SAGDFN

This is the PyTorch official implementation of SAGDFN: A Scalable Adaptive Graph Diffusion Forecasting Network for Multivariate Time Series Forecasting

Requirements

pip install -r requirements.txt

Data Preparation

The traffic data files for Los Angeles (METR-LA)should be put into the data/ folder. It is provided by DCRNN.

The Carpark1918 dataset can be downlaod from the GoogleDrive.

Run the following commands to generate train/test/val dataset at data/{METR-LA,Carpark}/{train,val,test}.npz.

# Unzip the datasets
unzip data/metr-la.h5.zip -d data/

# Create data directories
mkdir -p data/{METR-LA,Carpark}

# METR-LA
python -m scripts.generate_training_data --output_dir=data/METR-LA --traffic_df_filename=data/metr-la.h5

# Carpark1918
python -m scripts.generate_training_data --output_dir=data/Carpark --traffic_df_filename=data/carpark_05_06.h5

Train Model

When you train the model, you can run:

# Use METR-LA dataset
python train.py --config_filename=data/model/para_la.yaml

# Use Carpark1918 dataset
python train.py --config_filename=data/model/para_carpark.yaml

Hyperparameters can be modified in the para_la.yaml and para_carpark.yaml files.