GuangwenSi / MAMR

Code for Multi-agent reinforcement learning to unify order-matching and vehicle-repositioning in ride-hailing services

MAMR

Code for multi-agent reinforcement learning to unify order-matching and vehicle-repositioning in ride-hailing services. (https://doi.org/10.1080/13658816.2022.2119477)

Environment

The code is supposed to run in the environment:

python 3.7

CUDA 11.1

Pytorch 1.8.1

numpy 1.20.2

Instructions

step 1

Download data from https://outreach.didichuxing.com. Each taxi order consists of an order ID, pick-up/drop-off timestamps, and locations. Each driver track point consists of a driver ID, the timestamp and locations.

step 2

Preprocess the data.

python data/export_neighborhood_data.py
python data/create_city_state.py

Files generated including:

envs/driver_distribution.csv  
data/neighborhood.dill
data/city_states/city_states.dill
data/hex_bins/hex_bin_attributes.csv
data/hex_bins/hex_distances.csv

step 3

Train the model.

python train.py --num_agents ${num_agents}

step 4

Test the model.

python train.py --num_agents ${num_agents} --model_dir ${model_dir} --test True

Acknowledgement

Thanks to these repositories:

• light_mappo
• optimize-ride-sharing-earnings

Cite

@article{xu2022multi, title={Multi-agent reinforcement learning to unify order-matching and vehicle-repositioning in ride-hailing services}, author={Xu, Mingyue and Yue, Peng and Yu, Fan and Yang, Can and Zhang, Mingda and Li, Shangcheng and Li, Hao}, journal={International Journal of Geographical Information Science}, pages={1--23}, year={2022}, publisher={Taylor & Francis} }