Home

Awesome

[CIKM 2021 Best Resource Paper Runner-Up] DL-Traff: Survey and Benchmark of Deep Learning Models for Urban Traffic Prediction

DL-Traff-Graph: Graph-Based Deep Learning Models for Urban Traffic Prediction

Cite

@inproceedings{jiang2021dl,
  title={DL-Traff: Survey and Benchmark of Deep Learning Models for Urban Traffic Prediction},
  author={Jiang, Renhe and Yin, Du and Wang, Zhaonan and Wang, Yizhuo and Deng, Jiewen and Liu, Hangchen and Cai, Zekun and Deng, Jinliang and Song, Xuan and Shibasaki,
  Ryosuke},
  booktitle={Proceedings of the 30th ACM International Conference on Information \& Knowledge Management},
  pages={4515--4525},
  year={2021}
}

Introduction

English | 简体中文

DL-Traff is an open resourse project which offers a benchmark for traffic prediction on grid-based and graph-based models. DL-Traff-Graph is a part of graph-based project. This main branch works on Pytorch1.6. Different versions of Pytorch vary slightly in training time and performance. In this github, we integrate two traditional statistical methods(HistoricalAverage and CopyLastFrame), one time series models (LSTNet) and a large number of graph models into one platform. We maintain that all models are based on the same data processing, the same hyperparameters, and the same computing environment such as the version of Pytorch and Cudnn. Although this makes the models fail to achieve the final convergence effection, the performance of different network architectures under the same conditions will be fully reflected by our experiment. We will update the optimization results of each model in later work.

Installation Dependencies

Working environment and major dependencies:

Public data and models zoo

Datasets

Models

Components and user guide

Content

User guide

Download this project into your device, the code project will be downloaded into the current path where you type this powershell command:

git clone git@github.com:deepkashiwa20/DL-Traff-Graph.git

Use the STGCN model on METRLA dataset as an example to demonstrate how to use it.

cd /PEMSBAY
unzip pems-bay.h5

cd /workMETRLA

# Debug the model on video card 1 :
python STGCN.py 1

# Run the main program to train, prediction and test on video card 1:
python pred_STGCN3.py 1

# View the result after the operation is complete.
cd /save/pred_STGCN_METRLA_21061600