Play Flappy Bird by RL

this is a implement of DQN and Actor-Critic in game Flappy Bird
we use pytorch in python3.5

Requirement

pytorch
pygame
PIL

Usage

the main hyper-parameters are set in the config.py, gpu=-1 means use cpu

train

train dqn model:
python3 main_dqn.py

train actor-critic model:
python3 main_ac.py

note:the model will be saved each 100 episodes

test

train dqn model:
python3 test_dqn.py

train actor-critic model:
python3 test_ac.py

note: in test_*.py, you must set which episode_num of saved model should be load to be tested

Disclaimer

this work use the basic game env in the repo yenchenlin/DeepLearningFlappyBird