/HSTGNN

This repository is a pytorch version implementation of DEXA 2021 conference paper "Traffic Flow Prediciton through the Fusion of Spatial Temporal Data and Points of Interest".

Primary LanguagePython

HSTGNN

Inroduction

This repository is a pytorch version implementation of DEXA 2021 conference paper paper link "Traffic Flow Prediciton through the Fusion of Spatial Temporal Data and Points of Interest".

Run program

1.pakage dependencies

To run this code, you need to install the following packages:
torch==1.6.1
numpy==1.16.3
pandas==1.1.5
scipy==1.2.1
h5py==3.1.0

2.datasets

download the dataset from repository link, ZheYi Pan,KDD2019,"Urban traffic prediction from spatio-temporal data using deep meta learning"

copy the BJ_FLOW.h5,BJ_POI.h5 to data folder
-- bj_tfidf_poi.h5 is based on BJ_POI.h5 and has been processed by TF-IDF algorithm to calculate the importance of poi in each region.
-- cossimi_graph.npz is obtained from BJ_FLOW.h5 using cossine similarity to caculate the flow similarity of region pairs with a threshold to determine whether there is an edge between two regions, datails see /utils/generate_time_embedding.py, cossimi_graph serves as initial adjacent matrix to initialize the parameter of adaptive adjacent matrix.

3.Run

nohup python -u train.py > file.log 2>&1 &

If you find this repository is helpful to you, please cite our paper, thanks for your attention.