An implementation of an Asyc DQN to play Space Invaders using TensorFlow, Keras and Open-AI gym
Written in Python2
- Tensorflow
- Keras
- Gym
- numpy
- Gym-Retro
Git clone the repository into a directory of your choosing
- Inside of AsyncDQN.py you may specify the number of agents used in training, and whether or not to render each game
- As well, parameters such as how many episodes per summary saving/ checkpoint saving may also be altered
- Inside of the directory , run python AsyncDQN.py through shell or other means.
- If you have python 3 as default, run python2.
To come, finished implementation, training and hyperparameter tuning has begun
To come
The creation of this model was made possible through these resources:
- https://www.intel.ai/demystifying-deep-reinforcement-learning/#gs.CR036w2s
- http://videolectures.net/rldm2015_silver_reinforcement_learning/
- https://arxiv.org/abs/1602.01783
- Hands on Machine Learning with Scikit-Learn & Tensorflow, Chapter 16
- https://github.com/coreylynch/async-rl