pengzhangzhi/spatial-temporal-transformer

traffic flow prediction

Overview

The pytorch implementation of paper: Spatial-temporal Transformer Network with Self-supervised Learning for Traffic Flow Prediction.

Quick start

After download the repo and necessary dataset, you need to first generate training data and then lunach a training.

• clone this repo

git clone https://github.com/pengzhangzhi/spatial-temporal-transformer

• install packgaes

pip install -r requirements.txt 

• Download TaxiBJ dataset and put it in the path spatial-temporal-transformer/data/TaxiBJ/. You only need to download BJ16_M32x32_T30_InOut.h5, the rest raw files are aready in the folder spatial-temporal-transformer/data/TaxiBJ/.
• TaxiNYC dataset is already in the repo, you do not need to download it.
• generate training data
  • generate TaxiNYC training data: python prepareDataNY.py -c TaxiNYC.json .
  • generate TaxiBJ training data: python prepareDataNY.py -c TaxiBJ.json .
  • NOTE: make sure the config file TaxiBJ.json are in the config folder.
• lunach training: python train.py -c TaxiBJ.json, to train the model followed the hyper-parameters in the TaxiBJ.json file.

Citation

If you would like to use the code please cite my paper.