$\pi$ -Light: Programmatic Interpretable Reinforcement Learning for Resource-Limited Traffic Signal Control AAAI 2024
code for
- python=3.8.10
- torch=1.7.1+cu110
- numpy=1.21.2
- CityFlow=0.1.0
You need to install a modified version of CityFlow to run the code.
Then you need to unzip the data file.
python 02_run_MCTS.py --dataset=Jinanpython 02_run_MCTS.py --dataset=Hangzhou1 --generalization=True target=Manhattanpython 00_run_tiny_light.py --dataset=Jinanpython 01_run_baseline.py --dataset=Jinanpython 015_baseline_transfer.py --dataset=JinanRun VIPER
We also compare imitation learning-based VIPER, which distills the neural policy into a decision tree. We utilized MPLight as a teacher to generate state-action pairs for training the decision tree. Overall, VIPER's performance is close to that of MPlight.
python 03_run_viper.pyThis codebase is based on TinyLight's code.