README for Traffic Light Optimization using DQN
@article{zhang2018partially,
title={Partially Observable Reinforcement Learning for Intelligent Transportation Systems},
author={Zhang, Rusheng and Ishikawa, Akihiro and Wang, Wenli and Striner, Benjamin and Tonguz, Ozan},
journal={arXiv preprint arXiv:1807.01628},
year={2018}
}
- virtualenv --system-site-packages DQN_ENV
- need to inherit pysumo
- source DQN_ENV/bin/activate
- pip install --upgrade pip
- pip install --upgrade tensorflow
- if GPU
- pip install --upgrade tensorflow-gpu
- pip install -U -r requirements.txt
python run_rltl.py
- if you run on CPU, make sure to put --cpu
- default is sumo, in order to use pysumo make sure to put --pysumo
python run_rltl.py --mode test --load [weights.hdf5]
python run_multiagents_rltl.py
- if you run on CPU, make sure to put --cpu
- default is sumo, in order to use pysumo make sure to put --pysumo
python run_multiagents_rltl.py --mode test --load [weights.hdf5]
- currently, weights are saved as ~weights_[agent id].hdf5, so when you load, remove letters after weights
- e.g. DQN_SUMO_best_weights_0.hdf5 => DQN_SUMO_best_weights