_selfplay of selfplay_v2.py doesn't change anything. I'm too lazy to do do that right now. Will definitely work on it later though...

training_pipeline.py will not use selfplay_v2.py by default. However, I have managed to (and so can you) build the binary on Linux. Use onnxruntime version 1.2.0 (see here) and recent version of GSL.
If you have something to add, cough it up.

Name		Name	Last commit message	Last commit date
Latest commit History 71 Commits
Server		Server
standalone		standalone
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
c4game.py		c4game.py
dnn.py		dnn.py
dnn2.py		dnn2.py
mcts.py		mcts.py
mcts_v2.py		mcts_v2.py
onnx_converter.py		onnx_converter.py
play_vs_ai.py		play_vs_ai.py
play_vs_ai_ponder.py		play_vs_ai_ponder.py
schwi.py		schwi.py
selfplay.py		selfplay.py
selfplay_v2.py		selfplay_v2.py
training_pipeline.py		training_pipeline.py