BlackjackSM.py is a state machine for playing blackjack. It follows a maximal set of casino blackjack rules and features a number of useful support functions for reinforcement learning.
blackjack_DQN.py contains PyTorch code for a fully connected linear deep Q-network.
blackjack_eval.py converts the models created by blackjack_DQN.py into .csv files containing the derived optimal policies.
blackjack_QN.py contains code for traditional Q-learning, for comparison.
QN_eval.py converts the model created by blackjack_QN.py into a .csv containing the derived optimal policy.
policy_score.py compares my derived policies against the theoretically optimal policy for blackjack, simulates to find the expected reward of the derived and optimal policies in our model, then writes the results to a .csv file.
The paper folder contains my writeup for this project. The files folder contains various documents collected through the course of the project.