This repo is the code for our paper MSSTN: Multi-Scale Spatial-Temporal Network for Air Pollution Prediction. We provide pre-processed data and trained models that can reproduce main result listed in our paper. Please contact us at [email protected] if you have any question.

We have tested our code under centos7, python3, and tensorflow 1.8.0. Similiar environment and later versions may also work but we didn't test that.

We provide pre-processed data on Baidu NetDisk (Secure Code: 057p). Download data and replace the '/data/' folder before use.

Use following command to train models from scratch:

python3 main.py train

Use following command to load trained models and show result on test set:

python3 main.py test [InferenceModel]

where InferenceModel can be found below in config part.

You may modify config.yaml to tune the training process by yourself. For example, item 'Target_City' decide which city to optimize/test when 'city_number' is set to 1, and 'InferenceModel' claim the trained model to load. More specifically, this table shows the relationship between them: