Triple-BERT

Article: Zijian Zhao, Sen Li*, "Triple-BERT: Do We Really Need MARL for Order Dispatch on Ride-Sharing Platforms?", 2026 International Conference on Learning Representations (ICLR), 2026

1. Workflow

2. Dataset

The dataset used in this study is derived from the yellow taxi data in Manhattan.

The processed data can be found in the ./data directory.

Considering the copyright, we have removed the processed data. However, the data processing code is available in the ./data directory. Please download the dataset from the link provided above and use our code to process it.

3. How to Run

3.1 Stage 1: IDDQN

To run Stage 1, execute the following command in the ./pretrain directory:

python train.py --bi_direction 

3.2 Stage 2: TD3

To run Stage 2, execute the following command in the ./finetune directory:

python train.py --bi_direction --pretrain_model_path <path_to_trained_model_from_stage_1>

4. Model Parameters

The model parameters and training log files for both Stage 1 and Stage 2 are located in the ./parameters directory.

5. Reference

@article{zhao2025triple,
  title={Triple-BERT: Do We Really Need MARL for Order Dispatch on Ride-Sharing Platforms?},
  author={Zhao, Zijian and Li, Sen},
  journal={arXiv preprint arXiv:2510.03257},
  year={2025}
}