This repo implements an AI for the dice game Las Vegas based on deep-Q learning with experience replay. Currently this has only been tested with Python 3.6 using the Theano backend of Keras.
export PYTHONPATH=.
pip install -r requirements.txt
The repository contains two scripts bin/lasvegas-train and bin/lasvegas-eval.
Both can be run with --help to list the available options.
By default running lasvegas-train will train a neural network in a game against random players and save the model to model.h5.
Models can be evaluated by running lasvegas-eval -m model.h5 to produce the average score in games against random players.