/DQN-DoubleDQN-FlappyBird

Using DQN and Double DQN to play Flappy Bird

Primary LanguagePythonMIT LicenseMIT

DQN and Double DQN to play Flappy Bird in PyTorch

Dependencies:
  • Python 3.9
  • numpy
  • torch
  • opencv-python
  • pygame
How to run:
  • Run pip3 install -r requirements.txt to install dependencies.
  • Run python main.py --restore=./checkpoint/model_double_80000.pth to test the pretrained model.
  • Run python main.py --train=True --type=dqn to train the model from the beginning. You can also increase FPS in game/flappy_bird.py script for faster training.
Flags:
  • --restore: load pretrained weight to restore training or test.
  • --logs_path: path for reward log during training.
  • --checkpoint_path: path for saving model checkpoints.
  • --train: boolean, train or not, default=False.
  • --type: dqn or doubledqn, default=doubledqn.
  • --hparam_path: path of hyperparameter .json file.