1.The result is not good as origin code.It will increase 10% MAE now.
2.I don't use the same hyper-parameters as origin code. Because when i use the same hyper-parameters, it will appear gradient exploding(epsilon=10).
3.I don't use the same LR_SCHEDULER as origin code.
4.These problem will be address after 2 months because i am busy now.If you can address them now, i will appreciate it.(谢谢)

This version of implementation is only for learning purpose. For research, please refer to and cite from the following paper:

@inproceedings{
  oreshkin2020fcgaga,
  title={{FC-GAGA}: Fully Connected Gated Graph Architecture for Spatio-Temporal Traffic Forecasting},
  author={Boris N. Oreshkin and Arezou Amini and Lucy Coyle and Mark J. Coates},
  booktitle={AAAI},
  year={2021},
}

The code is built based on Python 3.9, PyTorch 1.12.0, EasyTorch and BasicTS.

After ensuring that PyTorch is installed correctly, you can install other dependencies via:

pip install -r requirements.txt

The code is now refactored based on BasicTS,and you can get kinds of datasets from it. Then unzip them to datasets/raw_data/.

You can pre-process all data via:

cd /path/to/your/project
bash scripts/data_preparation/all.sh

Then the dataset directory will look like this:

datasets
   ├─METR-LA
   ├─METR-BAY
   ├─PEMS04
   ├─raw_data
   |    ├─PEMS04
   |    ├─PEMS-BAY
   |    ├─METR-LA
   ├─README.md

python step/run.py --cfg='model/fc-gaga_$DATASET.py' --gpus='0'
# python step/run.py --cfg='model/fc-gaga_METR-LA.py' --gpus='0'
# python step/run.py --cfg='model/fc-gaga_PEMS-BAY.py' --gpus='0'

Replace $DATASET_NAME with one of METR-LA, PEMS-BAYas shown in the code above. Configuration file model/fc-gaga_$DATASET.py describes the forecasting configurations. Edit BATCH_SIZE and GPU_NUM in the configuration file and --gpu in the command line to run on your own hardware.