ghl3 / alphafour

ConnectFour AI

AlphaFour

Scripts and notebooks for playing, simulating, and building AIs for the game ConnectFour

There are two main components to this repository:

• The four.py script, which allows you to play ConnectFour, as well as simulate games and process data
• The training.ipynb, which generates ConnectFour "AI" models using TensorFlow

Four.py

four.py is a command-line script that provides the following capabilities:

This repository is designed to support the following steps

• play: Play a game of ConnectFour again an existing AI
• simulate: Simulate a game between AIs and save it as data
• visualize: Visualize a saved game
• process: Process a saved game into a data format suitable for model training

Training.ipynb

An jupyter notebook that reads in the processed form of games simulated using four.py, trains tensorflow models on them, and serializes the models out. A serialized tensforflow AI can then be used to play games or simulate new games using four.py

Typical Workflows:

Simulate games and process them into training data using as "RANDOM" AI

python four.py simulate --num-games 500000 --output-prefix simulations/random-`date '+%Y-%m-%d-%H:%M:%S'`/
python four.py process --output-prefix training_data/random-2017-10-21-13:41:47/ "simulations/random-2017-10-28-17:13:04/*.json"

Simulate games and process them into training data using an AI built with Tensorflow

python four.py simulate --num-games 100000 --red-model gen1-cov2d_beta_2017_10_29_150829 --yellow-model gen1-cov2d_beta_2017_10_29_150829 --output-prefix simulations/gen1-cov2d_beta_2017_10_29_150829-`date '+%Y-%m-%d-%H:%M:%S'`/
python four.py process --output-prefix training_data/gen-1-cov2d_beta_2017_10_22_142925/ "simulations/gen-1-cov2d_beta_2017_10_22_142925/*.json"

Play against our built and serialized AI model

python four.py play --ai gen2-cov2d_beta_2017_11_05_114919 --player-first false