HD-TTS

This folder contains the official code for the reproducibility of the experiments presented in the paper "Graph-based Forecasting with Missing Data through Spatiotemporal Downsampling" (ICML 2024). The paper proposes a graph-based time series forecasting architecture that handles missing data by computing representations at various spatiotemporal scales and adaptively combining them to generate forecasts based on the missing data pattern.

Authors: Ivan Marisca, Cesare Alippi, Filippo Maria Bianchi

---

Directory structure

The code directory is structured as follows:

.
├── config/
├── lib/
├── conda_env.yaml
└── experiments/
    ├── run_mso.py
    └── run_realworld.py

Datasets

In this paper, we introduce two new datasets: GraphMSO and EngRAD. In the following, we provide instructions on how to download and use them. The rest of the datasets used in the experiments are provided by the tsl library.

GraphMSO

The GraphMSO dataset is a synthetic dataset and is generated by the GraphMSO class in lib.datasets.mso.py. The dataset is generated on the fly and does not require any download. Please refer to the class documentation for more details.

EngRAD

The EngRAD dataset contains measurements of 5 different weather variables collected at 487 grid points in England from 2018 to 2020. The data has been provided by Open-Meteo and licensed under Attribution 4.0 International (CC BY 4.0).

The dataset is hosted on Zenodo and is downloaded automatically by instantiating the class EngRad in lib.datasets.engrad.py. Please refer to the class documentation for more details.

Configuration files

The config directory stores all the configuration files used to run the experiment using Hydra.

Requirements

To solve all dependencies, we recommend using Anaconda and the provided environment configuration by running the command:

conda env create -f conda_env.yml
conda activate hd-tts

Experiments

The scripts used for the experiments in the paper are in the experiments folder.

• run_mso.py is used to train and evaluate models on the new GraphMSO dataset. As an example, to run HD-TTS-IMP on the point missing setting do:

  (hd-tts) $ python experiments/run_mso.py model=hd_tts_imp dataset=mso_point 

• run_realworld.py is used to train and evaluate models on the real-world benchmarks. As an example, to run HD-TTS-AMP on EngRAD (Block–ST) do:

  (hd-tts) $ python experiments/run_realworld.py model=hd_tts_amp dataset=engrad dataset/mode=block_st

Citing

Please consider citing the paper if you find it useful for your research.

@inproceedings{marisca2024graph,
  title     = {Graph-based Forecasting with Missing Data through Spatiotemporal Downsampling},
  author    = {Marisca, Ivan and Alippi, Cesare and Bianchi, Filippo Maria},
  booktitle = {Proceedings of the 41st International Conference on Machine Learning},
  pages     = {34846--34865},
  year      = {2024},
  volume    = {235},
  series    = {Proceedings of Machine Learning Research},
  publisher = {PMLR}
}