KL4805 / CrossTReS

Repo for CrossTReS: Cross-city Transfer Learning for Traffic Prediction via Source Region Selection

CrossTReS

This is the repo for paper "Selective Cross-City Transfer Learning for Traffic Prediction via Source City Region Re-Weighting", KDD 2022.

Step 1: Data

Go to data repo and unzip the crosstres_data.zip file.

Step 2: Run the scripts in src

The structures of src are as follows:

• model.py: Contains implementation of base models.
• utils.py: Necessary utility functions.
• run_crosstres.py: The implementation of CrossTReS. The requirements are:
  • Python=3.8
  • PyTorch=1.9.0
  • DGL=0.6.1
  • sklearn
• run_crosstres_rt.py: The implementation of CrossTReS which uses RegionTrans for fine-tuning.
• gen_rt_dict.py: This script generates the dictionary for RegionTrans to do matching.

You can check the tunable parameters in run_crosstres.py and run_crosstres_rt.py.

Note: Runningrun_crosstres.py requires approximately 10GB GPU memory with batch_size=32. You can reduce batch_size to reduce memory cost.

Procedures to run the scripts

• run_crosstres.py: python run_crosstres.py --SET_PARAMETERS.
• run_crosstres_rt.py:
  • First, run python gen_rt_dict.py --metric poi --source [NY, CHI] --target [DC]. You will get a file under the src/rt_dict folder.
  • Then, run python run_crosstres_rt.py --SET_PARAMETERS --rt_dict poi.