qlemaire22 / reinforcement-learning-flappybird

Reinforcement Learning applied to the game Flappy Bird with TensorFlow.

Reinforcement Learning - Flappy Bird

Project of Deep Learning in Data science (KTH DD2424):

• Quentin Lemaire
• Quentin Vecchio
• Kevin Yeramian

Installation on Ubuntu

1. sudo apt-get update
2. sudo apt-get install python python3-pip (You can also use python2.7)
3. sudo apt-get install git
4. git clone https://github.com/ntasfi/PyGame-Learning-Environment
5. cd PyGame-Learning-Environment
6. pip install -e . && cd ..
7. pip install future tensorflow opencv-python (You can choose between tensorflow and tensorflow-gpu)
8. pip install pygame

Run training

python training.py

Run agent

python play.py

By default, the script play.py use the old network model. If you want to use your own model (trained with the training.py script) you have to edit the script play.py to use the Network class and not the NetworkOld class.

References

This work is based on the code of Yen-Chen Lin (https://github.com/yenchenlin/DeepLearningFlappyBird/blob/master/deep_q_network.py).