CS-GY 9223: Deep Learning Project
This project uses a modified version of RLCard for the RL environments. You can find the RLCard fork here
Contents
Implementations
DQNAgent.py
: DQN agentfully_connected.py
: fully-connected feed forward network with ReLU activationsmemory.py
: Memory buffers used during RL
Kuhn Poker
kuhn_poker_cfr_train.ipynb
: Train CFR agentkuhn_poker_dqn_train.ipynb
: Train DQN agentkuhn_poker_dqn_analyze.ipynb
: Examine the moves the DQN agent makeskuhn_poker_eval.ipynb
: Evaluate the agents against each other
Leduc hold'em
leduc_holdem_cfr_train.ipynb
: Train CFR agentleduc_holdem_dqn_train.ipynb
: Train the DQN agentleduc_holdem_dqn_analyze.ipynb
Examine the moves the DQN agent makesleduc_holdem_eval.ipynb
: Evaluate the agents against each other
Limit hold'em
limit_holdem_eval.ipynb
: Evaluate the agents against each otherlimit_holdem_dqn_train.ipynb
: Train the DQN agentlimit_holdem_human_dqn.py
: Play against the DQN agent
Other
vis.ipynb
: Visualizations for the reportimages/
: Folder containing images of visualizations for the reportspec-file.txt
: List of Conda dependencies for this project's environment
Results
The above notebooks and scripts reference files in the models
and experiments
folders. You can find these folders on Google Drive.