Training a Deep-Q Network to play retro games for the TRS-80. Work-in-progress.
Requires Python 3 and gcc installation.
Make sure you have a development version of freeglut installed
E.g.: sudo apt install freeglut3-dev
.
pip install pillow
pip install pyglet
pip install scikit-image
pip install keras
pip install tensorflow
pip install h5py
git clone --recursive https://github.com/apuder/trs-dqn.git
cd trs-dqn
export LD_LIBRARY_PATH=`pwd`
make
Let a human user play the game:
python main.py -m Play
Train the network. The optional parameter --no-ui
disables the UI during training:
python main.py -m Train [--no-ui]
Play a game by using the trained network:
python main.py -m Run --model=<model-file.h5>
Based on https://keras.io/examples/rl/deep_q_network_breakout/