This is the official code base of paper 'A Unified Model for Spatio-Temporal Prediction Queries with Arbitrary Modifiable Areal Units' (accepted by ICDE 2024).
One4All-ST is a framework that can conduct spatio-temporal prediction for arbitrary modifiable units using only one model.
It contains three parts namely Multi-scale Joint Learning, Optimal Combination Search and Index and Modifiable Areal Units Prediction.
This repository includes necessary open-source code to reproduce the results in our paper.
Our project based on the Open Source Spatio-Temporal Prediction Package UCTB. Thus we highly recommend you follow the tutorial here to install essential dependencies of UCTB. What's more, to address our problem targeted at spatio-temporal prediction for modifiable units, we use extra library geopandas and shapely to read and manipulate geometries.
As a result, the full installation process is as follows.
# follow the specific situation in the tutorial above
conda env create -f environment.yaml
# install `geopandas`
pip install geopandas
# install `shapely`
pip install shapelyThe scripts in the root directory of our repository are organized according to the three components of the framework in which exp.py is the entrace file for model defining, training and saving results, regular.py is the script responsible for rasterize arbitrary query into grids and test_query.py will includes how to obtain optimal combinations and show the final test results. Besides, we provide a script that contains the entire process execution named Runner_for_full_scripts.py
# obtain raterized query (more options can be refered to the regular.py)
python regular.py
# obtain unified model (more options can be refered to the exp.py)
python exp.py
# obtain optimal combinations and test results (more options can be refered to the test_query.py)
python test_query.pyIf One4All-ST is helpful for your work, please cite and star our project.
@article{chen2024unified,
title={A Unified Model for Spatio-Temporal Prediction Queries with Arbitrary Modifiable Areal Units},
author={Chen, Liyue and Fang, Jiangyi and Liu, Tengfei and Cao, Shaosheng and Wang, Leye},
journal={arXiv preprint arXiv:2403.07022},
year={2024}
}