Our article and code have been updated and the new code is avaliable at https://github.com/LiangZhang1996/DataLight.
Official code for article Data Might be Enough: Bridge Real-World Traffic Signal Control Using Offline Reinforcement Learning
First, download the memory from our google drive, memory . Or you can prepare your own memory with our provided codes.
run_offline.py
, then get the well-trained models;run_test.py
to test the models on each dataset;run_cycle.py
to get the DataLight-Cycle model;summary.py
to get the performance of each model.run_offline.py
to use different offline data.run_offline.py
to use different amounts of offline data.DataLight/generate_offline_data/
for generating offline data.For baseline methods, refer to Advanced_XLight and DynamicLight.
This project is licensed under the GNU General Public License version 3 (GPLv3) - see the LICENSE file for details.