This is an official Pytorch implementation of HiMoE.

Environment

python, numpy, torch

Data Preparation

We have provided download links for the PEMS04 and KNOWAIR datasets. However, the Beijing and TongZhou datasets cannot be made publicly available.

You need to download the data to ./data/PEMS04/ or ./data/KNOWAIR/ and follow the instructions in ./data/README.md to initialize the datasets.

The original datasets can be found at ASTGCN and PM2.5-GNN.

Run Pre-trained Model

python main.py --conf ./conf/himoe.json --dataset PEMS04 --data_dir ./data/PEMS04/ --gpuid 0 --mode test_only --best_model_path ./exp/pems04-best.pkl

python main.py --conf ./conf/himoe.json --dataset KNOWAIR --data_dir ./data/KNOWAIR/ --gpuid 0 --mode test_only --best_model_path ./exp/knowair-best.pkl

Model Training

python main.py --conf ./conf/himoe.json --dataset PEMS04 --data_dir ./data/PEMS04/ --gpuid 0

python main.py --conf ./conf/himoe.json --dataset KNOWAIR --data_dir ./data/KNOWAIR/ --gpuid 0