quinlan7 / dmstgcn

Dynamic and Multi-faceted Spatio-temporal Deep Learning for Traffic Speed Forecasting

This is a implementation of DMSTGCN: [Dynamic and Multi-faceted Spatio-temporal Deep Learning for Traffic Speed Forecasting, KDD2021].

Environment

• python 3.7.4
• torch 1.2.0
• numpy 1.17.2

Dataset

Step 1： Download the processed dataset from Baidu Yun (Access Code:luck) or Google Drive.

If needed, the origin dataset of PEMSD4 and PEMSD8 are available from ASTGCN.

Step 2: Put them into data directories.

Train command

# Train with PEMSD4
python train.py --data=PEMSD4

# Train with PEMSD8
python train.py --data=PEMSD8

# Train with England
python train.py --data=England