zqylen / MAGCN

MAGCN: A Multi-Adaptive Graph Convolutional Network for Traffic Forecasting

MAGCN

MAGCN: A Multi-Adaptive Graph Convolutional Network for Traffic Forecasting

Prerequisites

• Python3( >3.5)
• Pytorch (>1.0)
• Other Python libraries can be installed by pip install -r requirements.txt

Data Preparation

• Uncompress data file using tar -zxvf data.tar.gz

    -data\  
      -PEMS03\  
        -pems03.csv
        -pems03.npz
        -pems03.txt
        -pems03_data.csv
      -PEMS07\  
        -pems07.csv
        -pems07.npz
  

• Preprocess the data with

  python data_gen/pems_gendata.py --data_path ./data/PEMS03 --out_folder ./data/PEMS03 --config ./config/pems03/pems03_train.yaml

  python data_gen/pems_gendata.py --data_path ./data/PEMS07 --out_folder ./data/PEMS07 --config ./config/pems07/pems07_train.yaml

Training & Testing

Training

Change the config file depending on what you want.

`python main.py --config ./config/pems03/pems03_train.yaml`

`python main.py --config ./config/pems07/pems07_train.yaml`

Testing

To evaluate model on PEMS03, run

`python main.py --config ./config/pems03/pems03_test.yaml`

To evaluate model on PEMS07, run

`python main.py --config ./config/pems07/pems07_test.yaml`