HKUDS / AutoST

[WWW'2023] "AutoST: Automated Spatio-Temporal Graph Contrastive Learning"

Automated Spatio-Temporal Graph Contrastive Learning

This is the implementation of Automated Spatio-Temporal Graph Contrastive Learning (AutoST):

Requirements

Python = 3.7.13, pytorch = 1.12.0 with cuda = 10.2 and Tensorflow = 1.15.3 (crime prediction task (ST-SHN))

Data

Moblity data, crime data, census block data, POI data, house price of Chicago (https://data.cityofchicago.org/) and New York(https://opendata.cityofnewyork.us/). Also, we provide the process data and processing code for data preprocessing. And process data for traffic prediction task and crime prediction task can be loaded via the link https://drive.google.com/drive/folders/1MEi1YqB54-YzluFFp2As1aiEL_KpGXqG?usp=sharing

Data Preocessing

pre_s1.py # collecing positions of 180 (234) regions in Manhatton (a certain district) in New York (Chicago).
pre_s2.py # collecting traffic data
pre_s3.py # preprocessing the traffic data on correspongding regions
pre_s4.py # preprocessing the POI data and obtain POI-aware Region Graph
pre_poi_transformer.py # obatining the features of nodes by Transformer and Skip-gram
pre_spatial_graph.py # constructing the trajectory-based Regoin Graph and Distance-based Region Graph
pre_s5.py # constructing a hierarical graph
pre_s6_dataloader.py # obtaining the dataloder (dataset.pt file) for next step to get region embeddings

Hyperparameters

The dimensionality d of region representation is set as 96 . The depth of convolutional layers in GCN is set as 3. The learning rate is initialized as 0.0005 with the weight decay of 0.01. For the crime prediction backbone model, ST-SHN is configured with the learning rate of 0.001 and the weight decay of 0.96. The depth of the spatial path aggregation layers is set as 2. For the traffic prediction backbone model ST-GCN, the historical time window of all tests are set as 60 minutes with 12 observed data points that are utilized to forecast traffic conditions in the next 15, 30, 45 minutes. The baselines are implemented with the source code released by their original papers.

Obtaining region representations

cd data_augmentation
loader.py # adding the dataset.py file for training the model of the region representations
data_pre_4aug_fea.py   # training for obatining region representations

Crime prediction task

The code of the ST-SHN is https://github.com/akaxlh/ST-SHN

Replace the area embedding with the region represenation vetcor and run HG_ST_labcode.py.

Results of Crime Prediction task

Traffic prediction task

The code of the ST-GCN is https://github.com/VeritasYin/STGCN_IJCAI-18 ST-GCN predicts traffic in three stages, namely 15 minutes, 30 minutes, 45 minutes. Run main.py.

Results of Traffic Prediction task

House prediction task

We use Lasso Regression to perform house price evaluation task.

Citing

If our paper benefits to your research, please cite our paper using the bitex below:

@inproceedings{zhang2023automated,
title={Automated Spatio-Temporal Graph Contrastive Learning},
author={Zhang, Qianru and Huang, Chao and Xia, Lianghao and Wang, Zheng and Li, Zhonghang and Yiu, Siuming},
booktitle={Proceedings of the ACM Web Conference 2023},
pages={295--305},
year={2023}
}