Run TSC on MOSS

We provide examples for a classic traffic policy scenario: Traffic Signal Control. To run the examples provided in the moss-opt-showcases repository, you need to clone the repository and install the required dependencies.

git clone https://github.com/tsinghua-fib-lab/moss-opt-showcases.git
cd moss-opt-showcases
pip install -r requirements.txt

Then, you can use the traffic light control optimization cases as an example to run the optimization process.

Code and datasets are inside ./traffic_signal_control.

The hangzhou dataset contains 16 junctions. The manhattan dataset contains 48 junctions.

To visualize the datasets, use visualize.ipynb

To run baseline algorithms:

# Fixed time traffic signal control
python run_baseline.py --algo fixed_time --data data/hangzhou
# Max pressure traffic signal control
python run_baseline.py --algo max_pressure --data data/hangzhou
# SOTL traffic signal control
python run_baseline.py --algo sotl --data data/hangzhou
# Builtin version of fixed_time
python run_baseline.py --algo ft_builtin --data data/hangzhou
# Builtin version of max_pressure (considers phase fairness, different from above)
python run_baseline.py --algo mp_builtin --data data/hangzhou

To train DQN:

python run_dqn.py --data data/hangzhou

The training process can be visualized with tensorboard.