TrafficStream

Code for TrafficStream: A Streaming Traffic Flow Forecasting FrameworkBased on Graph Neural Networks and Continual Learning（IJCAI 2021). TrafficStream is a streaming traffic flow forecasting framework based on Graph Neural Networks (GNNs) and Continual Learning (CL), achieving accurate predictions and high efficiency.

Requirements

• python = 3.8.5
• pytorch = 1.7.1
• torch-geometric = 1.6.3

conda env create -f trafficStream.yaml

Data

Download raw data from this, unzip the file and put it in the data folder

Usages

• TrafficStream-STModel

python main.py --conf conf/trafficStream.json --gpuid 1

• Expansible-STModel (lower bound):

python main.py --conf conf/expansible.json --gpuid 1

• Retrained-STModel (upper bound):

python main.py --conf conf/retrained.json --gpuid 1

• Static-STModel:

python main.py --conf conf/static.json --gpuid 1